rapid research report on generative ai

Australian Council of Learned Academies Evidence-based Interdisciplinary Research

• Strategic Plan
• Member Academies
• Board of Directors
• ACOLA Committees
• Management Team
• Annual Reports
• ACOLA Code of Conduct
• Our History
• Our Process
• Horizon Scanning Series
• Australia’s Energy Transition Research Plan
• ACOLA Parliamentary Library Seminars
• KONEKSI: The Australia-Indonesia Knowledge Partnership
• Securing Australia’s Future
• Making interdisciplinary research work
• National Academies Forum
• All Publications
• Media Releases/News
• Events | Seminars
• Submissions

Contributors

Rapid response information report: generative ai.

• Other Programs
• Rapid Response Information Report: Generative…

Generative AI – language models and multimodal foundation models

The current ‘ChatGPT moment’ is provoking public conversation about the role AI should have in Australian society.

In February 2023, Australia’s National Science and Technology Council, at the request of the Minister for Industry and Science, the Hon Ed Husic MP, commissioned a report exploring the rapidly evolving area of Generative AI. The Australian Council of Learned Academies was pleased to develop this rapid report, convening contributions from the Academy of Technological Sciences and Engineering, the Academy of the Humanities, and the Academy of Science.

The report has been developed in the context of rapid change in the ecosystem and heightened expectations about the risks and possibilities of both ChatGPT in particular and generative AI more broadly. Generative AI raises questions about opportunities and risks of widespread adoption; the scope and adequacy of national strategic planning and policies; the fitness of legal and regulatory approaches; and the implications of increasing geopolitical competition and geo-specific regulation in AI-related technologies and industries.

The report responds to the following questions:

• What are the opportunities and risks of applying large language models (LLMs) and multimodal foundation models (MFMs) learning technologies over the next two, five and ten years?
• What are some examples of strategies that have been put in place internationally by other advanced economies since the launch of models like ChatGPT to address the potential opportunities and impacts of artificial intelligence (AI)?

The development of the report was supported by lead authors Distinguished Professor Genevieve Bell FAHA FTSE, Professor Julian Thomas FAHA, Professor Jean Burgess FAHA and Professor Shazia Sadiq FTSE with input from 24 expert contributors and 7 peer reviewers.

Project Management

Project funding and support.

The report was commissioned by Australia’s National Science and Technology Council. 

Acknowledgement of Country

ACOLA acknowledges the Traditional Owners and custodians of the lands on which our company is located and where we conduct our business. We pay our respects to Elders past and present.

Related Posts

Leave a Reply Cancel reply

Your email address will not be published. Required fields are marked *

Save my name, email, and website in this browser for the next time I comment.

Post comment

Suggestions or feedback?

MIT News | Massachusetts Institute of Technology

• Machine learning
• Social justice
• Black holes
• Classes and programs

Departments

• Aeronautics and Astronautics
• Brain and Cognitive Sciences
• Architecture
• Political Science
• Mechanical Engineering

Centers, Labs, & Programs

• Abdul Latif Jameel Poverty Action Lab (J-PAL)
• Picower Institute for Learning and Memory
• Lincoln Laboratory
• School of Architecture + Planning
• School of Engineering
• School of Humanities, Arts, and Social Sciences
• Sloan School of Management
• School of Science
• MIT Schwarzman College of Computing

Explained: Generative AI

Press contact :, media download.

*Terms of Use:

Images for download on the MIT News office website are made available to non-commercial entities, press and the general public under a Creative Commons Attribution Non-Commercial No Derivatives license . You may not alter the images provided, other than to crop them to size. A credit line must be used when reproducing images; if one is not provided below, credit the images to "MIT."

Previous image Next image

A quick scan of the headlines makes it seem like generative artificial intelligence is everywhere these days. In fact, some of those headlines may actually have been written by generative AI, like OpenAI’s ChatGPT, a chatbot that has demonstrated an uncanny ability to produce text that seems to have been written by a human.

But what do people really mean when they say “generative AI?”

Before the generative AI boom of the past few years, when people talked about AI, typically they were talking about machine-learning models that can learn to make a prediction based on data. For instance, such models are trained, using millions of examples, to predict whether a certain X-ray shows signs of a tumor or if a particular borrower is likely to default on a loan.

Generative AI can be thought of as a machine-learning model that is trained to create new data, rather than making a prediction about a specific dataset. A generative AI system is one that learns to generate more objects that look like the data it was trained on.

“When it comes to the actual machinery underlying generative AI and other types of AI, the distinctions can be a little bit blurry. Oftentimes, the same algorithms can be used for both,” says Phillip Isola, an associate professor of electrical engineering and computer science at MIT, and a member of the Computer Science and Artificial Intelligence Laboratory (CSAIL).

And despite the hype that came with the release of ChatGPT and its counterparts, the technology itself isn’t brand new. These powerful machine-learning models draw on research and computational advances that go back more than 50 years.

An increase in complexity

An early example of generative AI is a much simpler model known as a Markov chain. The technique is named for Andrey Markov, a Russian mathematician who in 1906 introduced this statistical method to model the behavior of random processes. In machine learning, Markov models have long been used for next-word prediction tasks, like the autocomplete function in an email program.

In text prediction, a Markov model generates the next word in a sentence by looking at the previous word or a few previous words. But because these simple models can only look back that far, they aren’t good at generating plausible text, says Tommi Jaakkola, the Thomas Siebel Professor of Electrical Engineering and Computer Science at MIT, who is also a member of CSAIL and the Institute for Data, Systems, and Society (IDSS).

“We were generating things way before the last decade, but the major distinction here is in terms of the complexity of objects we can generate and the scale at which we can train these models,” he explains.

Just a few years ago, researchers tended to focus on finding a machine-learning algorithm that makes the best use of a specific dataset. But that focus has shifted a bit, and many researchers are now using larger datasets, perhaps with hundreds of millions or even billions of data points, to train models that can achieve impressive results.

The base models underlying ChatGPT and similar systems work in much the same way as a Markov model. But one big difference is that ChatGPT is far larger and more complex, with billions of parameters. And it has been trained on an enormous amount of data — in this case, much of the publicly available text on the internet.

In this huge corpus of text, words and sentences appear in sequences with certain dependencies. This recurrence helps the model understand how to cut text into statistical chunks that have some predictability. It learns the patterns of these blocks of text and uses this knowledge to propose what might come next.

More powerful architectures

While bigger datasets are one catalyst that led to the generative AI boom, a variety of major research advances also led to more complex deep-learning architectures.

In 2014, a machine-learning architecture known as a generative adversarial network (GAN) was proposed by researchers at the University of Montreal. GANs use two models that work in tandem: One learns to generate a target output (like an image) and the other learns to discriminate true data from the generator’s output. The generator tries to fool the discriminator, and in the process learns to make more realistic outputs. The image generator StyleGAN is based on these types of models.  

Diffusion models were introduced a year later by researchers at Stanford University and the University of California at Berkeley. By iteratively refining their output, these models learn to generate new data samples that resemble samples in a training dataset, and have been used to create realistic-looking images. A diffusion model is at the heart of the text-to-image generation system Stable Diffusion.

In 2017, researchers at Google introduced the transformer architecture, which has been used to develop large language models, like those that power ChatGPT. In natural language processing, a transformer encodes each word in a corpus of text as a token and then generates an attention map, which captures each token’s relationships with all other tokens. This attention map helps the transformer understand context when it generates new text.

These are only a few of many approaches that can be used for generative AI.

A range of applications

What all of these approaches have in common is that they convert inputs into a set of tokens, which are numerical representations of chunks of data. As long as your data can be converted into this standard, token format, then in theory, you could apply these methods to generate new data that look similar.

“Your mileage might vary, depending on how noisy your data are and how difficult the signal is to extract, but it is really getting closer to the way a general-purpose CPU can take in any kind of data and start processing it in a unified way,” Isola says.

This opens up a huge array of applications for generative AI.

For instance, Isola’s group is using generative AI to create synthetic image data that could be used to train another intelligent system, such as by teaching a computer vision model how to recognize objects.

Jaakkola’s group is using generative AI to design novel protein structures or valid crystal structures that specify new materials. The same way a generative model learns the dependencies of language, if it’s shown crystal structures instead, it can learn the relationships that make structures stable and realizable, he explains.

But while generative models can achieve incredible results, they aren’t the best choice for all types of data. For tasks that involve making predictions on structured data, like the tabular data in a spreadsheet, generative AI models tend to be outperformed by traditional machine-learning methods, says Devavrat Shah, the Andrew and Erna Viterbi Professor in Electrical Engineering and Computer Science at MIT and a member of IDSS and of the Laboratory for Information and Decision Systems.

“The highest value they have, in my mind, is to become this terrific interface to machines that are human friendly. Previously, humans had to talk to machines in the language of machines to make things happen. Now, this interface has figured out how to talk to both humans and machines,” says Shah.

Raising red flags

Generative AI chatbots are now being used in call centers to field questions from human customers, but this application underscores one potential red flag of implementing these models — worker displacement.

In addition, generative AI can inherit and proliferate biases that exist in training data, or amplify hate speech and false statements. The models have the capacity to plagiarize, and can generate content that looks like it was produced by a specific human creator, raising potential copyright issues.

On the other side, Shah proposes that generative AI could empower artists, who could use generative tools to help them make creative content they might not otherwise have the means to produce.

In the future, he sees generative AI changing the economics in many disciplines.

One promising future direction Isola sees for generative AI is its use for fabrication. Instead of having a model make an image of a chair, perhaps it could generate a plan for a chair that could be produced.

He also sees future uses for generative AI systems in developing more generally intelligent AI agents.

“There are differences in how these models work and how we think the human brain works, but I think there are also similarities. We have the ability to think and dream in our heads, to come up with interesting ideas or plans, and I think generative AI is one of the tools that will empower agents to do that, as well,” Isola says.

Share this news article on:

Related links.

• Tommi Jaakkola
• Phillip Isola
• Devavrat Shah
• Computer Science and Artificial Intelligence Laboratory
• Laboratory for Information and Decision Systems
• Institute for Data, Systems, and Society
• Department of Electrical Engineering and Computer Science

Related Topics

• Artificial intelligence
• Computer science and technology
• Human-computer interaction
• Labor and jobs
• Computer Science and Artificial Intelligence Laboratory (CSAIL)
• Laboratory for Information and Decision Systems (LIDS)
• Electrical Engineering & Computer Science (eecs)

Related Articles

To excel at engineering design, generative AI must learn to innovate, study finds

From physics to generative AI: An AI model for advanced pattern generation

MIT CSAIL researchers discuss frontiers of generative AI

Previous item Next item

More MIT News

Featured video: Moooving the needle on methane

Read full story →

The many-body dynamics of cold atoms and cross-country running

Heather Paxson named associate dean for faculty of the School of Humanities, Arts, and Social Sciences

Preparing MIT’s campus for cardiac emergencies

Researching extreme environments

To build a better AI helper, start by modeling the irrational behavior of humans

• More news on MIT News homepage →

Massachusetts Institute of Technology 77 Massachusetts Avenue, Cambridge, MA, USA

• Map (opens in new window)
• Events (opens in new window)
• People (opens in new window)
• Careers (opens in new window)
• Accessibility
• Social Media Hub
• MIT on Facebook
• MIT on YouTube
• MIT on Instagram

Generative AI: A Review on Models and Applications

Ieee account.

• Change Username/Password
• Update Address

Purchase Details

• Payment Options
• Order History
• View Purchased Documents

Profile Information

• Communications Preferences
• Profession and Education
• Technical Interests
• US & Canada: +1 800 678 4333
• Worldwide: +1 732 981 0060
• Contact & Support
• About IEEE Xplore
• Accessibility
• Terms of Use
• Nondiscrimination Policy
• Privacy & Opting Out of Cookies

A not-for-profit organization, IEEE is the world's largest technical professional organization dedicated to advancing technology for the benefit of humanity. © Copyright 2024 IEEE - All rights reserved. Use of this web site signifies your agreement to the terms and conditions.

• Asia Pacific
• Latin America
• Middle East & Africa
• North America
• Australia & New Zealand

Mainland China

• Hong Kong SAR, China
• Philippines
• Taiwan, China
• Channel Islands
• Netherlands
• Switzerland
• United Kingdom
• Saudi Arabia
• South Africa
• United Arab Emirates
• United States

From startups to legacy brands, you're making your mark. We're here to help.

• Innovation Economy Fueling the success of early-stage startups, venture-backed and high-growth companies.
• Midsize Businesses Keep your company growing with custom banking solutions for middle market businesses and specialized industries.
• Large Corporations Innovative banking solutions tailored to corporations and specialized industries.
• Commercial Real Estate Capitalize on opportunities and prepare for challenges throughout the real estate cycle.
• Community Impact Banking When our communities succeed, we all succeed. Local businesses, organizations and community institutions need capital, expertise and connections to thrive.
• International Banking Power your business' global growth and operations at every stage.
• Client Stories

Prepare for future growth with customized loan services, succession planning and capital for business equipment.

• Asset Based Lending Enhance your liquidity and gain the flexibility to capitalize on growth opportunities.
• Equipment Financing Maximize working capital with flexible equipment and technology financing.
• Trade & Working Capital Experience our market-leading supply chain finance solutions that help buyers and suppliers meet their working capital, risk mitigation and cash flow objectives.
• Syndicated Financing Leverage customized loan syndication services from a dedicated resource.
• Employee Stock Ownership Plans Plan for your business’s future—and your employees’ futures too—with objective advice and financing.

Institutional Investing

Serving the world's largest corporate clients and institutional investors, we support the entire investment cycle with market-leading research, analytics, execution and investor services.

• Institutional Investors We put our long-tenured investment teams on the line to earn the trust of institutional investors.
• Markets Direct access to market leading liquidity harnessed through world-class research, tools, data and analytics.
• Prime Services Helping hedge funds, asset managers and institutional investors meet the demands of a rapidly evolving market.
• Global Research Leveraging cutting-edge technology and innovative tools to bring clients industry-leading analysis and investment advice.
• Securities Services Helping institutional investors, traditional and alternative asset and fund managers, broker dealers and equity issuers meet the demands of changing markets.
• Financial Professionals
• Liquidity Investors

Providing investment banking solutions, including mergers and acquisitions, capital raising and risk management, for a broad range of corporations, institutions and governments.

• Center for Carbon Transition J.P. Morgan’s center of excellence that provides clients the data and firmwide expertise needed to navigate the challenges of transitioning to a low-carbon future.
• Corporate Finance Advisory Corporate Finance Advisory (“CFA”) is a global, multi-disciplinary solutions team specializing in structured M&A and capital markets. Learn more.
• Development Finance Institution Financing opportunities with anticipated development impact in emerging economies.
• Sustainable Solutions Offering ESG-related advisory and coordinating the firm's EMEA coverage of clients in emerging green economy sectors.
• Mergers and Acquisitions Bespoke M&A solutions on a global scale.
• Capital Markets Holistic coverage across capital markets.
• Capital Connect
• In Context Newsletter from J.P. Morgan
• Director Advisory Services

Accept Payments

Explore Blockchain

Client Service

Process Payments

Manage Funds

Safeguard Information

Banking-as-a-service

Send Payments

• Partner Network

A uniquely elevated private banking experience shaped around you.

• Banking We have extensive personal and business banking resources that are fine-tuned to your specific needs.
• Investing We deliver tailored investing guidance and access to unique investment opportunities from world-class specialists.
• Lending We take a strategic approach to lending, working with you to craft the fight financing solutions matched to your goals.
• Planning No matter where you are in your life, or how complex your needs might be, we’re ready to provide a tailored approach to helping your reach your goals.

Whether you want to invest on your own or work with an advisor to design a personalized investment strategy, we have opportunities for every investor.

• Invest on your own Unlimited $0 commission-free online stock, ETF and options trades with access to powerful tools to research, trade and manage your investments.
• Work with our advisors When you work with our advisors, you'll get a personalized financial strategy and investment portfolio built around your unique goals-backed by our industry-leading expertise.
• Expertise for Substantial Wealth Our Wealth Advisors & Wealth Partners leverage their experience and robust firm resources to deliver highly-personalized, comprehensive solutions across Banking, Lending, Investing, and Wealth Planning.
• Why Wealth Management?
• Retirement Calculators
• Market Commentary

Who We Serve

Explore a variety of insights.

Global Research

• Newsletters

Insights by Topic

Explore a variety of insights organized by different topics.

Insights by Type

Explore a variety of insights organized by different types of content and media.  

• All Insights

We aim to be the most respected financial services firm in the world, serving corporations and individuals in more than 100 countries.

Key takeaways

• Generative AI tools reduce the money and time needed for content creation, boosting productivity and profitability.
• However, they could also lead to copyright infringement and increase data security risks.
• Overall, J.P. Morgan Research estimates generative AI could increase global GDP by $7–10 trillion, or by as much as 10%.
• The technology could result in a massive workforce productivity boom over the next one to three years, which could affect the shape of the economic cycle.

Disclaimer:  This article was not written using generative AI.

Generative AI — a category of artificial intelligence algorithms that can generate new content based on existing data — has been hailed as the next frontier for various industries, from tech to banking and media. Indeed, it is already being used for tasks such as content creation and data analysis, with widespread implications for the future of work.

“The advent of generative AI is a seminal moment in tech, more so than the Internet or the iPhone,” said Mark Murphy, Head of U.S. Enterprise Software Research at J.P. Morgan. “We see the potential for a massive workforce productivity boom over the next one to three years, which could affect the shape of the economic cycle. There could also be mass-scale white-collar job realignment four to eight years from now.”

At J.P. Morgan’s 5th Annual Global Machine Learning Conference, which took place in Paris in October 2023, investors were asked about their views on generative AI. Overall, investors believe that the uptake of the technology will be most prevalent in marketing (28%), followed by legal services and insurance (21%), media (20%), data analytics (18%) and then consumer technology (13%) — suggesting that sweeping changes lie ahead for these industries. 

Is generative AI a game changer?

Generative AI tools reduce the money and time needed for content creation, thereby boosting productivity and profitability.

However, they could also displace jobs, lead to copyright infringement and create data security risks.

Overall, J.P. Morgan Research estimates generative AI could increase global GDP by $7–10 trillion, or as much as 10%.

The technology could result in a workforce productivity boom over the next one to three years, which could affect the shape of the economic cycle.

“We see the potential for a massive workforce productivity boom over the next one to three years, which could affect the shape of the economic cycle.”

Mark Murphy

Head of U.S. Enterprise Software Research, J.P. Morgan

What are the advantages of generative AI?

“Fundamentally, generative AI reduces the money and time needed for content creation — across text, code, audio, images, video and combinations thereof,” said Gokul Hariharan, Co-Head of Asia Pacific Technology, Media and Telecom Research at J.P. Morgan. Businesses can produce more content at speed and at scale, boosting productivity and profitability.

The rise of generative AI could also breed innovation, paving the way for new business models and applications. While tools like ChatGPT are trained on generic data, companies are now creating generative AI systems designed for specific verticals and datasets. “ChatGPT is putting the wind in the sails of other companies, and hundreds of new startups are rushing to develop foundation models, build AI-native apps and stand up infrastructure,” Hariharan said. “Because of the potentially very large impact, a positive sentiment cycle here could well lead to a valuation bubble in related names.” 

Generative AI could revolutionize content creation

Generative AI’s output across text, code, images and video is expected to improve exponentially through 2030, surpassing what human workers can produce. 

What are the disadvantages of generative AI?

On the flip side, generative AI tools are not 100% accurate. ChatGPT is prone to “hallucinations,” or output that deviates from its training data. “Because of this, generative AI will not yet replace jobs entirely. Instead, it will augment existing jobs by automating repetitive tasks,” Hariharan said.

Generative AI tools also run the risk of plagiarism and copyright infringement, as they often repeat or paraphrase data sourced from elsewhere on the Internet. Then there are the data security risks involved, especially where client confidentiality is concerned. When new information is inputted into a generative AI system, it becomes part of its data repository and is made publicly available to other users. “Companies will understandably be guarded about this, so generative AI providers will need to create tools that are ringfenced — ensuring that all information is self-contained to each organization and isn’t commingled with the rest of the world,” Murphy said.

Despite these hurdles, generative AI could well be a game changer for business, dramatically redefining the way companies work. “Generative AI is the most important technological development of the last several decades. It is rapidly enabling use cases and scenarios that people once said would be impossible to achieve, and it’s only going to get smarter,” Murphy said. “It’s critical that generative AI is used responsibly and governed properly, so that it can amplify human potential instead of becoming too disruptive.”

Investing in generative AI  

Overall, J.P. Morgan Research estimates generative AI could increase global GDP by $7–10 trillion, or by as much as 10%. “We affirm our extreme bullish outlier view, established over a year ago, on the importance of generative AI. The technology is driving a powerful tech investment wave, with co-pilots and other generative AI product cycles launching in the near term,” Murphy said. “Indeed, VC investments are pivoting rapidly from cloud and crypto to generative AI, and a material percentage of Y Combinator companies are also in this space.”

While generative AI is undoubtedly a massive boon for the software industry, hardware companies also stand to benefit from the uptake of the technology. “The AI-related supply chain is buoyant, with continued strong demand for AI training hardware and emerging demand for various AI inference solutions — both likely to sustain AI demand into 2024 and beyond,” Hariharan noted. This will boost the global semiconductor market, which is forecast to see a 16% growth in revenue in 2024 and increased demand into 2025.

Investments in other AI infrastructure will also continue to grow, particularly for servers, switches and optics, according to Samik Chatterjee, Head of the IT Hardware and Networking Equipment research team at J.P. Morgan. “The leverage to AI for certain sections of our broader hardware coverage is likely to be a windfall moment,” Chatterjee said.

AI end-point devices such as smartphones and PCs will also likely see a pick-up in product launches in the second half of 2024, with more material adoption in 2025. “These will focus on local inferencing to enable low latency and personalized experiences, while also leveraging more complex audio and visual interactions such as gesture and facial recognition,” Chatterjee added. 

The generative AI adoption lifecycle

The generative AI adoption lifecycle consists of three stages. Stage 1 is largely focused on training demand for large language models; stage 2 sees organizations bringing co-pilot applications to the market; and stage 3 involves client devices carrying dedicated inference engines. 

What is generative AI?

Generative AI refers to a category of artificial intelligence algorithms that can generate new content based on the data they have been trained on. Their output — which includes text, images, videos, audio and more — typically resembles human-generated data. 

What are some examples of generative AI?

ChatGPT, an artificial intelligence chatbot, is a form of text-based generative AI. Other popular examples include DALL-E 2, which can generate digital images from natural language descriptions. 

What are the impacts of generative AI?

Generative AI reduces the money and time needed for content creation. It could also breed innovation, paving the way for new business models and applications. However, the technology could lead to copyright issues and data security risks.

Related insights

Leveraging cutting-edge technology and innovative tools to bring clients industry-leading analysis and investment advice.

The future of Big Tech

December 23, 2022

What challenges and opportunities lie ahead for Big Tech? J.P. Morgan Research explores the 2023 outlook.

Technology at our firm

This communication is provided for information purposes only. Please read J.P. Morgan research reports related to its contents for more information, including important disclosures. JPMorgan Chase & Co. or its affiliates and/or subsidiaries (collectively, J.P. Morgan) normally make a market and trade as principal in securities, other financial products and other asset classes that may be discussed in this communication.

This communication has been prepared based upon information, including market prices, data and other information, from sources believed to be reliable, but J.P. Morgan does not warrant its completeness or accuracy except with respect to any disclosures relative to J.P. Morgan and/or its affiliates and an analyst's involvement with any company (or security, other financial product or other asset class) that may be the subject of this communication. Any opinions and estimates constitute our judgment as of the date of this material and are subject to change without notice. Past performance is not indicative of future results. This communication is not intended as an offer or solicitation for the purchase or sale of any financial instrument. J.P. Morgan Research does not provide individually tailored investment advice. Any opinions and recommendations herein do not take into account individual client circumstances, objectives, or needs and are not intended as recommendations of particular securities, financial instruments or strategies to particular clients. You must make your own independent decisions regarding any securities, financial instruments or strategies mentioned or related to the information herein. Periodic updates may be provided on companies, issuers or industries based on specific developments or announcements, market conditions or any other publicly available information. However, J.P. Morgan may be restricted from updating information contained in this communication for regulatory or other reasons. Clients should contact analysts and execute transactions through a J.P. Morgan subsidiary or affiliate in their home jurisdiction unless governing law permits otherwise.

This communication may not be redistributed or retransmitted, in whole or in part, or in any form or manner, without the express written consent of J.P. Morgan. Any unauthorized use or disclosure is prohibited. Receipt and review of this information constitutes your agreement not to redistribute or retransmit the contents and information contained in this communication without first obtaining express permission from an authorized officer of J.P. Morgan. 

You're now leaving J.P. Morgan

J.P. Morgan’s website and/or mobile terms, privacy and security policies don’t apply to the site or app you're about to visit. Please review its terms, privacy and security policies to see how they apply to you. J.P. Morgan isn’t responsible for (and doesn’t provide) any products, services or content at this third-party site or app, except for products and services that explicitly carry the J.P. Morgan name.

Research with Generative AI

Resources for scholars and researchers

Generative AI (GenAI) technologies offer new opportunities to advance research and scholarship. This resource page aims to provide Harvard researchers and scholars with basic guidance, information on available resources, and contacts. The content will be regularly updated as these technologies continue to evolve. Your feedback is welcome.

Leading the way

Harvard’s researchers are making strides not only on generative AI, but the larger world of artificial intelligence and its applications. Learn more about key efforts.

The Kempner Institute

The Kempner Institute is dedicated to revealing the foundations of intelligence in both natural and artificial contexts, and to leveraging these findings to develop groundbreaking technologies.

Harvard Data Science Initiative

The Harvard Data Science Initiative is dedicated to understanding the many dimensions of data science and propelling it forward.

More AI @ Harvard

Generative AI is only part of the fascinating world of artificial intelligence. Explore Harvard’s groundbreaking and cross-disciplinary academic work in AI.

funding opportunity

GenAI Research Program/ Summer Funding for Harvard College Students 2024

The Office of the Vice Provost for Research, in partnership with the Office of Undergraduate Research and Fellowships, is pleased to offer an opportunity for collaborative research projects related to Generative AI between Harvard faculty and undergraduate students over the summer of 2024.

Learn more and apply

Frequently asked questions

Can i use generative ai to write and/or develop research papers.

Academic publishers have a range of policies on the use of AI in research papers. In some cases, publishers may prohibit the use of AI for certain aspects of paper development. You should review the specific policies of the target publisher to determine what is permitted.

Here is a sampling of policies available online:

• JAMA and the JAMA Network
• Springer Nature

How should AI-generated content be cited in research papers?

Guidance will likely develop as AI systems evolve, but some leading style guides have offered recommendations:

• The Chicago Manual of Style
• MLA Style Guide

Should I disclose the use of generative AI in a research paper?

Yes. Most academic publishers require researchers using AI tools to document this use in the methods or acknowledgements sections of their papers. You should review the specific guidelines of the target publisher to determine what is required.

Can I use AI in writing grant applications?

You should review the specific policies of potential funders to determine if the use of AI is permitted. For its part, the National Institutes of Health (NIH) advises caution : “If you use an AI tool to help write your application, you also do so at your own risk,” as these tools may inadvertently introduce issues associated with research misconduct, such as plagiarism or fabrication.

Can I use AI in the peer review process?

Many funders have not yet published policies on the use of AI in the peer review process. However, the National Institutes of Health (NIH) has prohibited such use “for analyzing and formulating peer review critiques for grant applications and R&D contract proposals.” You should carefully review the specific policies of funders to determine their stance on the use of AI

Are there AI safety concerns or potential risks I should be aware of?

Yes. Some of the primary safety issues and risks include the following:

• Bias and discrimination: The potential for AI systems to exhibit unfair or discriminatory behavior.
• Misinformation, impersonation, and manipulation: The risk of AI systems disseminating false or misleading information, or being used to deceive or manipulate individuals.
• Research and IP compliance: The necessity for AI systems to adhere to legal and ethical guidelines when utilizing proprietary information or conducting research.
• Security vulnerabilities: The susceptibility of AI systems to hacking or unauthorized access.
• Unpredictability: The difficulty in predicting the behavior or outcomes of AI systems.
• Overreliance: The risk of relying excessively on AI systems without considering their limitations or potential errors.

See Initial guidelines for the use of Generative AI tools at Harvard for more information.

• Initial guidelines for the use of Generative AI tools at Harvard

Generative AI tools

• Explore Tools Available to the Harvard Community
• Request API Access
• Request a Vendor Risk Assessment
• Questions? Contact HUIT

Copyright and intellectual property

• Copyright and Fair Use: A Guide for the Harvard Community
• Copyright Advisory Program
• Intellectual Property Policy
• Protecting Intellectual Property

Data security and privacy

• Harvard Information Security and Data Privacy
• Data Security Levels – Research Data Examples
• Privacy Policies and Guidelines

Research support

• University Research Computing and Data (RCD) Services
• Research Administration and Compliance
• Research Computing
• Research Data and Scholarship
• Faculty engaged in AI research
• Centers and initiatives engaged in AI research
• Degree and other education programs in AI

Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.

• View all journals
• Explore content
• About the journal
• Publish with us
• Sign up for alerts
• Published: 22 March 2024

Generative AI for designing and validating easily synthesizable and structurally novel antibiotics

• Kyle Swanson   ORCID: orcid.org/0000-0002-7385-7844 1   na1 ,
• Gary Liu 2   na1 ,
• Denise B. Catacutan 2 ,
• Autumn Arnold 2 ,
• James Zou   ORCID: orcid.org/0000-0001-8880-4764 1 , 3 &
• Jonathan M. Stokes   ORCID: orcid.org/0009-0001-8378-2380 2  

Nature Machine Intelligence volume  6 ,  pages 338–353 ( 2024 ) Cite this article

6544 Accesses

1 Citations

504 Altmetric

Metrics details

• Antibiotics
• Computational chemistry
• Phenotypic screening

The rise of pan-resistant bacteria is creating an urgent need for structurally novel antibiotics. Artificial intelligence methods can discover new antibiotics, but existing methods have notable limitations. Property prediction models, which evaluate molecules one-by-one for a given property, scale poorly to large chemical spaces. Generative models, which directly design molecules, rapidly explore vast chemical spaces but generate molecules that are challenging to synthesize. Here we introduce SyntheMol, a generative model that designs new compounds, which are easy to synthesize, from a chemical space of nearly 30 billion molecules. We apply SyntheMol to design molecules that inhibit the growth of Acinetobacter baumannii , a burdensome Gram-negative bacterial pathogen. We synthesize 58 generated molecules and experimentally validate them, with six structurally novel molecules demonstrating antibacterial activity against A. baumannii and several other phylogenetically diverse bacterial pathogens. This demonstrates the potential of generative artificial intelligence to design structurally novel, synthesizable and effective small-molecule antibiotic candidates from vast chemical spaces, with empirical validation.

This is a preview of subscription content, access via your institution

Access options

Access Nature and 54 other Nature Portfolio journals

Get Nature+, our best-value online-access subscription

24,99 € / 30 days

cancel any time

Subscribe to this journal

Receive 12 digital issues and online access to articles

111,21 € per year

only 9,27 € per issue

Buy this article

• Purchase on Springer Link
• Instant access to full article PDF

Prices may be subject to local taxes which are calculated during checkout

Similar content being viewed by others

SAVI, in silico generation of billions of easily synthesizable compounds through expert-system type rules

Hitesh Patel, Wolf-Dietrich Ihlenfeldt, … Marc C. Nicklaus

Accelerated antimicrobial discovery via deep generative models and molecular dynamics simulations

Payel Das, Tom Sercu, … Aleksandra Mojsilovic

Discovering highly potent antimicrobial peptides with deep generative model HydrAMP

Paulina Szymczak, Marcin Możejko, … Ewa Szczurek

Data availability

The data used in this paper, including training data and generated molecules, are available in the Supplementary Data . The data, along with trained model checkpoints and LC–MS and 1 H-NMR spectra, are available at https://doi.org/10.5281/zenodo.10257839 (ref. 68 ). The ChEMBL database can be accessed from www.ebi.ac.uk/chembl .

Code availability

Code for data processing and analysis, property prediction model training and SyntheMol molecule generation is available at https://github.com/swansonk14/SyntheMol (ref. 69 ). This code repository makes use of general cheminformatics functions from https://github.com/swansonk14/chemfunc as well as Chemprop model code from https://github.com/chemprop/chemprop .

Murray, C. J. et al. Global burden of bacterial antimicrobial resistance in 2019: a systematic analysis. Lancet 399 , 629–655 (2022).

Article   CAS   Google Scholar  

Rice, L. B. Federal funding for the study of antimicrobial resistance in nosocomial pathogens: No ESKAPE. J. Infect. Dis. 197 , 1079–1081 (2008).

Article   PubMed   Google Scholar  

Ma, Y. et al. Considerations and caveats in combating ESKAPE pathogens against nosocomial infections. Adv. Sci. 7 , 1901872 (2020).

Tacconelli, E. et al. Discovery, research, and development of new antibiotics: the WHO priority list of antibiotic-resistant bacteria and tuberculosis. Lancet Infect. Dis. 18 , 318–327 (2018).

Lee, C. R. et al. Biology of Acinetobacter baumannii : pathogenesis, antibiotic resistance mechanisms, and prospective treatment options. Front. Cell. Infect. Microbiol. 7 , 55 (2017).

Article   PubMed   PubMed Central   Google Scholar  

Carracedo-Reboredo, P. et al. A review on machine learning approaches and trends in drug discovery. Comput. Struct. Biotechnol. J. 19 , 4538–4558 (2021).

Article   CAS   PubMed   PubMed Central   Google Scholar  

Gaudelet, T. et al. Utilizing graph machine learning within drug discovery and development. Brief. Bioinform. 22 , bbab159 (2021).

Stokes, J. M. et al. A deep learning approach to antibiotic discovery. Cell 180 , 688–702.e13 (2020).

Rahman, A. S. M. Z. et al. A machine learning model trained on a high-throughput antibacterial screen increases the hit rate of drug discovery. PLoS Comput. Biol. 18 , e1010613 (2022).

Zeng, X. et al. Deep generative molecular design reshapes drug discovery. Cell Rep. Med. 3 , 100794 (2022).

Bilodeau, C., Jin, W., Jaakkola, T., Barzilay, R. & Jensen, K. F. Generative models for molecular discovery: recent advances and challenges. WIREs Comput. Mol. Sci. 12 , e1608 (2022).

Article   Google Scholar  

Bian, Y. & Xie, X. Q. Generative chemistry: drug discovery with deep learning generative models. J. Mol. Model. 27 , 71 (2021).

Article   CAS   PubMed   Google Scholar  

Liu, G. & Stokes, J. M. A brief guide to machine learning for antibiotic discovery. Curr. Opin. Microbiol. 69 , 102190 (2022).

Gao, W. & Coley, C. W. The synthesizability of molecules proposed by generative models. J. Chem. Inf. Model. 60 , 5714–5723 (2020).

Bradshaw, J., Paige, B., Kusner, M. J., Segler, M. H. S. & Hernández-Lobato, J. M. A model to search for synthesizable molecules. In Proc. 33rd International Conference on Neural Information Processing Systems (eds Wallach, H. M., Larochelle, H., Beygelzimer, A., d'Alché-Buc, F. & Fox, E. B.) 7937–7949 (Curran Associates Inc., 2019).

Bradshaw, J., Paige, B., Kusner, M. J., Segler, M. H. S. & Hernández-Lobato, J. M. Barking up the right tree: an approach to search over molecule synthesis DAGs. In Proc. 34th International Conference on Neural Information Processing Systems (eds Larochelle, H., Ranzato, M., Hadsell, R., Balcan, M. F. & Lin, H.) 6852–6866 (Curran Associates Inc., 2020).

Gottipati, S. K. et al. Learning to navigate the synthetically accessible chemical space using reinforcement learning. In Proc. 37th International Conference on Machine Learning (eds Daumé III, H. & Singh, A.) 3668–3679 (PMLR, 2020).

Gao, W., Mercado, R. & Coley, C. W. Amortized tree generation for bottom-up synthesis planning and synthesizable molecular design. In Proc. 10th International Conference on Learning Representations (2022); https://openreview.net/forum?id=FRxhHdnxt1

Pedawi, A., Gniewek, P., Chang, C., Anderson, B. M. & Bedem, H. van den. An efficient graph generative model for navigating ultra-large combinatorial synthesis libraries. In Proc. 36th International Conference on Neural Information Processing Systems (eds Oh, A. H., Agarwal. A., Belgrave, D. & Cho, K.) (2022); https://openreview.net/forum?id=VBbxHvbJd94

Kocsis, L. & Szepesvári, C. Bandit based Monte-Carlo planning. In Proc. European Conference on Machine Learning, ECML 2006 Vol. 4212 (eds Furnkranz, J. et al.) 282–293 (Springer, 2006).

Coulom, R. Efficient selectivity and backup operators in Monte-Carlo tree search. In Proc. International Conference on Computers and Games, CG 2006 Vol. 4630 (eds van den Herik, H. J. et al.) 72–83 (Springer, 2007).

Grygorenko, O. O. et al. Generating multibillion chemical space of readily accessible screening compounds. iScience 23 , 101681 (2020).

Stokes, J. M., Davis, J. H., Mangat, C. S., Williamson, J. R. & Brown, E. D. Discovery of a small molecule that inhibits bacterial ribosome biogenesis. eLife 3 , e03574 (2014).

van der Maaten, L. & Hinton, G. Visualizing data using t- SNE. J. Mach. Learn. Res. 9 , 2579–2605 (2008).

Google Scholar  

Mendez, D. et al. ChEMBL: towards direct deposition of bioassay data. Nucleic Acids Res. 47 , D930–D940 (2019).

Yang, K. et al. Analyzing learned molecular representations for property prediction. J. Chem. Inf. Model. 59 , 3370–3388 (2019).

RDKit: open-source cheminformatics. RDKit https://www.rdkit.org/ . Accessed 28 Mar 2022.

Breiman, L. Random forests. Mach. Learn. 45 , 5–32 (2001).

Silver, D. et al. Mastering the game of Go with deep neural networks and tree search. Nature 529 , 484–489 (2016).

Tversky, A. Features of similarity. Psychol. Rev. 84 , 327–352 (1977).

Rogers, D. & Hahn, M. Extended-connectivity fingerprints. J. Chem. Inf. Model. 50 , 742–754 (2010).

Arthur, D. & Vassilvitskii, S. K-Means++: the advantages of careful seeding. In Proc. Eighteenth Annu. ACM-SIAM Symp. Discrete Algorithms 1027–1035 (SIAM, 2007).

Maggiora, G., Vogt, M., Stumpfe, D. & Bajorath, J. Molecular similarity in medicinal chemistry: miniperspective. J. Med. Chem. 57 , 3186–3204 (2014).

Tanimoto, T. T. IBM Internal Report (IBM, 1957).

Nikaido, H. Molecular basis of bacterial outer membrane permeability revisited. Microbiol. Mol. Biol. Rev. 67 , 593–656 (2003).

Zurawski, D. V. et al. SPR741, an antibiotic adjuvant, potentiates the in vitro and in vivo activity of rifampin against clinically relevant extensively drug-resistant Acinetobacter baumannii . Antimicrob. Agents Chemother. 61 , e01239-17 (2017).

Eckburg, P. B. et al. Safety, tolerability, pharmacokinetics, and drug interaction potential of SPR741, an intravenous potentiator, after single and multiple ascending doses and when combined with β-lactam antibiotics in healthy subjects. Antimicrob. Agents Chemother. 63 , e00892-19 (2019).

Moffatt, J. H. et al. Colistin resistance in Acinetobacter baumannii is mediated by complete loss of lipopolysaccharide production. Antimicrob. Agents Chemother. 54 , 4971–4977 (2010).

O’Neill, A. J., Cove, J. H. & Chopra, I. Mutation frequencies for resistance to fusidic acid and rifampicin in Staphylococcus aureus . J. Antimicrob. Chemother. 47 , 647–650 (2001).

Björkholm, B. et al. Mutation frequency and biological cost of antibiotic resistance in Helicobacter pylori . Proc. Natl Acad. Sci. USA 98 , 14607–14612 (2001).

Nicholson, W. L. & Maughan, H. The spectrum of spontaneous rifampin resistance mutations in the rpoB Gene of Bacillussubtilis 168 spores differs from that of vegetative cells and resembles that of Mycobacterium tuberculosis . J. Bacteriol. 184 , 4936–4940 (2002).

Wu, Z. et al. MoleculeNet: a benchmark for molecular machine learning. Chem. Sci. 9 , 513–530 (2018).

Melo, M. C. R., Maasch, J. R. M. A. & de la Fuente-Nunez, C. Accelerating antibiotic discovery through artificial intelligence. Commun. Biol. 4 , 1050 (2021).

Yan, J. et al. Recent progress in the discovery and design of antimicrobial peptides using traditional machine learning and deep learning. Antibiotics 11 , 1451 (2022).

Mahlapuu, M., Håkansson, J., Ringstad, L. & Björn, C. Antimicrobial peptides: an emerging category of therapeutic agents. Front. Cell. Infect. Microbiol. 6 , 194 (2016).

Mahlapuu, M., Björn, C. & Ekblom, J. Antimicrobial peptides as therapeutic agents: opportunities and challenges. Crit. Rev. Biotechnol. 40 , 978–992 (2020).

Gómez-Bombarelli, R. et al. Automatic chemical design using a data-driven continuous representation of molecules. ACS Cent. Sci. 4 , 268–276 (2018).

Kang, S. & Cho, K. Conditional molecular design with deep generative models. J. Chem. Inf. Model. 59 , 43–52 (2019).

Krenn, M., Häse, F., Nigam, A., Friederich, P. & Aspuru-Guzik, A. Self-referencing embedded strings (SELFIES): a 100% robust molecular string representation. Mach. Learn. Sci. Technol. 1 , 045024 (2020).

Liu, Q., Allamanis, M., Brockschmidt, M. & Gaunt, A. L. Constrained graph variational autoencoders for molecule design. In Proc. 32nd International Conference on Neural Information Processing Systems (eds Wallach, H. M., Larochelle, H., Grauman, K. & Cesa-Bianchi, N.) 7806–7815 (Curran Associates Inc., 2018).

You, J., Liu, B., Ying, R., Pande, V. & Leskovec, J. Graph convolutional policy network for goal-directed molecular graph generation. In Proc. 32nd International Conference on Neural Information Processing Systems (eds Wallach, H. M., Larochelle, H., Grauman, K. & Cesa-Bianchi, N.) 6412–6422 (Curran Associates Inc., 2018).

Jin, W., Barzilay, R. & Jaakkola, T. Junction tree variational autoencoder for molecular graph generation. ICML 80 , 2323–2332 (2018).

Jin, W., Barzilay, R. & Jaakkola, T. Hierarchical generation of molecular graphs using structural motifs. ICML 119 , 4839–4848 (2020).

Bilodeau, C. et al. Generating molecules with optimized aqueous solubility using iterative graph translation. React. Chem. Eng. 7 , 297–309 (2022).

Sadybekov, A. A. et al. Synthon-based ligand discovery in virtual libraries of over 11 billion compounds. Nature 601 , 452–459 (2022).

Yang, X., Zhang, J., Yoshizoe, K., Terayama, K. & Tsuda, K. ChemTS: an efficient python library for de novo molecular generation. Sci. Technol. Adv. Mater. 18 , 972–976 (2017).

Qian, H., Lin, C., Zhao, D., Tu, S. & Xu, L. AlphaDrug: protein target specific de novo molecular generation. PNAS Nexus. 1 , pgac227 (2022).

Jin, W., Barzilay, R. & Jaakkola, T. Multi-objective molecule generation using interpretable substructures. ICML 119 , 4849–4859 (2020).

Segler, M. H. S., Preuss, M. & Waller, M. P. Planning chemical syntheses with deep neural networks and symbolic AI. Nature 555 , 604–610 (2018).

Coley, C. W. et al. A robotic platform for flow synthesis of organic compounds informed by AI planning. Science 365 , eaax1566 (2019).

Walters, W. P. & Murcko, M. Assessing the impact of generative AI on medicinal chemistry. Nat. Biotechnol. 38 , 143–145 (2020).

Corsello, S. M. et al. The Drug Repurposing Hub: a next-generation drug library and information resource. Nat. Med. 23 , 405–408 (2017).

Weininger, D. SMILES, a chemical language and information system. 1. Introduction to methodology and encoding rules. J. Chem. Inf. Model. 28 , 31–36 (1988).

CAS   Google Scholar  

Paszke, A. et al. PyTorch: an imperative style, high-performance deep learning library. In Proc. 33rd International Conference on Neural Information Processing Systems (eds Wallach, H. M., Larochelle, H., Beygelzimer, A., d'Alché-Buc, F. & Fox, E. B.) 8026–8037 (2019).

Pedregosa, F. et al. Scikit-learn: machine learning in Python. J. Mach. Learn. Res. 12 , 2825–2830 (2011).

MathSciNet   Google Scholar  

Daylight Theory. SMARTS - a language for describing molecular patterns. Daylight Chemical Information Systems Inc. www.daylight.com/dayhtml/doc/theory/theory.smarts.html (2022).

Wildman, S. A. & Crippen, G. M. Prediction of physicochemical parameters by atomic contributions. J. Chem. Inf. Comput. Sci. 39 , 868–873 (1999).

Swanson, K. et al. Generative AI for designing and validating easily synthesizable and structurally novel antibiotics: data and models. Zenodo https://doi.org/10.5281/zenodo.10257839 (2023).

Swanson, K. & Liu, G. swansonk/SyntheMol: SyntheMol. Zenodo https://doi.org/10.5281/zenodo.10278151 (2023).

Liu, G. et al. Deep learning-guided discovery of an antibiotic targeting Acinetobacter baumannii . Nat. Chem. Biol. 19 , 1342–1350 (2023).

Download references

Acknowledgements

This research was kindly supported by the Weston Family Foundation (POP and Catalyst to J.M.S.); the David Braley Centre for Antibiotic Discovery (J.M.S.); the Canadian Institutes of Health Research (J.M.S.); a generous gift from M. and M. Heersink (J.M.S.) and the Chan-Zuckerberg Biohub (J.Z.). We thank Y. Moroz for his help accessing and answering our questions about the Enamine REAL Space building blocks, reactions and molecules. We thank G. Dubinina for help obtaining generated compounds. We thank M. Karelina, J. Miguel Hernández-Lobato, A. Tripp and M. Segler for insightful discussions about our property prediction and generative methods. We thank J. Boyce and A. Wright for providing mutant strains of A. baumannii used in this study. K.S. acknowledges support from the Knight-Hennessy scholarship.

Author information

These authors contributed equally: Kyle Swanson, Gary Liu.

Authors and Affiliations

Department of Computer Science, Stanford University, Stanford, CA, USA

Kyle Swanson & James Zou

Department of Biochemistry and Biomedical Sciences, Michael G. DeGroote Institute for Infectious Disease Research, David Braley Centre for Antibiotic Discovery, McMaster University, Hamilton, Ontario, Canada

Gary Liu, Denise B. Catacutan, Autumn Arnold & Jonathan M. Stokes

Department of Biomedical Data Science, Stanford University, Stanford, CA, USA

You can also search for this author in PubMed   Google Scholar

Contributions

Conceptualization was carried out by K.S., G.L., J.Z. and J.M.S. Model development was performed by K.S. and G.L. Biological validation was carried out by D.B.C., A.A. and J.M.S. K.S., G.L., D.B.C., J.Z. and J.M.S. wrote the paper. J.Z. and J.M.S. supervised the work.

Corresponding authors

Correspondence to James Zou or Jonathan M. Stokes .

Ethics declarations

Competing interests.

These authors declare the following competing interests: K.S. is employed part-time by Greenstone Biosciences; J.Z. is on the scientific advisory board of Greenstone Biosciences and J.M.S. is cofounder and scientific director of Phare Bio. The other authors declare no competing interests.

Peer review

Peer review information.

Nature Machine Intelligence thanks Feixiong Cheng and the other, anonymous, reviewer(s) for their contribution to the peer review of this work.

Additional information

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Extended data

Extended data fig. 1 additional property prediction model development..

( a ) Normalized growth of A. baumannii ATCC 17978 in biological duplicate for each of the three training set chemical libraries. Note: compounds with normalized growth > 3 are removed for visual purposes. The R 2 values are the coefficient of determination. ( b ) Receiver operating characteristic (ROC) curves and ( c ) precision-recall (PRC) curves for the Chemprop, Chemprop-RDKit, and random forest models. For each model, the black lines show the performance of each of the ten models in the ensemble and the blue curve shows the average. ( d , e ) Model performance of each of our three property prediction models when generalizing between our three training set libraries. Values on the diagonal are the average test set performance of a model on a single library across tenfold cross-validation. Values on the off-diagonals are the result of applying an ensemble of ten models trained on one library to a different library and evaluating those predictions. ( d ) Performance measured by area under the receiver operating characteristic curve (ROC-AUC). ( e ) Performance measured by area under the precision-recall curve (PRC-AUC).

Extended Data Fig. 2 REAL Space Analysis, Comparisons to Training Set, and Reactions.

( a ) The cumulative percent of molecules in REAL Space that can be produced by each of the 169 REAL chemical reactions. ( b ) The percent of molecules in REAL Space that include each of the REAL building blocks. ( c ) The molecular weight distribution of a random sample of 25,000 REAL molecules (blue), the REAL building blocks (black), and our training set molecules (red). ( d ) The cLogP distribution of a random sample of 25,000 REAL molecules (blue), the REAL building blocks (black), and our training set molecules (red). ( e ) The remaining 5 REAL chemical reactions we used (first 8 in Fig. 4b ), along with the number and percent of REAL molecules produced by each reaction.

Extended Data Fig. 3 REAL Building Block and Full Molecule Scores from Chemprop-RDKit and Random Forest.

( a , b ) The distribution of antibacterial model scores on the REAL building blocks using the ( a ) Chemprop-RDKit or ( b ) random forest models. ( c , d ) The correlation between the antibacterial model score of a REAL molecule and the average antibacterial model score of its constituent building blocks using the ( c ) Chemprop-RDKit or ( d ) random forest models. The R 2 values are the coefficient of determination.

Extended Data Fig. 4 Comparison of Generated Sets with and without Building Block Diversity.

( a – c ) The frequency with which building blocks were used in the generated molecules of SyntheMol, with and without the building block diversity score penalty for ( a ) Chemprop, ( b ) Chemprop-RDKit, and ( c ) random forest.

Extended Data Fig. 5 Model Scores by Rollout from Chemprop-RDKit and Random Forest.

( a , b ) Violin plots of the distribution of antibacterial model scores for every 2,000 rollouts of the MCTS algorithm over 20,000 rollouts. SyntheMol uses the ( a ) Chemprop-RDKit or ( b ) random forest models for antibacterial prediction scores. The lines in each violin indicate the first quartile, the median, and the third quartile.

Extended Data Fig. 6 Additional Analysis of Chemprop Generated and Selected Sets.

( a ) The percent of building blocks that appear at different frequencies among the generated or selected compounds by SyntheMol with Chemprop. Building blocks are assigned to bins on the x-axis based on the number of generated or selected compounds that contain that building block, with the final bin including building blocks that appear in at least six compounds (max 137). ( b ) The distribution of chemical reactions used by the generated or selected compounds by SyntheMol with Chemprop. ( c ) A t-SNE visualization of the training set along with all generated and selected molecules from each of the three property predictor models.

Extended Data Fig. 7 Analysis of Chemprop-RDKit Generated and Selected Sets.

( a ) The percent of building blocks that appear at different frequencies among the generated or selected compounds by SyntheMol with Chemprop-RDKit. Building blocks are assigned to bins on the x-axis based on the number of generated or selected compounds that contain that building block, with the final bin including building blocks that appear in at least six compounds (max 185). ( b ) The distribution of chemical reactions used by the generated or selected compounds by SyntheMol with Chemprop-RDKit. ( c – f ) A comparison of the properties of the 25,828 molecules generated by SyntheMol with the Chemprop-RDKit antibacterial model and the 50 molecules selected from that set after applying post-hoc filters. ( c ) The distribution of nearest neighbour Tversky similarities between the generated or selected compounds and the active molecules in the training set. ( d ) The distribution of nearest neighbor Tversky similarities between the generated or selected compounds and the known antibacterial compounds from ChEMBL. ( e ) The distribution of Chemprop-RDKit antibacterial model scores on the generated or selected compounds, as well as on a random set of 25,000 REAL molecules. ( f ) The distribution of nearest neighbor Tanimoto similarities among the generated or selected compounds.

Extended Data Fig. 8 Analysis of Random Forest Generated and Selected Sets.

( a ) The percent of building blocks that appear at different frequencies among the generated or selected compounds by SyntheMol with random forest. Building blocks are assigned to bins on the x-axis based on the number of generated or selected compounds that contain that building block, with the final bin including building blocks that appear in at least six compounds (max 212). ( b ) The distribution of chemical reactions used by the generated or selected compounds by SyntheMol with random forest. ( c – f ) A comparison of the properties of the 27,396 molecules generated by SyntheMol with the random forest antibacterial model and the 50 molecules selected from that set after applying post-hoc filters. ( c ) The distribution of nearest neighbor Tversky similarities between the generated or selected compounds and the active molecules in the training set. ( d ) The distribution of nearest neighbor Tversky similarities between the generated or selected compounds and the known antibacterial compounds from ChEMBL. ( e ) The distribution of random forest antibacterial model scores on the generated or selected compounds as well as on a random set of 25,000 REAL molecules. ( f ) The distribution of nearest neighbor Tanimoto similarities among the generated or selected compounds.

Extended Data Fig. 9 Additional In Vitro Validation.

( a ) Gram-negative bacterial isolates tested for growth inhibition against SPR 741 or colistin. Experiments were performed in biological duplicate. Error bars represent absolute range of optical density measurements at 600 nm. ( b ) Heat map summarizing MICs of 58 randomly selected compounds from the REAL Space against A. baumannii ATCC 17978 in I) LB medium, II) LB medium + a quarter MIC SPR 741, and III) LB medium + a quarter MIC colistin. Compounds were tested at concentrations from 256 µg/mL to 4 µg/mL in two-fold serial dilutions. Lighter colours indicate lower MIC values for each random REAL molecule. No compounds displayed potent antibacterial activity using the threshold of MIC ≤ 8 µg/mL. Experiments were performed in at least biological duplicate. ( c , d ) Chequerboard analysis to quantify synergy, as defined by FICI, with SPR 741 or colistin against Gram-negative isolates. Chequerboard experiments were performed using two-fold serial dilution series with the maximum and minimum concentrations of the potentiator (x-axis) and compound (y-axis) shown in µg/mL. Darker blue represents higher bacterial growth. Experiments were performed in biological duplicate. The mean growth of each well is shown. ( c ) Chequerboard assays using the six bioactive compounds, in combination with colistin, against A. baumannii ATCC 17978. ( d ) Chequerboard assays using rifampicin – a control antibiotic – in combination with SPR 741 or colistin against a panel of Gram-negative bacterial species.

Extended Data Fig. 10 Toxicity Predictions.

Predictions of the probability of clinical toxicity using an ensemble of ten Chemprop-RDKit models trained on the ClinTox dataset. ‘Non-toxic compounds’ show the toxicity predictions of the model on the non-toxic molecules in the dataset (n = 1,372), where each molecule’s prediction score comes from the one model in the ensemble for which that molecule was in the test set. ‘Toxic compounds’ shows the same toxicity predictions for the toxic molecules in the dataset (n = 112). ‘Selected six’ shows the average prediction of the ensemble of ten models on the six potent generated molecules. Blue horizontal lines represent the mean predictions for each set.

Supplementary information

Supplementary information.

Supplementary Tables 1–6 and Extended Discussion.

Reporting Summary

Supplementary data 1–9.

Supplementary Data 1: Training sets for antibacterial activity and clinical toxicity. Supplementary Data 2: Antibiotic and antibacterial molecules from ChEMBL. Supplementary Data 3: Antibacterial Chemprop/Chemprop-RDKit/random forest models performance summary. Supplementary Data 4: Analysis of the Enamine REAL Space. Supplementary Data 5: cLogP Chemprop model performance summary trained for 1 and 30 epochs. Supplementary Data 6: Molecules generated by SyntheMol, using the antibacterial Chemprop model as a scoring function. Supplementary Data 7: Molecules generated by SyntheMol, using the antibacterial Chemprop-RDKit model as a scoring function. Supplementary Data 8: Molecules generated by SyntheMol, using the antibacterial random forest model as a scoring function. Supplementary Data 9: Molecules selected for synthesis by Enamine.

Supplementary Data

Quality control LC–MS data for SyntheMol-generated compounds.

Quality control LC–MS data for randomly selected control compounds.

Rights and permissions

Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.

Reprints and permissions

About this article

Cite this article.

Swanson, K., Liu, G., Catacutan, D.B. et al. Generative AI for designing and validating easily synthesizable and structurally novel antibiotics. Nat Mach Intell 6 , 338–353 (2024). https://doi.org/10.1038/s42256-024-00809-7

Download citation

Received : 10 March 2023

Accepted : 08 February 2024

Published : 22 March 2024

Issue Date : March 2024

DOI : https://doi.org/10.1038/s42256-024-00809-7

Share this article

Anyone you share the following link with will be able to read this content:

Sorry, a shareable link is not currently available for this article.

Provided by the Springer Nature SharedIt content-sharing initiative

Quick links

• Explore articles by subject
• Guide to authors
• Editorial policies

Sign up for the Nature Briefing newsletter — what matters in science, free to your inbox daily.

Considering the Environmental Impacts of Generative AI to Spark Responsible Development

April 10, 2024

New paper by MIT researchers explores the key drivers of Gen-AI's growth and why sustainability considerations are lacking – calling for more responsible development of Gen-AI using a comparative benefit-cost evaluation framework.

Use of generative artificial intelligence (Gen-AI) is expanding rapidly, and as with many large-scale technology-induced shifts, its benefits and high demand are taking center stage. While the perceived advantages are vast, neglecting to consider the negative effects alongside these potential benefits can lead to uncontrolled growth with lasting consequences on the environment. A new open access paper by an interdisciplinary team of MIT researchers explores some of the key drivers of this growth and why sustainability considerations for Gen-AI are lacking – ultimately calling for more responsible development of Gen-AI using a comparative benefit-cost evaluation framework. Published as a preprint by MIT Press, “ The Climate and Sustainability Implications of Generative AI ” highlights how the excitement around Gen-AI is leading to an incomplete consideration of value that ignores the potential costs. The resulting social and environmental impacts require detailed analysis, coordination, innovation, and adoption across diverse stakeholders to steer the direction toward responsible growth.

The power of potential at the cost of reality

Gen-AI’s perceived benefits are powerful. The reality of the environmental impacts, which include expanding demand for computing power, larger carbon footprints, shifts in patterns of electricity demand, and an accelerated depletion of natural resources, is largely overshadowed by the exciting potentials, such as enhanced productivity and boundless creativity, and a narrow focus on energy efficiency. While stakeholders across industries have invested in Gen-AI’s promising capabilities and transformative impact, potential does not always equate to practical outcomes.

And while there have been efficiency improvements in the AI space, these gains have not reduced Gen-AI’s overall energy consumption because of voracious demand and opportunity. Continuing to strive for efficiency gains is an important pre-requisite, but such gains will not solve the problem alone.

The paper’s authors – Noman Bashir , Priya Donti , James Cuff , Sydney Sroka , Marija Ilic , Vivienne Sze , Christina Delimitrou , and Elsa Olivetti – argue that responsible development of Gen-AI requires a focus on sustainability beyond only efficiency improvements, and that a narrow focus on energy efficiency can make Gen-AI’s challenges even more pronounced.

“The growth of Gen-AI is driving increased electricity demand, which runs counter to necessary efficiency gains to achieve net-zero greenhouse gas emissions,” said Noman Bashir, who leads the MCSC Data & Computing pathway. “Our current capacity to build sustainably also cannot keep pace with the datacenter construction necessary to support Gen-AI. Furthermore, this explosive growth exacerbates supply chain challenges, affecting essential goods and services that rely on the tech industry and potentially creating macroeconomic impacts.”

Sparking responsible growth

This is where the team’s proposal of a benefit–cost evaluation framework comes in.

Their framework has three elements:

• defining the scope and boundaries by aligning with life cycle assessment (LCA) methodology, but also leveraging methods beyond LCA;
• developing baseline and scenarios from which comparisons can be made to quantify the benefit as well as the burden; and
• building data inventory for accounting for the cost of operating and manufacturing computing devices and infrastructure, the benefit and cost of the Gen-AI application itself (such as a search query), as well as the benefit and cost at a societal level.

Implementing a robust life cycle assessment framework will prove complex and fraught with uncertainty, argue the researchers, but it is possible. Even taking steps towards quantifying the benefits and costs of Gen-AI would make a positive difference by providing the transparency needed to begin instilling accountability for more sustainable practices.

Developing the specifics of this framework would engage a wide range of stakeholders (e.g. technical and sociotechnical experts, corporate entities, policymakers, and civil society) who would each provide necessary considerations for social and environmental impacts into technological advancement assessments. In the paper, the authors dive into a range of stakeholders and identify action items catered to their specific roles, to build their proposed benefit–cost evaluation capability.

“The capability we describe would promote Gen-AI to develop in ways that can support social and environmental sustainability goals alongside economic opportunity, which is the type of responsible growth we are calling for,” added Bashir.

Researchers say the framework would be responsive to legislation designed to investigate the impacts of AI, such as the new bill put forth by Senator Edward J. Markey, Senator Martin Heinrich, Representative Anna Eshoo, and Representative Don Beyer. This newly introduced Artificial Intelligence Environmental Impacts Act of 2024 would investigate and measure the environmental impacts of AI – and the framework put forth in the MIT team’s paper could accelerate implementation.

A collaborative approach

The challenges surrounding Gen-AI’s rapid growth and its vast implications on energy use and emissions are multifaceted and complex, and addressing them requires deep, technical collaborations across disciplines and areas of expertise. In writing “The Climate and Sustainability Implications of Generative AI,” the authors drew on their different backgrounds and experiences to explore these challenges as they pertain to the information and communications technology sector – from computing algorithms and hardware to data centers all the way to networks and power systems. They take an integrated systems perspective to make progress toward reduced computational requirements for existing machine learning approaches. This perspective helps account for how innovations in computing algorithms might impact chip fabrication or how data center design and architecture enhance or hinder network or electrical grid optimization.

The process of writing “The Climate and Sustainability Implications of Generative AI” included pulling from a wide range of sources and drawing links across them, including two of the other selected MIT papers on Gen-AI , underscoring the MCSC’s collaborative approach to its work. In exploring the framework’s initial applications, the authors drew from some of the key concepts outlined in “ Closing the Execution Gap in Generative AI for Chemicals and Materials: Freeways or Safeguards ” and “ Learning from Nature to Achieve Material Sustainability: Generative AI for Rigorous Bio-Inspired Materials Design .” The first provided important background on incorporating the ways in which Gen-AI accelerates the determination of process pathways, reaction yields, synthesis byproducts, device performance, and manufacturing process into scenario and baseline development. The second was valuable in thinking about the applications of Gen-AI on discovery related to chemicals, such as innovations to advance chemicals for catalysis, more recyclable or biodegradable materials.

Within the MCSC, the theme of Data & Computing has been an ongoing emphasis through workshops and programming that engage member companies and the MIT community. In 2022, the MCSC hosted a two-day workshop on the Climate Implications of Computing & Communications, in collaboration with MIT-IBM Watson AI Lab and Schwarzman College of Computing, attracting hundreds of participants from academia and industry. The workshop presentations explored a host of energy-efficiency options, including specialized chip design, data center architecture, better algorithms, hardware modifications, and changes in consumer behavior. Later that year, the climate implications of computing was a central theme of the MCSC’s Annual Symposium.

The opportunity for furthering responsible use of Gen-AI is woven into all of the impact pathways , as increasingly, Gen-AI and foundation models are finding applications in such domains. The benefit-cost framework described in the paper can be leveraged to account for the implications of using machine learning and Gen-AI in sustainability applications such as monitoring soil carbon accrual, assessing the effectiveness of carbon offsets , tracking materials flows to drive beneficial circularity , or to build more resilient supply chains. As we apply Gen-AI methods in a broad set of sustainability cases, we must ensure the benefits outweigh the costs.

Investigating the social implications of Gen-AI at MIT

Last fall, MIT President Sally Kornbluth and Provost Cynthia Barnhart provided more than two dozen research teams from across MIT with seed funds  to develop impact papers that would articulate effective roadmaps, policy recommendations, and calls for action across the broad domain of generative AI and its effects on society. “The Climate and Sustainability Implications of Generative AI” was one of 27 across the Institute selected in the first round of funding. A second round of seed grants were awarded to 16 additional finalists this spring. The open access papers have been published as preprints online under the auspices of the MIT Open Publishing Services program from the MIT Press.

Get in the Know on Generative AI

Research report featuring decision-maker insights.

Research Report

Generative AI Challenges and Potential Unveiled: How to Achieve a Competitive Advantage

Generative AI is transforming the workplace – and society – at a whirlwind pace.

We asked 300 US-based decision makers about the challenges and opportunities their organizations face as they implement GenAI.

This report shares the results of this survey and AI best practices, including the optimal way to use the technology while accelerating innovation across your business, protecting your data and cultivating trust in changing times.

In this report, you will learn:

• The strategic investments US companies are making to harness the unparalleled potential of generative AI.
• How to identify your best generative AI use cases to deliver speedy return on investment.
• What data and AI techniques can help augment the accuracy and explainability of large language models (LLMs).
• How you can combine generative AI with an AI-based decisioning workflow system to facilitate measurable outcomes.
• How to proactively prepare your organization in an era of exponential innovation.

Interested in a global view? This US report is part of a global survey of 1,600 decision makers worldwide that is due out in July.

See how your peers worldwide are preparing their organizations for GenAI. Just check the box at the bottom of the registration form for the US report and you'll be alerted when the global results are available.

Simply complete the form below, and receive these valuable insights.

Have a SAS profile? To complete this form automatically Sign In

Edit Profile Log Out

Asking the better questions that unlock new answers to the working world's most complex issues.

Trending topics

AI insights

EY podcasts

EY webcasts

Operations leaders

Technology leaders

Marketing and growth leaders

Cybersecurity and privacy leaders

Risk leaders

EY Center for Board Matters

EY helps clients create long-term value for all stakeholders. Enabled by data and technology, our services and solutions provide trust through assurance and help clients transform, grow and operate.

Artificial Intelligence (AI)

Strategy, transaction and transformation consulting

Technology transformation

Tax function operations

Climate change and sustainability services

EY Ecosystems

Supply chain and operations

EY Partner Ecosystem

Explore Services

We bring together extraordinary people, like you, to build a better working world.

Experienced professionals

MBA and advanced-degree students

Student and entry level programs

Contract workers

EY-Parthenon careers

Discover how EY insights and services are helping to reframe the future of your industry.

Case studies

Energy and resources

How data analytics can strengthen supply chain performance

13-Jul-2023 Ben Williams

How Takeda harnessed the power of the metaverse for positive human impact

26-Jun-2023 Edwina Fitzmaurice

Banking and Capital Markets

How cutting back infused higher quality in transaction monitoring

11-Jul-2023 Ron V. Giammarco

At EY, our purpose is building a better working world. The insights and services we provide help to create long-term value for clients, people and society, and to build trust in the capital markets.

EY is now carbon negative

19-Sep-2022 Carmine Di Sibio

Our commitment to audit quality

13-Nov-2023 Julie A. Boland

No results have been found

Recent Searches

BEPS 2.0: as policies evolve, engagement is key

It remains to be seen whether the US will align its tax law with the OECD/G20’s global BEPS 2.0 rules. MNEs will feel the impact in 2024. Learn more.

How GenAI strategy can transform innovation

Companies considering or investing in a transformative GenAI strategy should tie generative artificial intelligence use cases to revenue, cost and expense. Learn more

Top five private equity trends for 2024

Read about the five key trends private equity firms will emphasize in 2024 as they create value

Select your location

close expand_more

EY Survey: AI adoption among financial services leaders universal, amid mixed signals of readiness

Press contact

Multidisciplinary professional services organization

• Send e-mail to EY Americas

One in five are not confident that their organizations are well-positioned to take advantage of the potential benefits AI might bring

Ernst & Young LLP (EY US) today released its 2023 Financial Services GenAI Survey. Nearly all (99%) of the financial services leaders surveyed reported that their organizations were deploying artificial intelligence (AI) in some manner, and all respondents said they are either already using, or planning to use, generative AI (GenAI) specifically within their organization.

Amid these universal adoption plans, just over one in five respondents said they are nervous or skeptical about the potential impact of GenAI on their organization. That same percentage also does not feel confident that their organizations are well-positioned to take advantage of the potential benefits AI might bring. Insurance is ranked highest in the nervous or skeptical category (24%), while banking and capital markets leaders had the lowest levels of skepticism (17%), followed by wealth and asset management (21%).

“While there’s a whole world of possibilities and efficiencies AI can create for financial services in areas ranging from data analysis to customer service optimization, blind optimism and hype around the technology can ultimately have a counterproductive impact on a business,” says  David Kadio-Morokro , EY Americas Financial Services Innovation Leader. “We like to take an ‘innovation intelligence’ approach to putting artificial intelligence to work — planning, education and an agile test and learn strategy for implementation are imperative for those looking to make the most of AI’s potential benefits.”

AI Outlook: Optimism Now and in the Future

The majority of respondents felt positive about AI, with more than half (55%) saying they felt supportive and optimistic about using AI in their organization. The long-term sentiment is even more optimistic, with 77% of executives viewing GenAI as an overall benefit to the financial services industry in the next 5-10 years. Leaders see a particular opportunity in customer and client experience, with 87% stating that they believe AI can bring improvements to this space.

AI Implementation: It’s Time to Get Back to Basics

Taking advantage of, and deploying, GenAI is not without its hurdles. When asked about the challenges financial organizations will face in taking advantage of GenAI, leaders ranked the following as top barriers:

• 40% said lack of proper data infrastructure and 35% lack technology infrastructure
• 36% said lack of clear commitment from leadership
• 33% said unclear governance and ethical framework

“Generative AI holds the potential to revolutionize a broad array of business functions,” says  Sameer Gupta , EY Americas Financial Services Organization Advanced Analytics Leader. “With each new wave of AI and analytic innovation, it becomes increasingly clear how important it is to have a tech stack with a solid foundation. Our role is to support financial services organizations in making sure their legacy data and technologies are unimpeachable before adding AI applications on top of existing systems.”

If organizations truly want to take advantage of AI’s benefits, a major factor in their success will be a better understanding of, and improvements in, their data infrastructure. The survey reflects this, with leaders citing the following as the top three ways AI will benefit the financial services industry:

• 46% said risk reduction from data processing
• 38% said creation of new offerings and hyper-personalized marketing
• 37% said improving data management process and accuracy

Winning the (AI) Talent Battle

“Focusing on the human role of AI implementation is just as important as technology infrastructure,” says EY Americas Financial Services Accounts Managing Partner  Michael Fox . “Our data showed that 44% of leaders cited access to skilled resources as a barrier to AI implementation, but there’s only so many already skilled professionals in existence. Part of the solution is deploying upskilling programs today that can equip your current workforce with the skills they need to help leaders and their business thrive in an increasingly AI-centric world.”

Implementing AI in financial services will undoubtedly change the way sector leaders learn, serve customers, process data and manage risk. Organizations that act now with investments in training, talent and infrastructure development while putting proper governance and controls in place will reap the benefits that AI offers their industry, continue driving the business case for investment and will likely find themselves to be a step ahead of their peers.

To find out more about EY and GenAI in Financial Services, including the recently announced unifying platform that the EY organization has created to help organizations confidently adopt AI, please visit  EY.ai .

EY exists to build a better working world, helping create long-term value for clients, people and society and build trust in the capital markets.

Enabled by data and technology, diverse EY teams in over 150 countries provide trust through assurance and help clients grow, transform and operate.

Working across assurance, consulting, law, strategy, tax and transactions, EY teams ask better questions to find new answers for the complex issues facing our world today.

EY refers to the global organization, and may refer to one or more, of the member firms of Ernst & Young Global Limited, each of which is a separate legal entity. Ernst & Young Global Limited, a UK company limited by guarantee, does not provide services to clients. Information about how EY collects and uses personal data and a description of the rights individuals have under data protection legislation are available via ey.com/privacy. EY member firms do not practice law where prohibited by local laws. For more information about our organization, please visit  ey.com .

Ernst & Young LLP is a client-serving member firm of Ernst & Young Global Limited operating in the US.

Methodology

Wakefield Research conducted this survey between August 15–28, 2023, among 300 Executive Directors, Managing Directors or higher at financial organizations with $2b+ in revenue. Financial organizations are defined as across the banking and capital markets, insurance, and wealth and asset management industries, with 100 responses collected from each sector.

• Connect with us
• Our locations
• Do Not Sell or Share My Personal Information
• Legal and privacy
• Accessibility
• Open Facebook profile
• Open X profile
• Open LinkedIn profile
• Open Youtube profile

EY refers to the global organization, and may refer to one or more, of the member firms of Ernst & Young Global Limited, each of which is a separate legal entity. Ernst & Young Global Limited, a UK company limited by guarantee, does not provide services to clients.

• Download PDF
• Share X Facebook Email LinkedIn
• Permissions

Clinical Reasoning of a Generative Artificial Intelligence Model Compared With Physicians

• 1 Department of Medicine, Beth Israel Deaconess Medical Center, Boston, Massachusetts
• 2 Department of Medicine, Massachusetts General Hospital, Boston
• 3 Department of Pulmonary and Critical Care, Brigham and Women’s Hospital, Boston, Massachusetts

Large language models (LLMs) have shown promise in clinical reasoning, but their ability to synthesize clinical encounter data into problem representations remains unexplored. 1 - 3 We compared an LLM’s reasoning abilities against human performance using standards developed for physicians.

Read More About

Cabral S , Restrepo D , Kanjee Z, et al. Clinical Reasoning of a Generative Artificial Intelligence Model Compared With Physicians. JAMA Intern Med. Published online April 01, 2024. doi:10.1001/jamainternmed.2024.0295

Manage citations:

© 2024

Artificial Intelligence Resource Center

Best of JAMA Network 2022

Browse and subscribe to JAMA Network podcasts!

Others Also Liked

Select your interests.

Customize your JAMA Network experience by selecting one or more topics from the list below.

• Academic Medicine
• Acid Base, Electrolytes, Fluids
• Allergy and Clinical Immunology
• American Indian or Alaska Natives
• Anesthesiology
• Anticoagulation
• Art and Images in Psychiatry
• Artificial Intelligence
• Assisted Reproduction
• Bleeding and Transfusion
• Caring for the Critically Ill Patient
• Challenges in Clinical Electrocardiography
• Climate and Health
• Climate Change
• Clinical Challenge
• Clinical Decision Support
• Clinical Implications of Basic Neuroscience
• Clinical Pharmacy and Pharmacology
• Complementary and Alternative Medicine
• Consensus Statements
• Coronavirus (COVID-19)
• Critical Care Medicine
• Cultural Competency
• Dental Medicine
• Dermatology
• Diabetes and Endocrinology
• Diagnostic Test Interpretation
• Drug Development
• Electronic Health Records
• Emergency Medicine
• End of Life, Hospice, Palliative Care
• Environmental Health
• Equity, Diversity, and Inclusion
• Facial Plastic Surgery
• Gastroenterology and Hepatology
• Genetics and Genomics
• Genomics and Precision Health
• Global Health
• Guide to Statistics and Methods
• Hair Disorders
• Health Care Delivery Models
• Health Care Economics, Insurance, Payment
• Health Care Quality
• Health Care Reform
• Health Care Safety
• Health Care Workforce
• Health Disparities
• Health Inequities
• Health Policy
• Health Systems Science
• History of Medicine
• Hypertension
• Images in Neurology
• Implementation Science
• Infectious Diseases
• Innovations in Health Care Delivery
• JAMA Infographic
• Law and Medicine
• Leading Change
• Less is More
• LGBTQIA Medicine
• Lifestyle Behaviors
• Medical Coding
• Medical Devices and Equipment
• Medical Education
• Medical Education and Training
• Medical Journals and Publishing
• Mobile Health and Telemedicine
• Narrative Medicine
• Neuroscience and Psychiatry
• Notable Notes
• Nutrition, Obesity, Exercise
• Obstetrics and Gynecology
• Occupational Health
• Ophthalmology
• Orthopedics
• Otolaryngology
• Pain Medicine
• Palliative Care
• Pathology and Laboratory Medicine
• Patient Care
• Patient Information
• Performance Improvement
• Performance Measures
• Perioperative Care and Consultation
• Pharmacoeconomics
• Pharmacoepidemiology
• Pharmacogenetics
• Pharmacy and Clinical Pharmacology
• Physical Medicine and Rehabilitation
• Physical Therapy
• Physician Leadership
• Population Health
• Primary Care
• Professional Well-being
• Professionalism
• Psychiatry and Behavioral Health
• Public Health
• Pulmonary Medicine
• Regulatory Agencies
• Reproductive Health
• Research, Methods, Statistics
• Resuscitation
• Rheumatology
• Risk Management
• Scientific Discovery and the Future of Medicine
• Shared Decision Making and Communication
• Sleep Medicine
• Sports Medicine
• Stem Cell Transplantation
• Substance Use and Addiction Medicine
• Surgical Innovation
• Surgical Pearls
• Teachable Moment
• Technology and Finance
• The Art of JAMA
• The Arts and Medicine
• The Rational Clinical Examination
• Tobacco and e-Cigarettes
• Translational Medicine
• Trauma and Injury
• Treatment Adherence
• Ultrasonography
• Users' Guide to the Medical Literature
• Vaccination
• Venous Thromboembolism
• Veterans Health
• Women's Health
• Workflow and Process
• Wound Care, Infection, Healing
• Register for email alerts with links to free full-text articles
• Access PDFs of free articles
• Manage your interests
• Save searches and receive search alerts

To revisit this article, visit My Profile, then View saved stories .

• Backchannel
• Newsletters
• WIRED Insider
• WIRED Consulting

Amanda Hoover

Students Are Likely Writing Millions of Papers With AI

Students have submitted more than 22 million papers that may have used generative AI in the past year, new data released by plagiarism detection company Turnitin shows.

A year ago, Turnitin rolled out an AI writing detection tool that was trained on its trove of papers written by students as well as other AI-generated texts. Since then, more than 200 million papers have been reviewed by the detector, predominantly written by high school and college students. Turnitin found that 11 percent may contain AI-written language in 20 percent of its content, with 3 percent of the total papers reviewed getting flagged for having 80 percent or more AI writing. (Turnitin is owned by Advance, which also owns Condé Nast, publisher of WIRED.) Turnitin says its detector has a false positive rate of less than 1 percent when analyzing full documents.

ChatGPT’s launch was met with knee-jerk fears that the English class essay would die . The chatbot can synthesize information and distill it near-instantly—but that doesn’t mean it always gets it right. Generative AI has been known to hallucinate , creating its own facts and citing academic references that don’t actually exist. Generative AI chatbots have also been caught spitting out biased text on gender and race . Despite those flaws, students have used chatbots for research, organizing ideas, and as a ghostwriter . Traces of chatbots have even been found in peer-reviewed, published academic writing .

Teachers understandably want to hold students accountable for using generative AI without permission or disclosure. But that requires a reliable way to prove AI was used in a given assignment. Instructors have tried at times to find their own solutions to detecting AI in writing, using messy, untested methods to enforce rules , and distressing students. Further complicating the issue, some teachers are even using generative AI in their grading processes.

Detecting the use of gen AI is tricky. It’s not as easy as flagging plagiarism, because generated text is still original text. Plus, there’s nuance to how students use gen AI; some may ask chatbots to write their papers for them in large chunks or in full, while others may use the tools as an aid or a brainstorm partner.

Students also aren't tempted by only ChatGPT and similar large language models. So-called word spinners are another type of AI software that rewrites text, and may make it less obvious to a teacher that work was plagiarized or generated by AI. Turnitin’s AI detector has also been updated to detect word spinners, says Annie Chechitelli, the company’s chief product officer. It can also flag work that was rewritten by services like spell checker Grammarly, which now has its own generative AI tool . As familiar software increasingly adds generative AI components, what students can and can’t use becomes more muddled.

Detection tools themselves have a risk of bias. English language learners may be more likely to set them off; a 2023 study found a 61.3 percent false positive rate when evaluating Test of English as a Foreign Language (TOEFL) exams with seven different AI detectors. The study did not examine Turnitin’s version. The company says it has trained its detector on writing from English language learners as well as native English speakers. A study published in October found that Turnitin was among the most accurate of 16 AI language detectors in a test that had the tool examine undergraduate papers and AI-generated papers.

Schools that use Turnitin had access to the AI detection software for a free pilot period, which ended at the start of this year. Chechitelli says a majority of the service’s clients have opted to purchase the AI detection. But the risks of false positives and bias against English learners have led some universities to ditch the tools for now. Montclair State University in New Jersey announced in November that it would pause use of Turnitin’s AI detector. Vanderbilt University and Northwestern University did the same last summer.

“This is hard. I understand why people want a tool,” says Emily Isaacs, executive director of the Office of Faculty Excellence at Montclair State. But Isaacs says the university is concerned about potentially biased results from AI detectors, as well as the fact that the tools can’t provide confirmation the way they can with plagiarism. Plus, Montclair State doesn’t want to put a blanket ban on AI, which will have some place in academia. With time and more trust in the tools, the policies could change. “It’s not a forever decision, it’s a now decision,” Isaacs says.

Chechitelli says the Turnitin tool shouldn’t be the only consideration in passing or failing a student. Instead, it’s a chance for teachers to start conversations with students that touch on all of the nuance in using generative AI. “People don’t really know where that line should be,” she says.

You Might Also Like …

In your inbox: The best and weirdest stories from WIRED’s archive

Jeffrey Epstein’s island visitors exposed by data broker

8 Google employees invented modern AI. Here’s the inside story

The crypto fraud kingpin who almost got away

Listen up! These are the best podcasts , no matter what you’re into