AI Wiki

Explore AI Wiki

Citation and evidence

Eric Horvitz

25 min full readUpdated 61 references

This article's verification

Report a problem with this article

More

Use this article

Raw MarkdownExplore connections

Improve this page

Suggest editRevision historyDiscussion

Browse categories

AI ResearchAI SafetyMicrosoftPeople

Cite this article

Eric Horvitz is an American computer scientist, holding both an MD and a PhD, who is the Chief Scientific Officer of Microsoft, a newly created post he has held since March 2020.[1][2] He joined Microsoft Research in 1993 and, before becoming Chief Scientific Officer, was director of Microsoft Research Labs, overseeing labs in Redmond, Cambridge (Massachusetts), New York City, Montreal, Cambridge (UK) and Bangalore.[1][2] His research centers on reasoning and decision making under uncertainty and with limited computational resources, Bayesian user modeling, and collaboration between people and AI systems; one of his early projects, Lumiere, served as the basis for the Office Assistant in Microsoft Office 97.[1][12] Outside the lab he founded the One Hundred Year Study on Artificial Intelligence (AI100) at Stanford, co-founded the Partnership on AI, served as president of the Association for the Advancement of Artificial Intelligence (AAAI), was a commissioner on the U.S. National Security Commission on AI, and served on President Joe Biden's Council of Advisors on Science and Technology (PCAST).[1][36] In September 2026 he was one of 22 authors of the working paper "What if automating AI R&D triggers an intelligence explosion?"[45]

Eric Horvitz
Current roleChief Scientific Officer, Microsoft (since March 2020)[2]
Joined Microsoft1993, as a principal researcher[2]
Previous Microsoft rolesManaging director, Microsoft Research Redmond; director, Microsoft Research Labs[9][2]
EducationUndergraduate studies in biophysics, Binghamton University; PhD and MD, Stanford University[6][1]
Doctoral dissertationComputation and Action Under Bounded Resources (1990)[5]
Research areasDecision making under uncertainty, bounded rationality, Bayesian user modeling, human-AI collaboration, AI in medicine[1]
Society leadershipAAAI president, 2007-2009[27]
Selected honorsNAE (2013), American Academy of Arts and Sciences (2011), ACM Fellow (2014), AAAI Feigenbaum Prize (2015), ACM-AAAI Allen Newell Award (2015)[49][50][53][52][51]

Expanded article table

Education and early career

Horvitz did undergraduate studies in biophysics at Binghamton University, where, according to a 2025 account of a talk he gave at MIT, he became interested in neuroscience; that interest carried into MD and PhD studies at Stanford University.[6] His interest in AI began with graduate classes on its technical and philosophical sides, and one question that drew him in was how an AI system should behave in time-critical, high-stakes settings such as the care of a dying patient.[6] His doctoral dissertation, Computation and Action Under Bounded Resources, was completed in 1990.[5] Stanford's announcement of AI100 in 2014 described him as a Stanford alumnus and said that Russ Altman, whom it named as AI100's faculty director, had studied computer science and medicine with Horvitz at Stanford in the late 1980s.[7]

Horvitz came to Microsoft with David Heckerman and Jack Breese. In Microsoft's oral history of its research lab, Heckerman recalled suggesting in 1992 that Nathan Myhrvold consider buying "one of the companies I had formed with Eric Horvitz and Jack Breese," and said Myhrvold brought all three to Redmond "under the guise of considering our company for purchase"; the interviews persuaded all three to join Microsoft Research. A colleague in the same history recalled the three arriving "from Knowledge Industries with Bayesian networks in their back pockets."[8] Microsoft Research colleagues in the same history credit the trio with starting the lab's machine-learning work and its Bayesian approach to handling uncertainty.[8]

Microsoft career

Horvitz started at Microsoft in 1993 as a principal researcher.[2] CNBC, citing his LinkedIn profile, reported in 2020 that he spent 16 years in that role.[3] By the mid-2010s he was managing director of Microsoft Research's Redmond lab: Microsoft Research used that title for him in January 2015, and Senator Ted Cruz's office listed him as "Managing Director, Microsoft Research Lab" in November 2016.[9][34] He later became director of Microsoft Research Labs, which his official biography describes as overseeing labs in Redmond, Cambridge (Massachusetts), New York City, Montreal, Cambridge (UK) and Bangalore.[1][2] GeekWire and CNBC also described him as a technical fellow at the time of his 2020 promotion.[2][3]

On March 11, 2020, Microsoft named Horvitz its first Chief Scientific Officer, a newly created role, and put Peter Lee, then corporate vice president of Microsoft Healthcare, in charge of Microsoft Research.[2][3] The reorganization followed the departure in February 2020 of Harry Shum, who had led Microsoft's AI and Research group.[3] A Microsoft spokesperson told CNBC that as chief scientist Horvitz would "provide cross-company leadership on advances and trends on scientific matters, and on issues and opportunities rising at the intersection of technology, people and society," and that his organization would advise on the company's scientific directions and investments.[3] As of September 2026, Microsoft's research site describes him as leading "initiatives at the intersection of science, technology, and society, with emphases on artificial intelligence, biosciences, and healthcare."[1]

PeriodRoleSource
1993Joined Microsoft Research as a principal researcher[2]
By January 2015 to at least November 2016Managing director, Microsoft Research Redmond[9][34]
Before March 2020Technical fellow and director, Microsoft Research Labs[2][3]
2016 onwardCo-founder (with Brad Smith) and chair of Microsoft's Aether committee[32]
March 2020 to present (as of September 2026)Chief Scientific Officer, Microsoft[2][1]

Expanded article table

Research contributions

Microsoft's biography summarizes his technical work as "harnessing probability and utility in machine learning and reasoning, developing models of bounded rationality, constructing systems that perceive and act via interpreting multisensory streams of information," and developing principles for human-AI collaboration, with fielded systems in healthcare, transportation, e-commerce, operating systems and aerospace.[1]

Reasoning under bounded resources

A thread that runs from his doctoral work onward is how a reasoning system with limited time and computation should decide how much to think before acting. His 1987 paper at the Third Conference on Uncertainty in Artificial Intelligence (UAI), "Reasoning About Beliefs and Actions Under Computational Resource Constraints," argued that when a complete normative analysis is impossible, a system should weigh "the expected costs and benefits of applying alternative approximation procedures and heuristics," and it introduced a notion of "real-time rationality."[10] In his Microsoft research statement he lists flexible or anytime computation, "ideal metareasoning for guiding computation," continual computation and bounded-optimal reasoning systems among the areas he has worked on.[61] The National Academy of Engineering cited this line of work when it elected him in 2013, "for computational mechanisms for decision making under uncertainty and with bounded resources."[49]

The same ideas were applied to interfaces for people making urgent decisions. A 1995 UAI paper with Matthew Barry introduced the "expected value of revealed information" and "expected value of displayed information" as measures for deciding what to show operators, motivated by work on decision-theoretic display control "for a time-critical monitoring application at the NASA Mission Control Center in Houston."[11] In 2015, with Samuel Gershman and Joshua Tenenbaum, he co-wrote "Computational rationality: A converging paradigm for intelligence in brains, minds, and machines" in Science.[17]

Bayesian user modeling: Lumiere and the Office Assistant

The Lumiere Project, described in a UAI 1998 paper by Horvitz, Jack Breese, David Heckerman, David Hovel and Koos Rommelse, built Bayesian models that infer a software user's goals and needs from the user's background, actions and queries. The paper covers Bayesian models of time-varying user goals, an event stream from applications, a language for turning system events into observations, persistent user-expertise profiles and an overall architecture for an intelligent interface. According to the paper, "Lumiere prototypes served as the basis for the Office Assistant in the Microsoft Office '97 suite of productivity applications."[12]

Mixed-initiative interaction, attention and alerting

His CHI 1999 paper "Principles of mixed-initiative user interfaces" addressed the debate between direct-manipulation interfaces and automated interface agents, reviewing principles for "an elegant coupling of automated services with direct manipulation," illustrated with the LookOut system for scheduling and meeting management.[13] A related UAI 1999 paper with Andy Jacobs and David Hovel, "Attention-Sensitive Alerting," set out utility-directed procedures that balance the cost of deferring an alert against the cost of interrupting a user, and described the Priorities system for ranking email by inferred criticality.[14] His Microsoft research statement lists "information triage and alerting that takes human attention into consideration," notification systems, surprise modeling and "psychological studies of interruption and recovery" among his interests, and describes leveraging "the complementarities of human and machine intelligence" as a key focus.[61]

Other fielded applications

  • Spam filtering. He is one of the four authors, with Mehran Sahami, Susan Dumais and David Heckerman, of "A Bayesian Approach to Filtering Junk E-Mail" (AAAI workshop, 1998), an early scholarly treatment of Bayesian spam filtering.[15] In a January 2026 essay he recalled the late 1990s and early 2000s as the period when "we were excited to launch the first machine-learning--powered junk-email filters and email urgency classifiers."[18]
  • Traffic prediction. A UAI 2005 paper, "Prediction, Expectation, and Surprise: Methods, Designs, and Study of a Deployed Traffic Forecasting Service," described a traffic forecasting service in public use.[16] The 2015 ACM citation for his Allen Newell Award credits him with "intelligent cloud services that make predictions about road traffic patterns and provide ideal route directions."[51]
  • Healthcare. The same citation credits computational models that help physicians with decisions about readmissions and infections.[51] Horvitz has described the Readmissions Manager as a system giving physicians "site-specific predictions about patients at highest risk for readmission within 30 days of hospital discharge," built with machine learning on local electronic health records.[18]

Large language models and medicine

Horvitz was among the 14 authors of "Sparks of Artificial General Intelligence: Early experiments with GPT-4" (March 2023), led by Sébastien Bubeck, which studied an early version of GPT-4 and argued that it was part of a new cohort of models showing "more general intelligence than previous AI models."[19] Two days earlier he was the last author of "Capabilities of GPT-4 on Medical Challenge Problems," which evaluated GPT-4 on official USMLE practice materials and the MultiMedQA benchmark suite without specialized prompting.[20] Later in 2023 he co-authored the Medprompt paper, "Can Generalist Foundation Models Outcompete Special-Purpose Tuning? Case Study in Medicine," which reported that general-purpose prompting strategies let GPT-4 "easily" top prior leading results on medical benchmarks, introducing the Medprompt technique.[21]

In June 2025 he was the last author of "Sequential Diagnosis with Language Models," which turned 304 New England Journal of Medicine clinicopathological conference cases into stepwise diagnostic encounters and introduced the MAI Diagnostic Orchestrator (MAI-DxO). The authors reported that MAI-DxO paired with OpenAI's o3 reached 80% diagnostic accuracy against a 20% average for the generalist physicians in the study, and 85.5% when configured for maximum accuracy.[22] He is also a co-author of a 25-author study of a large language model on physician reasoning tasks, whose preprint was titled "Superhuman performance of a large language model on the reasoning tasks of a physician" and which appeared in Science on April 30, 2026 as "Performance of a large language model on the reasoning tasks of a physician."[23] Other recent co-authored work includes JSONSchemaBench (2025), a benchmark for structured outputs in language models, and MEMENTO (April 2026), a method that teaches reasoning models to compress their own intermediate reasoning into summaries to reduce context and KV-cache use.[58][59]

Biosecurity

Horvitz is the last author, with Bruce Wittmann as first author, of "Strengthening nucleic acid biosecurity screening against generative protein design tools," a 12-author study published in Science on October 2, 2025.[24] In a Microsoft Research blog post he wrote that computer-based studies had found that AI protein design tools "could generate modified versions of proteins of concern, such as ricin," and that the reformulated proteins evaded the screening systems used by DNA synthesis companies.[25] The team treated the finding like a cybersecurity "zero-day," worked confidentially with synthesis companies, biosecurity organizations and policymakers to distribute "patches" before publication, and released sensitive data and code through a tiered-access system run with the International Biosecurity and Biosafety Initiative for Science (IBBIS).[25] In a June 4, 2026 post on Microsoft's policy blog he called nucleic acid synthesis screening "one of the most effective near-term defenses against biological misuse" and wrote that Microsoft "strongly supports" the Biosecurity Modernization and Innovation Act (S. 3741) introduced by Senators Cotton and Klobuchar.[26]

AI governance and policy work

AAAI presidency and the Asilomar study

Horvitz was president of AAAI from 2007 to 2009, between Alan Mackworth and Martha Pollack.[27] He has written that he made "Artificial Intelligence in the Open World" the theme of his presidency, that his July 2008 presidential lecture called for more robust systems that "understand their own limitations," and that he worked with Mackworth to change the society's name from the American Association for Artificial Intelligence to the Association for the Advancement of Artificial Intelligence while keeping the acronym.[18] As president he commissioned the AAAI Presidential Panel on Long-Term AI Futures, which he co-chaired with Bart Selman; its members met at Asilomar in Pacific Grove, California, on February 21-22, 2009, and its remit included "concerns about the potential for loss of human control of computer-based intelligences."[28]

One Hundred Year Study on AI

In a 2014 framing memo, Horvitz wrote that he approached Stanford in early 2014 with a proposal for a one-hundred-year program on AI and its influence on society, and that the proposal was accepted the following June.[29] Stanford announced the One Hundred Year Study on Artificial Intelligence (AI100) on December 16, 2014, calling it "the brainchild" of Horvitz, who "initiated AI100 as a private philanthropic initiative"; Russ Altman became faculty director and the first standing committee included Barbara Grosz, Alan Mackworth, Tom Mitchell, Deirdre Mulligan and Yoav Shoham.[7] Grosz chaired the standing committee that issued the first AI100 report in 2016; Horvitz served on the committee from 2015 to 2019.[30][31] The 2016 report's preface says AI100 was modeled on the 2008-2009 AAAI Asilomar study.[30] The AI Index began as a project of AI100, and the Alliance for Decision Education's 2026 announcement describes Horvitz as having co-founded it.[4][55]

Aether and responsible AI at Microsoft

According to his own account, Horvitz and Microsoft president Brad Smith co-founded the Aether committee ("AI and ethics in engineering and research") in 2016, and Horvitz has chaired it since.[32] Aether's work included developing and refining Microsoft's AI principles and setting up the company's Sensitive Uses review program, and in 2019 it helped form the Office of Responsible AI, which took over the Sensitive Uses program and wrote Microsoft's Responsible AI Standard.[32] Aether also ran cross-company "Laser studies" of emerging technologies, including studies of pre-release versions of GitHub Copilot and GPT-4.[32]

A 2019 Aether project on media provenance led to Project Origin, a collaboration of Microsoft with the BBC, CBC and The New York Times, and later to the Coalition for Content Provenance and Authenticity (C2PA) and the Content Credentials standard.[32] Horvitz has told the story as beginning in January 2019 at the World Economic Forum, where he showed the head of a major news organization deepfake videos of political leaders.[1][40]

Partnership on AI

When Amazon, DeepMind/Google, Facebook, IBM and Microsoft announced the Partnership on Artificial Intelligence to Benefit People and Society on September 28, 2016, Horvitz was pictured at the New York announcement as Microsoft's founding representative, alongside Francesca Rossi of IBM, Yann LeCun of Facebook and Mustafa Suleyman of DeepMind.[33] He served as interim co-chair and later as board chair.[34][1]

Congress, NSCAI and PCAST

  • First congressional hearing on AI (2016). On November 30, 2016, he testified as interim co-chair of the Partnership on AI at "The Dawn of Artificial Intelligence," a hearing of the Senate Commerce Subcommittee on Space, Science and Competitiveness chaired by Ted Cruz, which Cruz's office called "Congress' first ever hearing focused solely on artificial intelligence."[34]
  • National Security Commission on AI. He was a congressionally appointed commissioner on the National Security Commission on Artificial Intelligence and chaired its line of effort on ethical and trustworthy AI.[1] He was first author of the commission's "Key Considerations for the Responsible Development and Fielding of Artificial Intelligence."[35]
  • PCAST. President Biden named him to PCAST on September 22, 2021, as one of 27 members announced that day; the American Institute of Physics listed him among the council's information technology industry representatives.[36][37] The PCAST reports on his website run from 2022 to January 15, 2025, and include the April 2024 report "Supercharging Research: Harnessing Artificial Intelligence to Meet Global Challenges" and a February 2024 report on cyber-physical resilience.[38]
  • Senate Armed Services (2022). On May 3, 2022, he testified before the Senate Armed Services Subcommittee on Cybersecurity on the use of AI in Department of Defense cyber missions.[39]

His other service listed by Microsoft and the National Academy of Medicine includes the advisory boards of the Allen Institute for AI and Stanford HAI, the National Academies' Computer Science and Telecommunications Board, the Computing Community Consortium, the Board of Regents of the National Library of Medicine, and advisory committees for the NSF, NIH and DARPA.[1][54] He is a member of the Board of Sponsors of the Bulletin of the Atomic Scientists, and in February 2026 he joined the board of directors of the Alliance for Decision Education, a nonprofit that promotes decision-making skills in K-12 schools.[5][55]

Views on AI risk

In an April 2023 Fortune interview, published weeks after the open letter calling for a six-month pause on training powerful AI systems, he said he respected the signers but added, "I would prefer to see more knowledge, and even an acceleration of research and development, rather than a pause for six months, which I am not sure if it would even be feasible."[40] He said: "I'm least concerned with science-fiction-centric notions that scare people of A.I. taking over," and "I'm perhaps most concerned about the use of these tools for disinformation, manipulation, and impersonation."[40] He also said that "we're also probably going to need regulations."[40] In July 2023 he was one of 24 authors of "Frontier AI Regulation: Managing Emerging Risks to Public Safety," which proposed standard-setting, registration and reporting requirements, and compliance mechanisms for developers of highly capable foundation models.[41] In 2024 he co-authored the PNAS statement "Protecting scientific integrity in an age of generative AI."[42]

On June 4, 2026, Science published an editorial by Horvitz and Robert West, "A narrowing window to understand AI."[43] TechRadar quoted them as writing that "AI systems are now designed and refined by AI systems through recursive cycles that can outpace human understanding," and reported that they described an "interactional opacity" as AI systems increasingly communicate with each other, and a growing asymmetry in which long-running AI agents model people's preferences and "latent drivers such as fear, uncertainty, and the need for social belonging" while human understanding of the systems falls behind.[44] According to TechRadar, the authors argued that even partial understanding of AI is useful if it helps catch risks early.[44]

The intelligence explosion paper (2026)

On September 28, 2026, Horvitz appeared sixth in the 22-author list of "What if automating AI R&D triggers an intelligence explosion?", No. 2/2026 in the Frontier AI Working Paper Series of the Cambridge Programme on AI Science & Policy (CASP), also published by GovAI.[45][47] The paper lists his affiliation simply as Microsoft.[45] The co-authors include Geoffrey Hinton, Yoshua Bengio, Andrew Barto, OpenAI chief scientist Jakub Pachocki, Anthropic co-founder Jack Clark and Dawn Song; the corresponding authors are Alan Chan of GovAI and Sören Mindermann of CASP.[45] The accompanying executive summary names Horvitz, as Microsoft's Chief Scientific Officer, among the "industry leaders" on the author list.[46]

The paper argues that AI systems "are on track to automate most AI R&D work within a few years, and possibly all of it," and that if this triggers an intelligence explosion, "capabilities growth could accelerate far beyond what society can keep up with, humanity could lose control over superhuman AI systems," and checks on power could be "severely eroded."[45] It calls on policymakers to obtain visibility into AI R&D automation, develop ways to steer and constrain an intelligence explosion, and prepare society for its impacts.[45] The paper states that its views "are the authors' and do not necessarily represent the views of the organizations with which they are affiliated."[45] The Next Web noted that the authors "wrote in a personal capacity."[48] See Recursive self-improvement for the history of the intelligence explosion idea.

Honors and memberships

Honor or membershipYearCitation or noteSource
American Academy of Arts and SciencesElected 2011Computer sciences[50]
National Academy of EngineeringElected 2013"For computational mechanisms for decision making under uncertainty and with bounded resources"[49]
ACM Fellow2014Listed among 2014 ACM Fellows[53]
AAAI Feigenbaum Prize2015"For sustained and high-impact contributions to the field of artificial intelligence through the development of computational models of perception, reflection and action"[52]
ACM-AAAI Allen Newell Award2015"For contributions to artificial intelligence and human-computer interaction spanning the computing and decision sciences through developing principles and models of sensing, reflection, and rational action"[51]
CHI AcademyNot statedInducted for work at the intersection of AI and human-computer interaction[1][54]
American Philosophical SocietyNot statedElected member[1][37]
Fellow of AAAI, AAAS and the American College of Medical InformaticsNot statedListed in official biography[1]

Expanded article table

The University of Washington, where he was an affiliate professor, reported the 2015 Newell Award in April 2016.[60] In 2026 the Journal of the American Medical Informatics Association published his article "AI, humanity, and the open world of health care: enduring imperatives for the next century," which his homepage lists as a "Second Century award and address."[57][4] In May 2026 Columbia University School of Nursing named him its commencement speaker.[56]

References

  1. ^1 ^2 ^3 ^4 ^5 ^6 ^7 ^8 ^9 ^10 ^11 ^12 ^13 ^14 ^15 ^16 ^17Microsoft Research. "Eric Horvitz, Chief Scientific Officer" (profile). Accessed September 30, 2026. microsoft.com/...horvitz
  2. ^1 ^2 ^3 ^4 ^5 ^6 ^7 ^8 ^9 ^10 ^11 ^12Soper, Taylor. "Microsoft taps Eric Horvitz as first chief scientific officer; Peter Lee takes over Microsoft Research." GeekWire, March 11, 2020. geekwire.com/...peter-lee-takes-microsoft-research
  3. ^1 ^2 ^3 ^4 ^5 ^6Novet, Jordan. "Microsoft unifies research groups as it appoints a science chief." CNBC, March 11, 2020. cnbc.com/...-groups-as-it-appoints-a-science-chief
  4. ^1 ^2Horvitz, Eric. "Home page of Eric Horvitz." Accessed September 30, 2026. erichorvitz.com
  5. ^1 ^2 ^3Bulletin of the Atomic Scientists. "Eric Horvitz" (Board of Sponsors biography). Accessed September 30, 2026. thebulletin.org/...eric-horvitz
  6. ^1 ^2 ^3Wang, Hongning. "Microsoft Chief Science Officer Dr. Eric Horvitz discusses the promises and perils of AI." The Tech (MIT), November 14, 2025. thetech.com/...horvitz-csail-talk
  7. ^1 ^2Stanford Engineering. "Stanford to host 100-year study on artificial intelligence." December 16, 2014. engineering.stanford.edu/...rtificial-intelligence
  8. ^1 ^2Microsoft Research. "The Meteoric Rise of Microsoft Research: An Oral History." September 26, 2006. microsoft.com/...e-microsoft-research-oral-history
  9. ^1 ^2 ^3Microsoft Research. "Eric Horvitz Receives AAAI Feigenbaum Prize; Shares Reflections On AI Research." January 27, 2015. microsoft.com/...shares-reflections-on-ai-research
  10. ^Horvitz, Eric J. "Reasoning About Beliefs and Actions Under Computational Resource Constraints." Proceedings of the Third Conference on Uncertainty in Artificial Intelligence (UAI 1987). arXiv:1304.2759. arxiv.org/...1304.2759
  11. ^Horvitz, Eric J.; Barry, Matthew. "Display of Information for Time-Critical Decision Making." Proceedings of the Eleventh Conference on Uncertainty in Artificial Intelligence (UAI 1995). arXiv:1302.4959. arxiv.org/...1302.4959
  12. ^1 ^2Horvitz, Eric J.; Breese, John S.; Heckerman, David; Hovel, David; Rommelse, Koos. "The Lumiere Project: Bayesian User Modeling for Inferring the Goals and Needs of Software Users." Proceedings of the Fourteenth Conference on Uncertainty in Artificial Intelligence (UAI 1998). arXiv:1301.7385. arxiv.org/...1301.7385
  13. ^Horvitz, Eric. "Principles of mixed-initiative user interfaces." Proceedings of CHI '99, pp. 159-166. doi.org/...302979.303030 (PDF: erichorvitz.com/chi99horvitz.pdf)
  14. ^Horvitz, Eric J.; Jacobs, Andy; Hovel, David. "Attention-Sensitive Alerting." Proceedings of the Fifteenth Conference on Uncertainty in Artificial Intelligence (UAI 1999). arXiv:1301.6707. arxiv.org/...1301.6707
  15. ^Sahami, Mehran; Dumais, Susan; Heckerman, David; Horvitz, Eric. "A Bayesian Approach to Filtering Junk E-Mail." AAAI Workshop Papers 1998. aaai.org/...055-ws98-05-009
  16. ^Horvitz, Eric, et al. "Prediction, Expectation, and Surprise: Methods, Designs, and Study of a Deployed Traffic Forecasting Service." Proceedings of the Twenty-First Conference on Uncertainty in Artificial Intelligence (UAI 2005). arXiv:1207.1352. arxiv.org/...1207.1352
  17. ^Gershman, Samuel J.; Horvitz, Eric J.; Tenenbaum, Joshua B. "Computational rationality: A converging paradigm for intelligence in brains, minds, and machines." Science 349 (6245): 273-278, July 17, 2015. doi.org/...science.aac6076
  18. ^1 ^2 ^3Horvitz, Eric. "Reflections as AAAI 2026 Concludes." LinkedIn, January 28, 2026. linkedin.com/...-2026-concludes-eric-horvitz-1pivc
  19. ^Bubeck, Sébastien, et al. "Sparks of Artificial General Intelligence: Early experiments with GPT-4." arXiv:2303.12712, March 22, 2023. arxiv.org/...2303.12712
  20. ^Nori, Harsha; King, Nicholas; McKinney, Scott Mayer; Carignan, Dean; Horvitz, Eric. "Capabilities of GPT-4 on Medical Challenge Problems." arXiv:2303.13375, March 20, 2023. arxiv.org/...2303.13375
  21. ^Nori, Harsha, et al. "Can Generalist Foundation Models Outcompete Special-Purpose Tuning? Case Study in Medicine." arXiv:2311.16452, November 28, 2023. arxiv.org/...2311.16452
  22. ^Nori, Harsha, et al. "Sequential Diagnosis with Language Models." arXiv:2506.22405, June 27, 2025. arxiv.org/...2506.22405
  23. ^Brodeur, Peter G., et al. "Performance of a large language model on the reasoning tasks of a physician." Science 392 (6797): 524-527, April 30, 2026. doi.org/...science.adz4433 (preprint: "Superhuman performance of a large language model on the reasoning tasks of a physician," arXiv:2412.10849, arxiv.org/...2412.10849)
  24. ^Wittmann, Bruce J., et al. "Strengthening nucleic acid biosecurity screening against generative protein design tools." Science 390 (6768): 82-87, October 2, 2025. doi.org/...science.adu8578
  25. ^1 ^2Horvitz, Eric. "When AI Meets Biology: Promise, Risk, and Responsibility." Microsoft Research Blog, October 6, 2025. microsoft.com/...y-promise-risk-and-responsibility
  26. ^Horvitz, Eric. "Strengthening biosecurity in the era of AI." Microsoft On the Issues, June 4, 2026. blogs.microsoft.com/...iosecurity-in-the-era-of-ai
  27. ^1 ^2AAAI. "Past AAAI Officers." Accessed September 30, 2026. aaai.org/...past-aaai-officers
  28. ^AAAI. "AAAI Presidential Panel on Long-Term AI Futures: 2008-2009." Accessed September 30, 2026. aaai.org/...anel-on-long-term-ai-futures-2008-2009
  29. ^Horvitz, Eric. "One Hundred Year Study on Artificial Intelligence: Reflections and Framing." 2014. ai100.stanford.edu/...reflections-and-framing
  30. ^1 ^2One Hundred Year Study on Artificial Intelligence. "Preface" (2016 report). ai100.stanford.edu/...preface
  31. ^One Hundred Year Study on Artificial Intelligence. "Standing Committee History." Accessed September 30, 2026. ai100.stanford.edu/standing-committee-history
  32. ^1 ^2 ^3 ^4 ^5Horvitz, Eric. "Aether Committee at Microsoft." Accessed September 30, 2026. erichorvitz.com/Aether_Committee_Microsoft
  33. ^DeepMind. "Announcing the Partnership on AI to Benefit People & Society." September 28, 2016. deepmind.google/...on-ai-to-benefit-people-society
  34. ^1 ^2 ^3 ^4Office of Senator Ted Cruz. "Sen. Cruz Chairs First Congressional Hearing on Artificial Intelligence." November 30, 2016. cruz.senate.gov/...ring-on-artificial-intelligence
  35. ^Horvitz, Eric; Young, Jessica; Elluru, Rama G.; Howell, Chuck. "Key Considerations for the Responsible Development and Fielding of Artificial Intelligence." Report of the National Security Commission on AI. arXiv:2108.12289. arxiv.org/...2108.12289
  36. ^1 ^2American Institute of Physics, FYI. "Biden Names President's Council of Advisors on Science and Technology." 2021. aip.org/...council-advisors-science-and-technology
  37. ^1 ^2The White House (archived). "Eric Horvitz" (PCAST member page). bidenwhitehouse.archives.gov/...eric-horvitz
  38. ^Horvitz, Eric. "PCAST Documents and Reports from the Years 2020 to 2025." Accessed September 30, 2026. erichorvitz.com/PCAST_reports_2020-2025
  39. ^Microsoft On the Issues. "Applications for artificial intelligence in Department of Defense cyber missions." May 3, 2022. blogs.microsoft.com/...t-of-defense-cyber-missions
  40. ^1 ^2 ^3 ^4Mathews, Jessica. "Microsoft's chief scientific officer, one of the world's leading A.I. experts, doesn't think a 6 month pause will fix A.I., but has some ideas of how to safeguard it." Fortune (via Yahoo Finance), April 30, 2023. finance.yahoo.com/...entific-officer-one-230000103
  41. ^Anderljung, Markus, et al. "Frontier AI Regulation: Managing Emerging Risks to Public Safety." arXiv:2307.03718, July 6, 2023. arxiv.org/...2307.03718
  42. ^Blau, Wolfgang, et al. "Protecting scientific integrity in an age of generative AI." PNAS 121 (22), May 21, 2024. doi.org/...pnas.2407886121
  43. ^Horvitz, Eric; West, Robert. "A narrowing window to understand AI." Science 392 (6802): 1003, June 4, 2026. doi.org/...science.aei3167
  44. ^1 ^2Udinmwen, Efosa. "Microsoft CSO acknowledges that humans are struggling to keep up with AI advancement, reckons we've got a 'narrowing window to understand AI' before it's, well, too late." TechRadar, June 20, 2026. techradar.com/...stand-ai-before-its-well-too-late
  45. ^1 ^2 ^3 ^4 ^5 ^6 ^7Chan, Alan; Winter, Christoph; Barto, Andrew; Pachocki, Jakub; Hinton, Geoffrey; Horvitz, Eric; et al. "What if automating AI R&D triggers an intelligence explosion?" Frontier AI Working Paper Series No. 2/2026, Cambridge Programme on AI Science & Policy, September 2026. casp.ac/...intelligence-explosion.pdf (also arXiv:2609.36054, arxiv.org/...2609.36054)
  46. ^Cambridge Programme on AI Science & Policy. "What if automating AI R&D triggers an intelligence explosion? Executive Summary." September 2026. casp.ac/...intelligence-explosion-summary.pdf
  47. ^GovAI. "What If Automating AI R&D Triggers an Intelligence Explosion?" September 28, 2026. governance.ai/...riggers-an-intelligence-explosion
  48. ^Stan, Alina Maria. "Hinton, Bengio and AI lab scientists warn of an intelligence explosion." The Next Web, September 28, 2026. thenextweb.com/...per-hinton-bengio-pachocki-clark
  49. ^1 ^2 ^3National Academy of Engineering. "Dr. Eric Horvitz" (member page). Accessed September 30, 2026. nae.edu/...Dr-Eric-Horvitz
  50. ^1 ^2American Academy of Arts and Sciences. "Eric Joel Horvitz" (member page). Accessed September 30, 2026. amacad.org/...eric-joel-horvitz
  51. ^1 ^2 ^3 ^4ACM. "2015 ACM Technical Awards." awards.acm.org/...2015-technical-awards
  52. ^1 ^2AAAI. "AAAI Feigenbaum Prize." Accessed September 30, 2026. aaai.org/...aaai-feigenbaum-prize
  53. ^1 ^2ACM SIGAI. "ACM Fellows." Accessed September 30, 2026. sigai.acm.org/...acm_fellows
  54. ^1 ^2National Academy of Medicine. "Eric Horvitz, MD, PhD." Accessed September 30, 2026. nam.edu/...eric-horvitz-md-phd
  55. ^1 ^2Alliance for Decision Education. "Alliance for Decision Education Appoints Microsoft Chief Scientific Officer Eric Horvitz to Board of Directors." GlobeNewswire, February 18, 2026. globenewswire.com/...Horvitz-to-Board-of-Directors
  56. ^Columbia University School of Nursing. "AI Researcher Eric Horvitz is Commencement Speaker." May 12, 2026. nursing.columbia.edu/...rvitz-commencement-speaker
  57. ^Horvitz, Eric. "AI, humanity, and the open world of health care: enduring imperatives for the next century." Journal of the American Medical Informatics Association 33 (8): 1567-1568, June 17, 2026. doi.org/...ocag099
  58. ^Geng, Saibo, et al. "JSONSchemaBench: A Rigorous Benchmark of Structured Outputs for Language Models." arXiv:2501.10868, January 18, 2025. arxiv.org/...2501.10868
  59. ^Kontonis, Vasilis, et al. "MEMENTO: Teaching LLMs to Manage Their Own Context." arXiv:2604.09852, April 10, 2026. arxiv.org/...2604.09852
  60. ^Paul G. Allen School of Computer Science & Engineering. "UW CSE affiliate professor Eric Horvitz of Microsoft Research receives ACM-AAAI Allen Newell Award." April 27, 2016. news.cs.washington.edu/...-aaai-allen-newell-award
  61. ^1 ^2Microsoft Research. "Research Activities: Eric Horvitz, Chief Scientific Officer." Accessed September 30, 2026. microsoft.com/...research-activities

Improve this article

Add missing citations, update stale details, or suggest a clearer explanation. Every suggestion is reviewed for sourcing before it goes live.

1 revision · v2 · 4,906 words · full history

Fact-checks are independent of edits: a reviewer re-verifies the article against its sources and stamps the date. How we verify

Research and drafting on this wiki are AI-assisted, under named human editorial standards. How AI is used here

Reviewer note: Independent verification V4 (xg14, 30 Sep 2026): ~110 claims vs Microsoft profile, GeekWire, CNBC, AIP, AI100, UAI paper, honors pages, arXiv; 0 material, 5 minor fixed; privacy rules checked

Cite this page: AI Wiki. "Eric Horvitz." aiwiki.ai, updated 30 Sept 2026, fact-checked 30 Sept 2026. CC BY 4.0. https://aiwiki.ai/wiki/eric_horvitz

Suggest edit