@erichorvitz
Chief Scientific Officer, Microsoft
Collaboration gap! #COLM2026 @im_td @adamfourney @SaleemaAmershi @cervisiarius @ecekamar @EPFL @MSFTResearch
“How to do something useful” @JohnCLangford shares broader context for today’s ML research—including “change stress” and “review stress” in his invited talk at #COLM2026 @MSFTResearch @Microsoft
Seems like yesterday. We can achieve these visions now with the core capabilities of today’s AI models, but need to innovate with designs for human-AI collaboration and safety, reliability, privacy, and transparency. @MSFTResearch @MicrosoftAI
Gaurav Devsarmah@g_devsarmah·Incredible that this was laid out in 2001 by @erichorvitz , the tech is finally scalable now and personal agents are running hot but we still haven't nailed the right human-AI interface youtube.com/watch?v=1KYxJM…
Human cognition in the AI era. Podcast: podcasts.apple.com/us/podcast/the…
I enjoyed this wide-ranging discussion with Annie Duke, and hope you'll find it interesting. We recorded it several months ago, before the recent surge of concern about AI. The ideas and themes we discussed feel more important than ever. @AnnieDuke @AllDecisionEd @Microsoft @decisionpros @kevin_scott @AnthropicAI @openai
Alliance for Decision Education@AllDecisionEd·What do we lose when a machine can think for us? On the latest episode of The Decision Education Podcast, @erichorvitz, Chief Scientific Officer at @Microsoft, joins @AnnieDuke to explore what it really means for humans to stay "in the driver's seat" as headlines about an AI takeover proliferate the news. Drawing on decades of research into human-AI collaboration, Eric argues against the passive framing of an AI "takeover," making the case that governance, design, and clear values determine whether these tools sharpen human judgment or quietly erode it. Plus, hear Eric draw on lectures he’s given for the Society of Decision Professionals, arguing that the best AI decision tools help people clarify what they value. Listen now! t.co/j164LBjQRB
A pleasure to share reflections on directions ahead for human-AI collaboration at #HCOMP2026 + #CI2026 Great energy at the meeting. Nice to catch up with long-term colleagues and to meet new folks.
Kenneth Huang #HCOMP2026 Sep27-30@DC@windx0303·Dr. Eric Horvitz (@erichorvitz) is delivering the opening keynote at #HCOMP2026 #CI2026!
Such a pleasure to spend time with Judea @yudapearl and colleagues on his 90th birthday. What an inspiring day to look back—and also forward—on the influences and impact of Judea’s contributions.
Elias Bareinboim@eliasbareinboim·On September 6 we celebrated Judea Pearl's 90th birthday at UCLA, and the morning session is now online. It was a day to say thank you to Judea, and an honor to organize it with Rina Dechter and Hector Geffner. Students, collaborators, and friends came from across the world, and the affection in the room was the real subject of the day. The morning had a keynote by @tdietterich, talks by Chris Hitchcock, @DaphneKoller, and me, a panel with @DanRothNLP, @erichorvitz , and @pmddomingos, remarks by Chancellor @UCLAchancellor, and personal reflections from students, collaborators, and friends. t.co/tqL6UbNYaU Thank you to Zvi Galil, Leonard Kleinrock, Deborah Estrin, Moises Goldszmidt, Kaoru Mulvihill, Ayala Or-El, and Jason Cong, who spoke in the room, and to Moshe Vardi, Stuart Russell, Uri Sivan, David Harel, vint cerf, Warshel Arieh, Bernhard Schölkopf, Steven Sloman, guido imbens, and Dan Geiger, who sent messages, and to everyone who came. The afternoon session will follow.
Informative conversation. Great job, @polynoamial. @OpenAI
Dwarkesh Patel@dwarkesh_sp·New episode with @polynoamial We talk about multi-agent, Navier-Stokes, and what the current explosion of maths progress tells us about what happens once you automate AI research. And we also discuss how we will know if the models are actually aligned before we kick off RSI. 0:00:00 – Multi-agent and Navier-Stokes 0:15:28 – How will AI firms work? 0:22:02 – What math progress tells us about recursive self improvement 0:40:22 – Hugging Face and alignment 1:01:18 – The internal/external model gap 1:08:34 – Chain of thought is degrading 1:14:12 – How will we know when alignment is solved?
"Any pursuit of superintelligence has to be grounded in the core principle that if the AI we build is not helping humanity and under human control, it's not worth pursuing."
Satya Nadella@satyanadella·Any pursuit of superintelligence has to be grounded in the core principle that if the AI we build is not helping humanity and under human control, it's not worth pursuing. We also need to accelerate and spread the benefits of AI, such that they are diffused broadly across countries, communities, and companies. This requires a frontier ecosystem in which both closed and open-source models can thrive. And for firms, it’s imperative that they retain full control over their unique and tacit knowledge. Every organization should be able to build its own continuous learning loop/hill climbing machine, without becoming dependent on any one model provider, and have the ability to embed its own knowledge into models and weights they control. So, in this context, we welcome the research, focus, and deliberate pacing needed to get alignment right as the design goal. We also welcome ideas like "embedded evaluators" and the broader efforts to develop the mechanisms to make this more than just talk. The key is that this cannot be controlled by a handful of entities, but must have broad representation across the ecosystem, countries, and fields, including academia. This is the approach we are taking: broad access and choice at every layer of the AI stack; enterprise control of learning loops and models; and the “Code of Conduct” that underlies our own first party MAI models that we’ll publish tomorrow for public consultation.
Blast to the past. Coverage of the AAAI Asilomar meeting: nytimes.com/2009/07/26/sci…
A narrowing window to understand & guide AI: science.org/doi/10.1126/sc…
Thoughtful essay about steps forward for AI safety by @DarioAmodei.
Dario Amodei@DarioAmodei·We Must Pace the Frontier: I’ve written a new essay on why the AI industry should slow down, with a three-part plan for doing so. Anthropic is unilaterally committing to the first of these steps. We’ll provide third-party evaluators with permanent, employee-level access to our systems, so that they can verify adherence to our safety measures, report on incidents, and assess models’ alignment during training. You can read the full post here: t.co/OGyPb7yaYt
Learn about the Paraphrase project and efforts by @Microsoft, @OpenAI, @Anthropic and other organizations to address biosecurity concerns with the rising powers of AI. podcasts.apple.com/us/podcast/hea…
More on directions with bolstering biosecurity: blogs.microsoft.com/on-the-issues/…
Calling out concerns with trends on understanding and guiding AI.
Bob West@cervisiarius·6/ This is why @erichorvitz and I warned in @ScienceMagazine about "A narrowing window to understand AI". Human understanding—even if only partial and high-level—is crucial, and it won’t emerge on its own; it must be an explicit training objective. science.org/doi/10.1126/sc…
Tandem training: An approach aimed at incentivizing models to be understandable. Considered "understanding" by weaker models--but had human interpretability of machine intelligence in mind.
Bob West@cervisiarius·7/ One proposal: tandem training (@ashton_anderson @erichorvitz), where a strong "senior" (Astra?) co-generates solutions with a frozen weaker "junior" (~humans). To get reward, the senior's reasoning must be something the junior can pick up and continue. aclanthology.org/2026.eacl-long…
Astra's model card shows that monitoring for abuse via CoT may become increasingly difficult. Beyond efforts on alignment & refusals, impt. to invest in maintaining monitorability. R&D on efforts like "tandem training" will be important. More from @cervisiarius below. @sama @DarioAmodei
Bob West@cervisiarius·1/ The state-of-the-art way of monitoring state-of-the-art models is to read tea leaves. Left: a model reasoning to itself. Right: the same model talking to other agents. Every frontier lab depends on this, @sama @DarioAmodei @demishassabis
Here's a demo of Lumiere, the research prototype that inspired Clippy (the Office Assistant): t.co/tHWwZcBhOB As background: The shipped version operated quite differently. The Office team decided to execute the Bayesian inference component only when there was a user-initiated search. So there was no autonomous initiative for the Bayesian component. So where did the infamous Clippy behaviors come from? The activations that folks noted and joked about were based in a separate, legacy system that was hooked up to Clippy--relatively simplistic rule-based tips that were a holdover from the "Tip Wizard" system that had been used in earlier versions of Office, where the rules gently lit up a lightbulb icon on a toolbar. So, there were actually two minds behind Clippy--one Bayesian and only executing when a question was asked, and the other a legacy rule-based system designed originally for a quieter bell icon.
Holostaff AI@holostaffai·The model under Clippy watched for a pause after activity, a hunt through several menus, an undo. Only then did it score the odds you wanted help, against a threshold you set yourself. Eric Horvitz published the whole thing in 1998. Paper in the reply.
Here's the demo of Lumiere, on the core technology that led the Office Assistant (aka "Clippy"): t.co/tHWwZcBhOB As background: The Microsoft Research prototype inspired the product, but the shipping product was quite different because of challenges with scaling the service. A lightweight variant of the Bayesian inference was used to predict the best assistance to provide based on streams of events and words. And a design choice was made by the Office team: the Bayesian inference component would only come to life when there was a user-initiated search. So there was no autonomous initiative for the Bayesian component. The product team didn't compute the probability that a user needed help, the way we did in the prototype. So where did the infamous Clippy behaviors come from? All of the activation that folks noted and joked about was based in rules firing in a separate, legacy system: the relatively simplistic tips that were a holdover from the "Tip Wizard" system that had been used in earlier versions of Office--where the rules gently lit up a lightbulb icon on a toolbar. A decision was made when building the shipping version of Clippy to connect those legacy rules to Clippy (and friends). So, there were actually two minds behind Clippy--one Bayesian and only executing when a question was asked, and the other a legacy rule-powered system designed for a quieter bell icon.
Holostaff AI@holostaffai·The model under Clippy watched for a pause after activity, a hunt through several menus, an undo. Only then did it score the odds you wanted help, against a threshold you set yourself. Eric Horvitz published the whole thing in 1998. Paper in the reply.
Nature Digital podcast on clinician-AI collaboration. open.spotify.com/episode/3mta95…
That graphic makes me feel like a member of the Mod Squad. Looking forward to sharing my conversation with @AnnieDuke on the Decision Education podcast. Skills for decision making under uncertainty will only grow with the increasing influence of AI in our lives. I'm deeply supportive of the goals & efforts of the Alliance for Decision Education. @AllDecisionEd
Alliance for Decision Education@AllDecisionEd·We make thousands of decisions every day. How many do we actually think about? A new season of The Decision Education Podcast is coming to help you learn about the science behind decision-making and how the choices we make today can transform our future. Throughout the season, host @AnnieDuke will talk to experts including @erichorvitz, Jenny Just, and Corinne Low, and more, exploring what better decision-making looks like in life, work, and society. Subscribe now to never miss an episode, and check out our Season 7 trailer: t.co/tlCCibBFYd Plus, stay tuned for Navigating Uncertainty, a limited series we’re launching this fall hosted by our Executive Director David Samuelson! New episodes will drop in The Decision Education Podcast feed.
RadFusion: Confidence-based control of clinical report generation t.co/gOEClCQFn6 AI-generated radiology report generation has focused on creating a single, best report. But clinical settings have different costs for misses vs. false alarms. Thus, we may want different “operating points,” that is, different thresholds for when findings should be reported.
RadFusion makes that tradeoff explicit & controllable. It combines the strengths of classification + generation: calibrated classifier scores provide threshold control, a radiology report generator provides clinical detail, and an LLM reconciles the two. Change the thresholds for reporting findings → change the sensitivity/specificity tradeoff → generate a report consistent with the selected operating points. Details: t.co/gOEClCQFn6 Collaboration with Ying Jin, @NoelCodella, John Corring, Mu Wei, Dinei Florencio @Microsoft
Just recorded an exciting session with host @AnnieDuke. Looking forward to sharing the conversation.
Alliance for Decision Education@AllDecisionEd·We make thousands of decisions every day. How many do we actually think about? A new season of The Decision Education Podcast is coming to help you learn about the science behind decision-making and how the choices we make today can transform our future. Throughout the season, host @AnnieDuke will talk to experts including @erichorvitz, Jenny Just, and Corinne Low, and more, exploring what better decision-making looks like in life, work, and society. Subscribe now to never miss an episode, and check out our Season 7 trailer: t.co/tlCCibBFYd Plus, stay tuned for Navigating Uncertainty, a limited series we’re launching this fall hosted by our Executive Director David Samuelson! New episodes will drop in The Decision Education Podcast feed.
High-stakes decisions with language models We introduce 'decision-analytic steering' in this study: t.co/sRpbqujSIX In real-world decisions, different types of errors rarely have equal consequences. So, a system making recommendations under uncertainty must do more than predict what is likely to be true. It must also reflect the relative costs and benefits of acting or failing to act. As an example, in emergency medical triage, missing a patient who needs immediate care is often far more costly than sending someone to the emergency department who ultimately did not need to go. It's critically important to consider that asymmetry. We set out to investigate opportunities for decision-analytic steering by generating ROC curves for families of model. These sample curves showing the true positive and false positive rate for two models on a set of emergency medicine triage cases.
A recent Nature Medicine study provided a striking example: ChatGPT Health was found (Feb, 2026, @NatureMedicine) to have under-triaged more than half of the cases that physicians judged to require immediate emergency care. Was the underlying model simply unable to recognize the emergencies? Our analyses suggest a more nuanced and actionable answer.
Exciting results for AI in mathematics and theoretical computer science @OpenAI openai.com/index/ten-adva…
Nice summary. msn.com/en-us/technolo…
Full editorial: science.org/doi/10.1126/sc…
Congratulations, Wajeeha. Enjoyed our collaborations. Onward! @wajeeha__ahmad x.com/wajeeha__ahmad…
New study on AI and biosecurity: We tested whether updated screening tools can detect short fragments of AI-engineered sequences. The results are encouraging, but also underscore the need for sustained red-teaming, research, and safeguards as AI-biology capabilities advance. Article: t.co/pak563SIkE @IBBIS_bio @Microsoft
Podcast on clinician-AI collaboration study: cassyni.com/events/Cfw9Yuc…
A few years back. @MSFTResearch x.com/hbou/status/20…
@NoamShazeer Noam central!
Test of time awards are the most impressive honors for publications. Longuet-Higgins is the 10yr impact award for CVPR papers. Congrats to the ResNet team. Seems like yesterday. @MSFTResearch x.com/msftresearch/s…
Important finding. x.com/polynoamial/st…
In diagnostic settings, it's important to understand how AI models weigh uncertainties and tradeoffs--for example, the cost of false alarms vs. cost of missed diagnoses. @KhurramYam describes methods for identifying implicit utilities, and for testing whether models can follow x.com/KhurramYam/sta…
Editorial: “A narrowing window to understand AI,” published in this week’s issue of Science science.org/doi/10.1126/sc… @cervisiarius @EPFL @MSFTResearch
Editorial: “A narrowing window to understand AI,”published in this week’s issue of Science science.org/doi/10.1126/sc… @cervisiarius @EPFL @MSFTResearch
Strengthening biosecurity in the era of AI blogs.microsoft.com/on-the-issues/… @MSFTIssues
Patrick Winston’s last lecture on “How to Speak.” In 2019, as I was getting mic’d to give a lecture at MIT, Patrick slowly came down the auditorium steps with a cane. He handed me a copy of his new book, and we caught up briefly. My host later told me he was very ill and had
At Build, we announced the general availability of Microsoft Discovery, a platform for accelerating scientific and engineering R&D. Discovery orchestrates leading-edge scientific models through cycles at the heart of science: evidence → hypotheses → analysis & experimentation
AI and the firm: Empowering organizations to climb the gradient to their own AI frontiers. @satyanadella x.com/satyanadella/s…
On related work & directions with agents: opportunities and prototypes described in this video are still as fresh as they were decades ago. Enjoy! microsoft.com/en-us/research… x.com/erichorvitz/st…
Seems like yesterday! x.com/tgupj/status/2…
