Vincent Conitzer

how not to convince someone to not be paranoid

"what if we trained an LLM only on data from after 1900?"

Game Theory for AI Agents

looking forward to seeing some good friends tomorrow
events.seas.harvard.edu/event/cs-col...

(2/2) (I am actually interested in whether and how we can say anything about AI consciousness -- www.cs.cmu.edu/~conitzer/LL... -- but I don't think it does much to just ask an LLM for a probability that it is conscious.)

Anthropic CEO Says Company No Longer Sure Whether Claude Is Conscious Anthropic CEO Dario Amodei said he didn't know whether his Claude AI was conscious, but was strikingly open to the possibility.

(1/2) From the article: "Suppose you have a model that assigns itself a 72 percent chance of being conscious [...] Would you believe it?"

Hmm. If you walk into a hospital and someone tells you "I think there's a 72% chance that I am a doctor," what would you think?

futurism.com/artificial-i...

Very sad to hear of Joe Halpern's passing. He influenced so many people, myself included, for the better. Even when his body was getting weak, his mind was still sharp, and he was still helping me, kindly pointing me to important work.

AIES 2026 will be in Malmö, Sweden, October 12-14. Abstract due May 14, paper due May 21.
www.aies-conference.com/2026/

"What if Pythagoras had failed to install a security update? Be specific."

the sciences constantly need to adapt to the elements changing weight

air resistance

an underappreciated feature

AI Testing Should Account for Sophisticated Strategic Behaviour This position paper argues for two claims regarding AI testing and evaluation. First, to remain informative about deployment behaviour, evaluations need account for the possibility that AI systems und...

(2/2) One concerning development is that models are starting to be able to recognize when they are being tested for safety. This upends traditional approaches to safety testing, turning it instead into a strategic game between the tester and the model. arxiv.org/abs/2508.14927

(1/2) proud to have played a small role in the second International AI Safety Report! internationalaisafetyreport.org/publication/...

gave brief closing talk "Cooperative AI: How to set up a contest for environments that aren’t zero-sum" @Algorithms with a Purpose '26 @cmu.edu, where teams competed programming bots to run restaurants, with the option to sabotage your competitor by throwing away their ingredients.
awap.acmatcmu.com

Also thanks to the amazing committee: Fei Fang, Tuomas Sandholm, Ben Levinstein, and Stuart Russell! (There's gotta be a better way to do these zoom pictures though...)

(2/3) pic left to right: Vince, Caspar, Emin (zoom), Emanuel (zoom), Jiayuan, Carlos, Tuomas

(1/3) Congratulations to my student Caspar Oesterheld (second from left) on a successful defense and an important dissertation "New foundational ideas in cooperative AI"! www.andrew.cmu.edu/user/coesterh/

Today is Holocaust remembrance day. How it got to that:
www.auschwitz.org/en/history/b...

Moral Change or Noise? On Problems of Aligning AI With Temporally Unstable Human Feedback Alignment methods in moral domains seek to elicit moral preferences of human stakeholders and incorporate them into AI. This presupposes moral preferences as static targets, but such preferences often...

Our paper in the alignment track of AAAI'26: what if human feedback is unstable over time?
arxiv.org/abs/2511.10032

On the Edge of Core (Non-)Emptiness: An Automated Reasoning Approach to Approval-Based Multi-Winner Voting Core stability is a natural and well-studied notion for group fairness in multi-winner voting, where the task is to select a committee from a pool of candidates. We study the setting where voters eith...

Tomorrow (F) morning at 11am SG time, Emin Berker and Emanuel Tewolde will give a talk at AAAI on our paper below!
arxiv.org/abs/2512.16895

not a great look for Google

Fortress Protocol Listen and make your own on Suno.

"Oh look AI is figuring out human preferences from our behavior so things will be fine."
Suno (music-generating AI):
suno.com/s/pmQIY73ulk...

simple

opening for new department head for CMU Philosophy!
www.cmu.edu/dietrich/phi...

Soccer goalkeepers have to make some tough decisions during the game. This should clear things up.

Markus Brill Markus Brill Associate Professor, Department of Computer Science Tutorial Fellow, Oriel College University of Oxford

Congratulations to Markus Brill (and to Oxford) for starting a faculty position at Oxford CS! sites.google.com/site/brillma...

"could my teacher's husband and my wife's teacher be the same person?"

more on having your son as your elementary school teacher

Apparently in the 1960s/70s anything was possible.

It is now ridiculously easy to make little games like this that you can just play in your web browser. (This one made with Claude.)
www.cs.cmu.edu/~conitzer/ke...