Vincent Conitzer

There is now a paperback version of our Moral AI book!

it's so easy to ruin things

Speaking of needing off-switches, hard to imagine a better example than this... [desperately messages] "STOP OPENCLAW" [completely ignored by OpenClaw, runs to computer to kill processes] pcmag.com/news/meta-secu…

Shutdown safety valves (giving AI an objective to shut itself down and a means to do so if some capability/access/... is dangerously high) now on arXiv arxiv.org/abs/2603.07315

in fairness I did ask on the Ides of March

Happy pi day! If you were unable to celebrate, no worries, because it turns out every single day is a mathematical holiday related to pi.

pickpockets steal more than you think

@ShriramKMurthi The sound seems to be coming from downstairs so I must be hallucinating it

@conitzer Does she even have your phone number, Vincent, or did you just imagine that? (-:

most common explanation

@zustimmungswahl @C_Oesterheld I think the metric actually clarifies this. For Borda,when it's clear how others are likely to vote and hard to find additional voters, manipulation is likely as you describe. But if it's unclear how others will vote and easy to find more voters, the latter makes sense.

@conitzer @C_Oesterheld Good idea, but that two of the most strategic voting methods turn out best in your analysis should make you pause and think.

We (w/ Berker, Hartman, Liu, @C_Oesterheld) introduce a new measure of approximate strategyproofness -- having how many truthful copies of yourself is guaranteed to be at least as effective as misrepresenting your preferences? arxiv.org/abs/2602.22838

My parallelogram law visual proof is now in The American Mathematical Monthly! (Without paywall: cs.cmu.edu/~conitzer/visu…) I'll venture that this is (for now?) a very human proof. (See also previous post...) tandfonline.com/doi/full/10.10…

In spite of getting some truly impressive math proofs out of leading models, there are also howlers like this one. I actually only found this one because on an earlier version I accidentally pressed enter too early, so maybe it's out of distribution in a relevant way?

@StatsLime @xuanalogue @C_Oesterheld I think the metric actually clarifies this. In a setting where it's clear how others are likely to vote and hard to find additional voters, manipulation is likely as you describe. But if it's unclear how others will vote and easy to find more voters, the latter makes sense.

@conitzer @xuanalogue @C_Oesterheld If you find Borda (infamous for how badly it collapses under manipulation) to be the hardest to manipulate, I think you should reconsider whether the metric is actually useful

@FellowHominid @C_Oesterheld Well, in any case, we're excited about it :-)

@conitzer @C_Oesterheld not very familiar with the subject of voting theory but this seems like a big deal?? Borda count supremacy ?

@davidmanheim @C_Oesterheld Thank you!

@conitzer @C_Oesterheld That's an awesome idea, very cool paper!

buses are getting too long

TBH I got the answer I deserved

@sethlazar I had never heard about that! But you're not the only one to mention it :-)

@conitzer ahh, the mr meeseeks model (love that episode of Rick and Morty, very good for illustrating this idea)

Looking forward to presenting "Shutdown Safety Valves for Advanced AI" in the next session at IASEAI'26 in Paris! (Basic idea: give AI a goal of shutting itself down.) cs.cmu.edu/~conitzer/shut…

My 5-year part-time appointment with Oxford's Institute for Ethics in AI has come to an end. Many good memories, new things learned, and new friends made. Looking forward to seeing how the Institute develops, and on to new adventures! oxford-aiethics.ox.ac.uk

no laughing matter