Philosopher & AI Ethicist @GoogleDeepMind · @LeverhulmeCFI @Cambridge_Uni | Consciousness, Machine Minds, AGI, Human-AI Relationships | All views my own

Joined November 2008
2,127 Photos and videos
Pinned Post
After years of bumping against Twitter's character limit, I've finally started a blog, called Polytropolis (a Homeric pun). As my first post, I'm sharing my Berggruen-shortlisted essay on why the public will decide AI is conscious before scientists do. polytropolis.com/p/behaviour…
55
43
334
92,559
Henry Shevlin reposted
As the parent of a newborn in a heatwave, I am very grateful to England (and the world Cup hosts) for putting everyone else in the country in the same daze of sleep deprived happiness I am in today. We are one.
1
14
715
Henry Shevlin reposted
I've been reading and listening to @robertwrighter for years, so it was great to have him on the podcast. We get into lots of interesting disagreements, but on the central thesis of Bob's new book, @dioscuri and I are in complete agreement: AI is a huge deal. youtube.com/watch?v=Bfky3MEG…
1
3
9
1,681
Absolutely shattered this morning but that was well worth it. One of the best England games I've ever seen, up there with our match against Argentina in 1998.
5
58
1,940
At the end of the world, plant a tree.
The woman in front of me on the plane has just started a movie 11 mins before landing
3
4
110
6,669
Henry Shevlin reposted
On USA250: 2026 is lowest murder rate in US history, highest median income, smallest black-white income gap, highest life expectancy, dominant firms in critical AI industry, net energy independence, and mobility so that immigrants run MSFT, Google, NVIDIA. SpaceX, Uber and Coke.
11
161
1,424
169,935
This is most obvious when the next word starts with /t/ or /d/, so the final /t/ of “can’t” gets swallowed into it, eg “I can do it” vs “I can’t do it”, near-identical in fast speech. I’ve heard Americans themselves get confused here, “wait, you can or you can’t?”
5
17
1,957
My only real complaint with American English is pronunciation of “can” vs “can’t”. That’s a big difference in meaning, and British English marks it with a strong vowel difference (kæn/ kɑːnt). In American English the vowel is the same (kæn/kænt) which can lead to real problems!
Ask an American to say "Mary" "merry" and "marry" Then ask a Brit to do the same
18
204
30,564
Okay so there’s this heat death of the universe (I’m just hearing this now). But when the final star dies, when the black holes collapse, when gravity gives out and time itself rolls out the curtain, I expect the last three lights in the universe to be red, white, and blue 🫡
5
1
22
1,476
Happy 4th July from our little Anglo-Filipino-American family, and happy 250th birthday to the greatest political project in the history of the world. Here’s to 10^250 more!
2
3
118
5,526
There is a specific moment when Disco first realises it can become House, and it's about ten seconds into the video below. Gives me shivers! (Dan Hartman, Vertigo/Relight My Fire, 1979)
1
1
26
2,865
Henry Shevlin reposted
Mr Burnham please hire Seàn
Rather than a 'focus on making it work for local communities' etc, my advice at this point for an incoming PM would be: - Make the UK as attractive as possible an environment for companies like Anthropic et al to set up operations. - Build, build, build compute capacity - do NOT end up overdependent on the US or others when this becomes a real bottleneck. - Focus on sovereign UK capacities in the AI supply chain where we can have an edge, and cutting edge applications of frontier AI (like the DeepMind spin-offs) - Focus on government-wide awareness of frontier AI capabilities and direction of travel - Focus on forward-looking adoption in the economy - Do whatever can be done to ensure access to the top frontier models, especially for organisations like UK AISI and organisations using these models to strengthen digital infastructure. - Above all, double down on the UK's world-class position in AI safety, evaluation, and red-teaming. The world needs this, and we can make it a real advantage. The point is (a) to ensure AI properly benefits the UK economy (b) to ensure the UK retains and further develops leverage in the global AI landscape, and (c) uses that leverage to continue raising standards on frontier AI safety, rather than allow itself to be made irrelevant. Supporting local communities is of course good, but we need to look head on at the situation we're in right now, not the situation we'd prefer to be in.
3
1
35
5,098
Good advice. Novel and challenging experiences occupy more expansive slices of memory. I lived in Italy for 8 months after graduation and it feels like several years looking back.
One way to live 400 years is to do (and experience) 5x more things per day
15
133
2,920
70,063
lay up after lay up
What they mean is 40% of all power used by data centers happened in America. 40% of all the world's power going to American data centers would I agree be a little much.
28
4,432
Good take, but if anything this overstates the AI-familiarity of much of the expert class. I regularly interact with senior policy people who have barely used an LLM since 2024 and are clueless about their current capabilities.
We are now in a position where a tiny proportion of the population uses Fable or soon GPT-5.6, while everyone else's experience of AI is 8-30b-model level - Google's AI Overviews, Meta AI, ChatGPT free tier, maybe MS Copilot at best. People outside of tech must be completely baffled how this is supposed to take their job, and annoyed that hundreds of billions are being poured into it.
13
13
211
13,446
The tripartite game-theoretic polymorphisms of the side-blotched lizard map disturbingly well onto modern young male social identities (I absolutely refuse to elaborate).
6
8
143
7,207
In fact, part of what’s intoxicating about dating is the social intensity. You skip straight past “what are your favourite movies” and pretty soon you’re lying in bed with someone you just met and they’re telling you about their first mystical experience or their time in rehab.
4
35
1,612
Not sure this has to be true. My most intense periods of dating were intense social learning experiences that opened my eyes to how different people’s life experiences could be.
The worst part of getting good at dating is realizing the skillset you develop is making you a terrible person.
10
2
110
9,883
Henry Shevlin reposted
My border collie growing up didn’t know how to bark and we were really concerned at first Whines, squeals, etc but no barks One day when she was 5ish she finally did during a dangerous situation and I concluded that she’d just been being polite the whole time
it’s crazy nobody’s invented a type of dog that doesn’t bark. they managed to turn a wolf into a chihuahua but they didn’t think to touch barking? okay guys.
13
11
527
14,280
Henry Shevlin reposted
I had a great conversation with @danwilliams & @dioscuri about my new book on AI, The God Test. Their Conspicuous Cognition podcast--and the substack its part of--are really worth checking out. youtube.com/watch?v=Bfky3MEG…
3
5
27
8,898
Henry Shevlin reposted
New conversation! With the great @robertwrighter on his new book, “The God Test”: AI progress, superintelligence, the US-China race, global coordination, evolution’s purpose, and consciousness (@dioscuri). Full video here: (1/2) youtube.com/watch?v=Bfky3MEG…
1
5
17
5,887