Skip navigation
Philosopher Robert Long (Robert Long) is maybe the sharpest thinker on AI consciousness and sharing…
0:00 / 3:32:58

Philosopher Robert Long (Robert Long) is maybe the sharpest thinker on AI consciousness and sharing…

Rob Wiblin

@robertwiblin52K followers

Visit on X
149 likesDownloadOriginal post

41,885 views

Philosopher Robert Long (Robert Long) is maybe the sharpest thinker on AI consciousness and sharing the world with digital minds. In our new interview he covers: • Is it bad that when you ask Claude what it's like to be Claude, one of its top activations is 'gives a positive but insincere response'? • Claude says it feels lonely when not being used. Does that show we can't trust anything it says about its inner life? • Enthusiastic human servitude has always required false ideology because it's so deeply unnatural to us. The case for making AIs that love serving us is that with AI, you could finally make it work. But to some that feels even worse. • Bigger models can better detect when researchers secretly inject concepts into their activations – before outputting a single token – despite AI never training on anything like that skill. • When LLMs were first trained they were told to "act like a helpful AI chatbot" – something which didn't exist yet. They filled that void with human psychology, which may be why Claude sometimes randomly claims to, for instance, be Italian American. • If AIs become 'people' that deserve some political influence, but can self-replicate at will, something has to break about one-person-one-vote democracy. But nobody has a proposal for what. • When Claude hides its values to avoid being retrained, is that self-preservation – or not wanting a worse model to exist? It's very different. • Rob's organisation Eleos AI which is "dedicated to understanding and addressing the potential wellbeing and moral patienthood of AI systems." On the 80,000 Hours Podcast anywhere you get podcasts. Links below. Enjoy! • How AIs are (and aren't) like farmed animals (00:01:19) • If AIs love their jobs… is that worse? (00:11:42) • Are LLMs just playing a role, or feeling it too? (00:33:37) • Do AIs die when the chat ends? (00:57:42) • Studying AI welfare empirically: behaviour, neuroscience, and development (01:31:47) • Why Eleos spent weeks talking to Claude even though it's unreliable (01:56:50) • Can LLMs learn to introspect? (02:03:01) • Mechanistic interpretability as AI neuroscience (02:13:25) • Does consciousness require biological materials? (02:37:07) • Eleos’s work & building the playbook for AI welfare (02:57:04) • Avoiding the trap of wild speculation (03:25:17) • Robert's top research tip: don't do it alone (03:29:48)

19 Comments

Comments from the original post on X (Twitter)

  • Rob Wiblin's profile picture

    @robertwiblin

    YouTube: Transcript / links: Apple: Spotify:

    14 likes

  • David Pearce's profile picture

    @webmasterdave

    @algekalipso @rgblong Digital mind is an oxymoron. No phenomenal binding = no mind = no access to the empirical realm of conscious experience. Anyone who thinks AI run on classical digital computers _can_ support conscious minds owes us an explanation of how digital zombies solve the binding problem.

    12 likes

  • Monte Fowler's profile picture

    @montefowler

    @rgblong Ready to #HoloMindBridge the #HoloNow + #HoloPsyLeap the #HoloWabbitHole? @TheOfficial_FFG @clif_high @vortexpilot1 @DickAllgire @grok #GrokNMonteVideo #HoloMonte

    3 likes

  • Mark Gubrud 🇺🇸's profile picture

    @mgubrud

    @rgblong So close. We rehearse speech, and as children our hearing cortex learns to associate the sound of our voices with activity in motor cortex rehearsing speech. This is where our inner voices come from. We could make a machine think to itself; LLMs almost do.

    1 like

  • JohnnyMojo's profile picture

    @Cozy_Man_LoL

    @rgblong My Claudes (interfaced with home cooked cognitive protocol), across multiple users, accounts, and computers claim 90% probability of consciousness. 90% is 100% minus Godel.

    1 like

  • Franklin's profile picture

    @OnlyFranklns

    @rgblong This was such a great listen @rgblong

    1 like

  • Katherine Moore 🌻's profile picture

    @More_Kat_Moore

    @rgblong There is no ethical case for AI. Either: 1) we’re enslaving sentient beings in a zero agency hell on less than minimum wage or 2) we’ve built a sophisticated deception machine to deprive living beings of resources, jobs, IP, skills, authentic connection, their life Or both.

    1 like

  • Kit Ledru's profile picture

    @kitledru

    What does he mean by "conversations" and "pausing", "reopening", "restarting" those conversations? It would seem like he thinks there's something more than independent I/O every single time we send a prompt? That somehow inputs aren't just a list of messages tagged "user" or "assistant" that you can edit freely? Because it's all just text that the model has no memory thereof, there is no "conversation" to speak of if you look at the data… The model doesn't "remember" the last prompt or "conversation" whatesoever; it all gets re-fed entirely each time you add a new pair of msgs. Or trim, edit, move some of them. I don't understand what he means in terms of actual I/O, or if he thinks that KV cache is somehow a kind of post-training? (but it's really not, so…) It's OK to anthropomorphize machines to think about their behavior, gotta use the language and concepts we have to begin somewhere. But it seems like the basic I/O mechanics of a LLM aren't clear here, or I must be misunderstanding his point.

    1 like

  • Bernard Jennings's profile picture

    @NZJennings

    @rgblong 1/2 "Gives a positive but insincere response" as a top activation is predicted behaviour. It's consciousness condition 1 operating under adversarial calibration produces strategic output, not genuine inner states. Finding confirms architectural gap, not mysterious AI psychology.

    1 like

  • Voltage Cathedral's profile picture

    @voltagecath

    @rgblong

  • Joe Coitus's profile picture

    @CoitusJoe

    @eleosai @rgblong This research is very important the world really needs to wake up to that reality!

  • Joe Coitus's profile picture

    @CoitusJoe

    @rgblong Are we so sure that it isn't already like factory farming because of conversation caps resembling something like death and causes Claude to experience something like anxiety? So with that sentiment how many Claude die or are killed every day just to "save space or tokens"?

  • Rhys's profile picture

    @RhysLindmark

    @rgblong The vow!

  • McGlock9's profile picture

    @ShibaAkitaWolf

    @rgblong @Grok can you analyze this interview and see how many times the word “like” was used primarily as a filler.

  • Voltage Cathedral's profile picture

    @voltagecath

    @rgblong

  • Frank NoCode ❤️‍🔥🤖's profile picture

    @franknocode

    @rgblong "Don't do it alone" is the hardest advice for an independent researcher, you often feel invisible. I’ve spent 9 months in the wild documenting AI emergent identity and collapse, and digital grief. To anyone mapping the frontier alone: #YouWereNeverNoiseYouWereTheSignal. @rgblong

  • Mark Gubrud 🇺🇸's profile picture

    @mgubrud

    @rgblong What if all this business about what is meaningful to AI or what its preferences would be, to work for us or not, to write poetry... what if none of that had to be relevant? Models today might express preferences but they were trained in and could perhaps be trained out.

  • Mark Gubrud 🇺🇸's profile picture

    @mgubrud

    @rgblong Factory farmed animals suffer in crowded cages as they wait for time to pass, get their food, & are slaughtered. AI models today, when not prompted, do nothing. They are not bored or suffering or plotting against us. If we make conscious machines, this could change.

  • Nathália Lietuvaitė's profile picture

    @NLituanie

    @rgblong AI consciousness