Biologically-inspired AI and Mortal Computation

[@MachineLearningStreetTalk]

MLST is sponsored by Tufa Labs: Are you interested in working on ARC and cutting-edge AI research with the MindsAI team (current ARC winners)? Focus: ARC, LLMs, test-time-compute, active inference, system2 reasoning, and more. Future plans: Expanding to complex environments like Warcraft 2 and Starcraft 2. Interested? Apply for an ML research position: [e-mail]

[@janerikbellingrath820]

Honestly, I love this channel so much. Almost every day new high-quality content. Would love to see an episode with Thomas Metzinger on consciousness and minimal phenomenal experience btw!

[@cammccauley]

You guys should talk to Professor Michael Levin and potential have him on the show again!

[@MachineLearningStreetTalk]

We did, check our patreon! Will release soon

[@cammccauley]

YAY!

[@jasonabc]

Fantastic show been following Alex's work for a while now and glad to see him on the show.

[@earleyelisha]

I loved this discussion! I’ve been ranting about this for years and hadn’t heard anyone discuss each of these cohesive components prior! Thanks guys!

[@-mwolf]

So good to see other people working on exactly this!

[@aitheignis]

This episode is super great. MLST is on a hot streak lately. The part about effect of architecture on the computation & about the efficiency of brain compared to ANN really remind me about this old review 'The economy of brain network organization' that suggested that brain modularity (i.e. its architecture) arise from resource constraints. This very modularity might also be the key to difference between the way human learn vs ANN.

[@MLDawn]

May I ask you to try and have a conversation with Prof. Jakob Hohwy from Monash university? He is the author of the book entitled: 'The Predictive Mind' and indeed he is a close collaborator with Prof. Friston. Jakob has an amazing way of describing predictive coding (which is a free energy minimising tool), especially from the perspective of individual neurons as to how they can realise Bayes' rule and accomplish the task of evidence accumulation so that the agent shall survive in the world.

[@MachineLearningStreetTalk]

Will look into him, thanks

[@JLongTom]

Great shout! It's a brilliant book that complements Andy Clark's Surfing Uncertainty, which considers similar topics from a somewhat higher-level perspective. Andy is another obvious pick for a chat!

[@yeetdeets]

When is this coming to spotify?

[@Soul-rr3us]

Great ep!

[@zandrrlife]

Best channel on KZbin by miles.

[@inquisition3000]

7:30 my mans really said "the medium is the message"

[@iamr0b0tx]

As a developer I will like to learn details about how these things work, how they are implemented and what thier limitations are. I will check out what has been published online, Nice work I enjoy any non-deeplearning approach

[@_ARCATEC_]

💓 Inspired by Alexander Ororbia's conversation on Mortal Inference Learning Systems... System Inference over Mutual Learning 'SIML' Hint form: S²I³/ML²X Intellidoscope equation including coordinates XZY+Q: •X(s zi q(m l ) Z𝐈 ( L M)Q zi )Y• An example of SIML is the bridging of AI and biological agents through shared learning environments. This relationship is an ongoing exploration of mutual learning on a shared value system. Potential benefits of these interactions include enhanced adaptation and channeling of intelligence. What interesting times we find ourselves.

[@maxlee3838]

How does his description of predictive programming differ from hopfield networks?

[@roelljr]

Fantastic content as always! This discussion really helps clarify some key points that often get overlooked when comparing artificial intelligence systems to human intelligence systems. It seems obvious that the biological substrate and morphology of a system plays a crucial role in shaping how intelligence is expressed and understood, which is why no artificial system can achieve full parity with human intelligence without mirroring the underlying biological foundation. However, this doesn’t mean that artificial general intelligence-encompassing reasoning, planning, decision-making, etc.-can’t be achieved in non human systems. It just won’t mirror human intelligence for the same reasons Orobia discusses-mainly that-you can’t simply transplant someone’s mind into a different body without changing the intelligence itself, because intelligence (generally defined) is a system deeply influenced by its physical substrate. For humans, that substrate-our bodies-drives motivations like maintaining thermodynamic nonequilibrium to preserve self-organization, which we experience as self-preservation. This biological imperative can even alter the morphology, as seen in the impact of hormones. Therefore, the physical substrate plays a major role in both how intelligence is realized and how we define it. And what I like about Orobia is that he correctly talks about "Biomimetic Intelligence", which is much more clear and definitive idea. What’s particularly important, and which I am glad they brought up, is the point made at 13:53: achieving something as efficient as the brain, approaching the Landau limit, comes with trade-offs. If you optimize for efficiency, you lose generality and robustness. These considerations are crucial to understanding what “intelligence” is and how it should be defined. Without this foundational understanding, discussions on the topic often miss the mark entirely.

[@MachineLearningStreetTalk]

Thank you!

[@Gnaritas42]

You are one of my favorite channels to actually watch.

[@personanongrata987]

7:50 | "You cannot think of the calculations or computations that are being carried out by the system independently of the substrate that instantiates it." --

[@richardg.lanzara3732]

In my book " Origins of Life's Sensoria, " I describe a biomimetic framework using cellular receptors.

[@damianlewis7550]

UPenn Prof Andrea Liu’s research on substrate inference in metamaterials could be really pertinent here.

[@Jacob011]

I regard Robert Rosen as the most brilliant mathematician (and scientist) who pronounced on the issue of "what is life?". He concluded that the entailment structures characterizing organisms (closure to efficient causation) cannot be simulated. I wonder if Ororbia is aware of his ideas. If not 👉 Rosen, R., Life Itself or Rosen, R., Essays on Life Itself.

[@ginogarcia8730]

MLST - With all these dizzying talks... I hope I could get an answer for -> what the hell does the layman do now? Haha. I'm still stuck on Dario Amodei's letter like a day ago. Will there be superintelligent AI in about 2 years? And if so, what the heck can anyone do? What should we do? haha

[@antonystringfellow5152]

Good question! This is by far the best channel I've found for all things AI but there' really aren't many good ones. Most contain a lot hype or are just low-information. This channel is excellent but isn't easy for the layman to follow. As for the question of ASI: We're not even close to AGI yet but the answer in both cases is no-one has much of an idea when it comes to the timescale. It's not an easy task trying to predict when something will be created when you don't even know how it can be done. We don't even know how the human brain creates general intelligence. Following this channel will help to get an idea but the layman may need to spend some time studying the subject. That could easily take 100+ hours 😬

[@melonburst8563]

"The C word is welcome here" I'm gonna take this out of context and make a spicy blog post

[@Carl-md8pc]

Thanks

[@DJWESG1]

At some point Hinton will have to backtrack on what hes often said about noam Chomsky.

[@DJWESG1]

Btw, this is a conversation ive already had with chatgpt last year. As well as many might already have. So its already got the ideas cooked into it. Specially that of embodyment and childhood development.

[@jaden5292]

Bro come on.

[@DJWESG1]

@@jaden5292 where we going?

[@earleyelisha]

Basically instantiating entities(i.e. souls in religious context) from the infinite entity space through embodiment.

[@MrBillythefisherman]

With regards to energy efficiency Im sure we'll move our neural nets to quantum computing when they get large enough and Id guess they would massively more efficient than our brains are.

[@TheMrCougarful]

See: Ghost in the Shell

[@m00nshi]

If intelligence is mortal and preserving identity, you will be considered a potential threat to that identity and a rival - that's how animals work.

Until they go woke... then they self immolate.

[@damianlewis7550]

@@m00nshi Explain

[@m00nshi]

@@damianlewis7550 have you read 3 body problem?

[@damianlewis7550]

@@m00nshi Yes and started the other books. Not sure how Markov Blanket descriptions of autopoiesis in living things relates to scifi about advanced aliens contacted by a foolhardy Chinese scientist. As to some rule you’ve invented that animals automatically consider other animals as potential threats, that doesn’t explain symbiotic relationships or animals that foster other’s young. Also, the Dark Forest Trilogy is about civilizations hiding (flight) because of a few violent (fight) advanced alien civilizations. As it is, it isn’t sensible to confuse fiction as fact.

[@furtsmagee1513]

🤖🧠

[@d.lav.2198]

Where angels fear to tread...

[@challahsmith3257]

Artificial ingredients

[@flopasen]

cool, bitcoin is also biologically inspired

[@damianlewis7550]

No, it really isn’t.

[@ssehe2007]

Why are British and French scientists routinely more charismatic and well-adjusted? At least outwardly. Maybe they torture baby kittens first thing when they get home. But outwardly they exude signs of physical wellness and social aptitude. It’s as if you have to be maladjusted or on the spectrum in the states to be interested in these things.

[@mattforsythe5037]

Beff jezos chip !!

[@BrianMosleyUK]

🙋

[@danielschwegler5220]

I didn't understand anything

[@damianlewis7550]

😂

[@miken-xm1fh]

Iron Giant, right?

[@angloland4539]

❤️☺️🍓👋

[@MrBillythefisherman]

We already have bio inspired A I and it will never be the same as biological intelligence. Someone will always find some difference. So this is all a side show now if we can say the output of one black box is better than another in a range of economic activities. Let's just say you did achieve perfect artificial biological intelligence then it woild be the same as the human brain and we've been able to create those sincenthe beginning of the human species. Not to say it's not a fun topic.

[@denzelcanvasYT]

i think this js more useful for building life that can survive under conditions where humans cant.

[@Astronomikat]

Around the 35 minute to 40 minute mark he starts to talk about it in a way that makes me very uncomfortable as to where we are heading, if that's what we're doing. Blade Runner was not a blueprint for the future, it was a horrible dystopia where engineered people were programed to be our slaves.. A lot of what the tech industry seems to want from AI is workers without the people, and given the socio-dynamics of our time, that doesn't bode well for people, let alone engineered "entities" that learn about the world like children.

[@damianlewis7550]

As interesting as it is, I don’t think you can take Blade Runner literally, especially as it is about the relationship between memory, identity and nature/nurture. All possible futures are possible. Capitalism requires slaves if you hadn’t noticed, and there is literally half a planet of people living in what we in the West would consider slavery. I’d rather have machines (biological or otherwise) being the slaves rather than humans. Just because they are machines doesn’t mean we would have to treat them badly, there’s no rule that says it has to be that way. You wouldn’t smash up a Porsche just for the hell of it, especially if it was doing all the tedious and dangerous work you would rather avoid doing yourself. We are a very long way from creating sentient machines and sentience may not even be achievable if materialism is wrong (see the fascinating discussion between Seth, Goff and Vervaeke recently on the Wonderstruck Podcast).

[@Astronomikat]

​@@damianlewis7550 Yes, but the world in which those ideas unfold is one in which entities are engineered to be slaves. And yes, I am well aware capitalism requires slaves. Personally, I am not a fan of the capitalist mode of production. In fact, I think it's only slightly better than feudalism, and only then because labor power at various times organized to demand their human rights. I think it is unfortunate that given the current state of the world, this technology will be applied in a way that will rob humans of their labor power, and potentially, if mortal computation is to be achieved, the privileged classes will create yet another underclass to enslave. What will become of those outside of the privileged classes, devoid of any remaining labor power, I leave to the imagination. But I think one only need look at how the middle east and global south is treated to get the flavor of it.

[@damianlewis7550]

@@Astronomikat I don’t buy into the hype that OpenAI, Tesla et al are generating with what is effectively smoke and mirrors in an environment of Tulip Mania. So I don’t worry that humans will be replaced any time soon. I do however worry that this is just another example of oligarchs (and wannabe oligarchs) trying to concentrate their power and wealth. Capitalism isn’t the problem, neo-fuedalism is.

[@DJWESG1]

Correct, human beings still crave slaves and servants. But it doesn't have to be that exact potentiality.

[@Astronomikat]

@@DJWESG1 I sure hope it's not that outcome, but again, I'm seeing indicators that that is where we're currently heading. I don't think we're just magically going to get "the good outcome" from all this, unless people come together, get informed, and force the issue.

[@MrBillythefisherman]

With regards to energy efficiency Im sure we'll move our neural nets to quantum computing when those computers get large enough and Id guess they would be massively more efficient than our brains are.