Dr. MAXWELL RAMSTEAD - The Physics of Survival

Equilibrium and Existence!
Patreon: / mlst
Discord: / discord
Join us for a fascinating discussion of the free energy principle with Dr. Maxwell Ramstead, a leading thinker exploring the intersection of math, physics, and philosophy and Director of Research at VERSES. The FEP was proposed by renowned neuroscientist Karl Friston, this principle offers a unifying theory explaining how systems maintain order and their identity.
The free energy principle inverts traditional survival logic. Rather than asking what behaviors promote survival, it queries - given things exist, what must they do? The answer: minimizing free energy, or "surprise." Systems persist by constantly ensuring their internal states match anticipated states based on a model of the world. Failure to minimize surprise leads to chaos as systems dissolve into disorder.
Thus, the free energy principle elucidates why lifeforms relentlessly model and predict their surroundings. It is an existential imperative counterbalancing entropy. Essentially, this principle describes the mind's pursuit of harmony between expectations and reality. Its relevance spans from cells to societies, underlying order wherever longevity is found.
Our discussion explores the technical details and philosophical implications of this paradigm-shifting theory. How does it further our understanding of cognition and intelligence? What insights does it offer about the fundamental patterns and properties of existence? Can it precipitate breakthroughs in disciplines like neuroscience and artificial intelligence?
Dr. Ramstead completed his Ph.D. at McGill University in Montreal, Canada in 2019, with frequent research visits to UCL in London, under the supervision of the world’s most cited neuroscientist, Professor Karl Friston (UCL).
Pod: podcasters.spotify.com/pod/sh...
scholar.google.ca/citations?u...
spatialwebfoundation.org/team...
/ maxwell-ramstead-43a19...
/ mjdramstead
VERSES AI: www.verses.ai/
Intro: Tim Scarfe (Ph.D)
Interviewer: Keith Duggar (Ph.D MIT)
TOC:
00:00:00 - Intro and philosophy
00:06:16 - Intro to Maxwell
00:09:50 - FEP
00:20:58 - Markov Blankets
00:43:05 - Verses AI / Applications of FEP
00:57:45 - Potential issues with deploying FEP
01:02:40 - Shared knowledge graphs
01:06:19 - XRisk / Ethics
01:16:47 - Strength of Verses
01:20:20 - Misconceptions about FEP, Physics vs philosophy/criticism
01:36:31 - Emergence / consciousness
References:
Principia Mathematica
www.abebooks.co.uk/servlet/Bo...
Andy Clark's paper "Whatever Next? Predictive Brains, Situated Agents, and the Future of Cognitive Science" (Behavioral and Brain Sciences, 2013)
pubmed.ncbi.nlm.nih.gov/23663...
"Math Does Not Represent" by Erik Curiel
• "Math Does Not Represe...
A free energy principle for generic quantum systems (Chris Fields et al)
arxiv.org/pdf/2112.15242.pdf
Designing explainable artificial intelligence with active inference
arxiv.org/abs/2306.04025
Am I Self-Conscious? (Friston)
www.frontiersin.org/articles/...
The Meta-Problem of Consciousness
philarchive.org/archive/CHATM...
The Map-Territory Fallacy Fallacy
arxiv.org/abs/2208.06924
A Technical Critique of Some Parts of the Free Energy Principle - Martin Biehl et al
arxiv.org/abs/2001.06408
WEAK MARKOV BLANKETS IN HIGH-DIMENSIONAL, SPARSELY-COUPLED RANDOM DYNAMICAL SYSTEMS - DALTON A R SAKTHIVADIVEL
arxiv.org/pdf/2207.07620.pdf

[@BrianMosleyUK]

I didn't get this the first viewing, making more sense second time around. One of the few long form channels I'll enjoy watching again. 🙏👍

[@rezamirkhani4747]

Thanks for the delightful treat you have prepared this Sunday ❤

[@pennyjohnston8526]

Great video - Thanks MLST! A multi domain consolidation point, a useful framework with associated terminology, tools and standards to enable the creation of complex, context understanding systems that are transparent !

[@nonchai]

Absolutely fascinating interview and my first exposure to this amazing FEP stuff- was encouraged towards the end of the interview that it could lead to interpretable ( interrogable? ) neural nets - AIs' etc - in the future

[@_ARCATEC_]

1:50:00 💓 "At the heart of every boundary is infinity, of which is boundless." In the Tillichian sense, utilizing what can be considered a hypergraph to point towards the approximate value of 0, while simultaneously conceptualizing the position of the computational limit, which is functionally referred to as a holographic screen providing an image of the object within what would mathematically be described as a Markov blanket, is difficult but fundamental. There is so much to say about this topic, but for now, I would like to share two numbers that really stand out to me. One is 10¹⁵, and perhaps the most difficult number to model of all is 0. it becomes evident that accepting approximation is intrinsic to the modeling process." Further contemplation... Holographic Screen over the Boundary of Unisonant Mind HS²/BU² Utilising the novel algebraic calculus of Intellidoscope we can construct the following formula: •(()())• the balance nested parentheses provide a syntactic framing. Let "XZY+Q" be coordinates, Let "H" be Holographic, "S" be Screen, "B" be boundary and "U" be Unisonant Mind. •X h(s z qb(u ) Z ( U)BQ z S)H Y• Progressing towards a conclusion: "H" has come to symbolise many different things from heat to holographic and even the Heart in it's most quintessential context. And although some forms of infinity remain undisclosed through our modelling we can still point towards them. That said there is an aspect of infinity somewhat disclose to us in that we can reasonably conclude there is a holistic and although we cannot fully fathom it's extent we can acknowledge it's temporalitie and dimensionality. d7•X d6 ( zd5 q(d1d2) Zd5 (d3 D4)Q zd5 ) d6 Y•d7 "To truly know something we must enter into it"

[@MMMM-sv1lk]

The whole part trying to explain the FEP could be better summarized by a musical concept - Harmony. Things that exist together must harmonize. That's what representing a part of the outside systen in your system means. Or so I would propose.

[@earleyelisha]

A reflection of Snarky Puppy’s - Lingus performance would point to your perspective. I share this mode of thinking.

[@MMMM-sv1lk]

@@earleyelisha yeah indeed such an exceptional performance 😊 glad you could relate... Tesla famously said: if you want to understand the universe think in terms of energy, frequency and vibration. Well these are the fundamental building blocks of music and the universe alike and that is both the universe within and without. I bet people a lot knowledgeable than me in FEP could figure out some analogies between these concepts. 🤔😌

[@pauloabelha]

Piaget was trying through Genetic Epistemology to bridge the gap from physics through biology to intelligent agents. A single epistemological framework that would unite emergent proprieties and explain how knowledge, agency and concepts can form that relate back to Reality. The FEP might just be the basis for such a neo-Materialism through a natural epistemology, where knowledge and agents are not separate, but an intrinsic part of Nature.

[@alexkiefer432]

solid discussion!

[@behrad9712]

Interesting 🙏

[@aldousd666]

I am fascinated! This seems, (as Dr Ramstead rather derisively implied with his comment about the EU AI law,) to be the only way to actually reason about correctness. Even if not, it sure beats playing RHLF whack-a-mole. I now have a bunch of reading to catch up on. I see shades of Jeff Hawkins work here too, at least in spirit. Interesting theory on qualia too. I kinda have a 'suspicion' that you are right about that, but I'm too naive in the theory to be able to justify that thought.

[@MLDawn]

How great this video was! I had no idea that MAxwell was Prof. Friston's PhD student!!! A quick question for you guys: At 13:20, does 'tracking' refer to tracking the ever-escaping expected posterior out there in the world, by capturing the dynamics of the world? More technically, through the generalised coordinates of motion? If so, it makes sense to me, since if I can track the world around me, then I can predict my sensations, and if I can do that, then it means I can actively and dynamically update my generative model of the world. Is this accurate? Thanks.

[@noomade]

Sesquipedalian and I love it! 😃

[@PeanutB]

scarfe always speaks with an almost poetic pedantry. i kind of love it.

[@earleyelisha]

I’m in Target looking at blankets.

[@MMMM-sv1lk]

😅

[@olivercroft5263]

Could this be related in some way to the RDoC (research domain criteria), deployed in the national institute for mental health for more broad mental health queries?

[@raminsafizadeh]

How is this notion different from enthalpy?

[@consistent1]

The FEP is new in its mathematical symbolic formulation. It was, however, a hot topic of debate among ‘mystics’ for millennia. My favorite encoding is the Toltec one, but the maps co-align if you know mmm the principal. BTW, from this POV the FEP amounts to having everything figured out with one notable exception - the sort of blankets that define an agent necessarily come in pairs. This is a crucial point, pun intended. Outer blankets define an agent’s boundaries while providing a certain degree of self-reflection for free. That’s a part the FEP crowd figured out in an amazing fashion. Brilliantly done!!! Maybe you are on to it, and I never came across (or did not recognize) this idea in the FEP ecosystem, but the Toltecs emphasize it to no end, so I will mention it because I agree. It is worth emphasizing - An outer blanket is coupled, necessarily, with an inner blanket, which is (under normal circumstances) way smaller. A point, if you wish. The inner blanket acts (necessarily) as the point where (drumroll, please) awareness is assembled. Accordingly, it is referred to as “the assemblage point”, for short. Think of it as a kind of shorthand to a serialization of the state the agent is in, if you wish, giving rise to coherence, continuity, and (with some help from ‘the outside’) awareness. Put another way, one implication of the whole being less than its parts is that the whole hides a virtually endless number of islands (of stability). Each island defines other possible wholes hidden from view in plain sight. What defines the (consistency of the) whole is a specific entanglement of free variables which creates the view, the active model, which the whole gets to experience from the inside. Each island is a possible position of the AP. If the AP moves to a different position (in relation to the outside blanket) the result is a completely different, yet completely coherent, perception of ‘reality’. Systems that have an AP at the same place, relative to their outer blankets, share features and may communicate. We all share the same place of our APs, and hence we all agree about the features of ‘physical reality’. According to those guys, from way back when, the FEP can and should be modeled as a map containing 4 aspects of (any system’s) being - Actions, generally placed in the north, represent the blankets from the outside. Rules/Reason, generally placed in the east, represent a serialization of the blankets and their interdependence (as a hypergraph). Resources/Options, generally placed in the south, represent the blankets from the inside. Intent, generally placed in the west, represent the interface between levels of abstraction. That’s an amazingly powerful ontology, in conjunction with some added rules of thumb, most notably the existence of the AP. The ontology is considered ‘energetically irreducible’, as far as we can tell, so it is the gold standard that may be used by modern ‘shamans’ to make dealing with the abstract, with awareness itself, fit our human concepts, sort of. That’s probably way more than is reasonable for a YT comment. :) Wow. Amazing work!!! What a great episode!

[@wp9860]

Regarding consciousness discussed at the closing stages of the video. Clearly, addressing consciousness within the FEP is in a speculative stage, which is no different than the stage consciousness is in by any other approach. I would still like to hear more discussion of consciousness. What is its function. Display was discussed, but what is the purpose of that display? Why consciousness, at all? ... Is consciousness a computational component of the human system? Bone, muscle, and sinew are not computational components. Is consciousness non-computational like those? If so, the FEP could never explain qualia because FEP is strictly a computational model. Does explaining consciousness even lend itself to an objective approach given that it only exists as a subjective phenomenon?

[@haldanesghost]

*reads the title* me, before watching the episode: "This is giving me some Lotka vibes. Did MLST get some physical biology in?" Excited to see this episode.

[@siarez]

I wish at around minute 50 Maxwell had give credit to Terrence Deacon for the "engine example." kzbin.info/www/bejne/eKepdmR3qpiGbrc

[@jamesg5804]

I got to 1:44... " Marketing Idealists." [ Young, material, marketing idealists. ] What basis of evidence have you provided? (Perhaps it's in the portion past 1:44...but 'I doubt it.')

[@tomripley7148]

The problem i have with the free energy principle is that its draws heavy on schools in cybernetics but never angelogenes it, yes I look at you sir ...

[@listentome4488]

Great video! Was just wondering if you guys were going to make a video about how I can reduce my reduce my electricity bill?🗿🦍 It getting kind of high and I really would like those savings to hit my bank account! Looking forward to those savings!🤞

[@RadekMirski]

1:59:20

[@matthewcurry3565]

If they also have an episode on how to avoid the upcoming draft, and meat grinder leave me a comment!

[@CandidDate]

I must be stupid. What in the heck are you talking about? The obvious is not some inverted obvious. The obvious is the obvious.

[@PremierSullivan]

That is a glorious mustache.

[@EasyLawBot2]

Thanks @Machine Learning Street Talk for posting this video about affirmative action / supreme court. Here are the viewpoints expressed by Supreme Court justices regarding affirmative action. 1) This case is about a group called Students for Fair Admissions (SFFA) who sued Harvard College and the University of North Carolina (UNC). They said that these schools were not fair in their admissions process because they were using race as a factor, which they believed was against the law. The law they referred to is the Civil Rights Act of 1964 and the Equal Protection Clause of the Fourteenth Amendment*. 2) The Equal Protection Clause is a part of the Fourteenth Amendment that says that every person should be treated equally by the law, no matter their race, color, or nationality. The SFFA believed that by considering race in admissions, Harvard and UNC were not treating all applicants equally. 3) The Court looked at the history of the Fourteenth Amendment and how it has been used in the past. They also looked at how other cases involving race and college admissions were handled. They found that while diversity in a student body can be a good thing, it must be handled in a way that treats all applicants fairly and equally. 4) The Court also looked at the idea of "strict scrutiny*". This is a way for the courts to look at laws to see if they are fair and necessary. If a law or policy is found to be unfair or unnecessary, it may not pass strict scrutiny and could be considered unconstitutional. 5) The Court found that the admissions systems at Harvard and UNC did not pass strict scrutiny. They said that the schools' use of race in admissions was not clear or specific enough, and it resulted in fewer admissions for certain racial groups. They also said that the schools' use of race in admissions seemed to stereotype certain racial groups, which is not allowed. 6) The Court also said that the schools' admissions systems did not have a clear end point. This means that there was no clear plan for when the schools would stop using race as a factor in admissions. This was another reason why the Court said the schools' admissions systems were not fair. 7) The Court decided that the admissions systems at Harvard and UNC were not fair and did not follow the Equal Protection Clause of the Fourteenth Amendment. They said that the schools' use of race in admissions was not clear, specific, or fair enough to be allowed. 8) However, the Court also said that schools can consider how race has affected an applicant's life. They can look at how an applicant's experiences with their race have shaped them and what they can bring to the school because of those experiences. 9) In the end, the Court decided that the admissions systems at Harvard and UNC were not fair and did not follow the law. They said that the schools' use of race in admissions was not allowed because it was not clear, specific, or fair enough. 10) So, the Court decided that the SFFA was right. They said that Harvard and UNC were not treating all applicants equally in their admissions process, which is against the law. They said that the schools needed to change their admissions systems to be fair to all applicants, no matter their race. *The Equal Protection Clause is a part of the Fourteenth Amendment that says that every person should be treated equally by the law, no matter their race, color, or nationality. *Strict scrutiny is a way for the courts to look at laws to see if they are fair and necessary. If a law or policy is found to be unfair or unnecessary, it may not pass strict scrutiny and could be considered unconstitutional.

[@azsx299]

"... kindling mental synapses with fresh insights" Bruh

[@isajoha9962]

Interesting to try out what an AI based on that model can do, compared to what exists now. 🤔🤓

[@isajoha9962]

I guess it might be like eg a LLM with "limited world view focused on specific things" kind of "entity". Like a "system filtered" LLM where the layers are optimised to work within similar context by being able to form how the LLM should act through self regulation. 😊 Creating a more accurate behaviour? 🤔

[@waylonbarrett3456]

I am building such a model

[@alexijohansen]

At 1:17:00 you discuss AI being damaging. Can AI at this stage really be blamed? It does not act by itself. These are actions taken by humans/corporations to extract wealth at scale with disregard for the common good. If damage is done right now the blame lies elsewhere no?

[@mayoai197]

I mean that's a good question. AI being damaging doesn't entail "it" should be blamed. If you train a model, it will produce what you trained it with and what you trained it for, with no regard for context. there are ways to minimize biases, but those have to be engineered into the models.

[@artandculture5262]

Their children are harmed if it results in spoiling of life on Earth. I think that’s the only causal law left in this bungled selfish destructive human project.

[@christopherdeslandes]

You guys really need to get Bernardo Kastrup on this podcast

[@J.D.1.]

And Benny Harvey

[@calebmorgan6939]

I thought Ramstead might be a poet with his "verses". You want "versus". Versus Versus, meaning “against, opposed to” or “in contrast to,” is often abbreviated to vs. in sports coverage and to v. in legal documents. Versus and its abbreviations are not italicized... I know you computer people learn many ungodly languages and so never feel that you have to learn your native tongue.

[@FutureGuy47]

As a non-English speaker, i had a ton of trouble understand even 50% of the intro. Overly complicated and convoluted language with a ton of buzzwords

[@SallyMangos]

"Bard....ELI5" lol

[@pedromarcelo4028]

I respectfully disagree. I find its rich language inspiring and refreshing in this era where people speak by approximation.

[@jcc4tube]

This sounds like somebody used chat GPT to write their philosophy thesis. WTF?

[@yoloswaginator]

This is because it presumes knowledge of quite some concepts that will sound rather empty statements without it. But the FEP is a real thing and increasingly applied in theoretical neuroscience, robotics, AI and systems dynamics. Usually people having this reaction either lack the adequate context or are in a very opposing state of mind.

[@juicespace6338]

It's exactly what happened.

[@juicespace6338]

These people can't communicate with anybody outside of their bubble and if that is not bad enough, they live so detached and unhealthy in any possible way, that their science is left utterly useless for dealing with their own life. It's idiocracy on an academic level.

[@doctorcrankyflaps1724]

Was the opening 5 minutes a pseuds corner parody? Stop trying to impress your mum. You are the only science and tech channel I struggle to understand.

[@HappySlapperKid]

Someone apply the free energy principle to the what is a woman debate so I can understand it in real terms