Time, Space and Causality
September 26, 2019 6:32 AM Subscribe
The Genius Neuroscientist Who Might Hold the Key to True AI - "Karl Friston's free energy principle might be the most all-encompassing idea since Charles Darwin's theory of natural selection. But to understand it, you need to peer inside the mind of Friston himself." (via)
Friston first became a heroic figure in academia for devising many of the most important tools that have made human brains legible to science. In 1990 he invented statistical parametric mapping, a computational technique that helps—as one neuroscientist put it—“squash and squish” brain images into a consistent shape so that researchers can do apples-to-apples comparisons of activity within different crania. Out of statistical parametric mapping came a corollary called voxel-based morphometry, an imaging technique that was used in one famous study to show that the rear side of the hippocampus of London taxi drivers grew as they learned “the knowledge.”also btw...
A study published in Science in 2011 used yet a third brain-imaging-analysis software invented by Friston—dynamic causal modeling—to determine if people with severe brain damage were minimally conscious or simply vegetative.
When Friston was inducted into the Royal Society of Fellows in 2006, the academy described his impact on studies of the brain as “revolutionary” and said that more than 90 percent of papers published in brain imaging used his methods. Two years ago, the Allen Institute for Artificial Intelligence, a research outfit led by AI pioneer Oren Etzioni, calculated that Friston is the world’s most frequently cited neuroscientist. He has an h-index—a metric used to measure the impact of a researcher’s publications—nearly twice the size of Albert Einstein’s. Last year Clarivate Analytics, which over more than two decades has successfully predicted 46 Nobel Prize winners in the sciences, ranked Friston among the three most likely winners in the physiology or medicine category...
For the past decade or so, Friston has devoted much of his time and effort to developing an idea he calls the free energy principle. (Friston refers to his neuroimaging research as a day job, the way a jazz musician might refer to his shift at the local public library.) With this idea, Friston believes he has identified nothing less than the organizing principle of all life, and all intelligence as well. “If you are alive,” he sets out to answer, “what sorts of behaviors must you show?”
[...]
First the bad news: The free energy principle is maddeningly difficult to understand. So difficult, in fact, that entire rooms of very, very smart people have tried and failed to grasp it. A Twitter account with 3,000 followers exists simply to mock its opacity, and nearly every person I spoke with about it, including researchers whose work depends on it, told me they didn’t fully comprehend it.
But often those same people hastened to add that the free energy principle, at its heart, tells a simple story and solves a basic puzzle. The second law of thermodynamics tells us that the universe tends toward entropy, toward dissolution; but living things fiercely resist it. We wake up every morning nearly the same person we were the day before, with clear separations between our cells and organs, and between us and the world without. How? Friston’s free energy principle says that all life, at every scale of organization—from single cells to the human brain, with its billions of neurons—is driven by the same universal imperative, which can be reduced to a mathematical function. To be alive, he says, is to act in ways that reduce the gulf between your expectations and your sensory inputs. Or, in Fristonian terms, it is to minimize free energy.[1]
- Facebook's Latest Purchase Gets Inside Users' Heads—Literally - "The social media company acquires CTRL-Labs, a 'brain-machine-interface' startup that lets users control devices by tapping signals off a wristband."[2,3]
- Inside DeepMind's epic mission to solve science's trickiest problem - "DeepMind's AI has beaten chess grandmasters and Go champions. But founder and CEO Demis Hassabis now has his sights set on bigger, real-world problems that could change lives. First up: protein folding."
- How to Build Artificial Intelligence We Can Trust - "Computer systems need to understand time, space and causality. Right now they don't."
- A critique of pure learning and what artificial neural networks can learn from animal brains - "Artificial neural networks (ANNs) have undergone a revolution, catalyzed by better supervised learning algorithms. However, in stark contrast to young animals (including humans), training such networks requires enormous numbers of labeled examples, leading to the belief that animals must rely instead mainly on unsupervised learning. Here we argue that most animal behavior is not the result of clever learning algorithms—supervised or unsupervised—but is encoded in the genome. Specifically, animals are born with highly structured brain connectivity, which enables them to learn very rapidly. Because the wiring diagram is far too complex to be specified explicitly in the genome, it must be compressed through a 'genomic bottleneck'. The genomic bottleneck suggests a path toward ANNs capable of rapid learning."
- AI That Evolves in the Wild - "Alison Gopnik said how nobody reads past the one sentence in Turing's 1950 paper. They never read past his 1936 paper to his 1939 'Systems of Logic Based on Ordinals', which is much more interesting. It's about non-deterministic computers, not the universal Turing machine but the second machine he wrote his thesis on in Princeton, which was the oracle machine—a non-deterministic machine. Already he realized by then that the deterministic machines were not that interesting. It was the non-deterministic machines that were interesting. Similarly, we talk about the von Neumann architecture, but von Neumann only has one patent, and that patent is for non-von Neumann architecture. It's for a neuromorphic computer that can do anything, and he explains that, because to get a patent you have to show what it can do. And nobody reads that patent."
- The Explosive Evolution of Consciousness - "Some philosophers identify consciousness with the complex, reflective, self-conscious experiences that we have when, say, we are sitting in an armchair and thinking about consciousness. As a result, they argue that even babies and animals aren't really conscious. At the other end of the spectrum, some philosophers have argued for 'pan-psychism', the idea that consciousness is everywhere, even in atoms. Recently, however, a number of biologists and philosophers have argued that consciousness was born from a specific event in our evolutionary history: the Cambrian explosion. A new book, 'The Evolution of the Sensitive Soul' by the Israeli biologists Simona Ginsburg and Eva Jablonka, makes an extended case for this idea."
In mathematics there’s this deep, old problem called the continuum hypothesis. We have an infinite number of different infinities, but they divide into only two kinds: countable infinities and uncountable infinities. My analogy for that is how at the end of a conference when you look for a t-shirt, there are only extra small t-shirts and extra large. There are no medium t-shirts. The continuum hypothesis—and there is a difference between being true and being provable—has not been proved. It says you will never find a medium-sized infinity. All the infinities belong to one side or the other.[4,5,6]
Two very interesting things are happening. What this means is that for any uncountable infinity, say, a line, there’s an infinite number of points between any two points, and then if you cut a piece of that line, it still has an infinite number of points. That, I believe, is analogous to organisms. All organisms do their computing with continuous function. In nature we use discrete functions for error correction in genetics, but all control systems in nature are analog. The smallest analog system has the full power of the continuum.
On the other side, you have the constructible infinities. What’s interesting there is that we’re trying to prove this by doing it. We’re doing our best to create a medium-sized infinity. So, you can say, "Well, it exists. We’ve made it." The current digital universe is growing by 30 trillion transistors per second, and that’s just on the hardware side, so we have this medium-sized infinity, but it still legally belongs to the countable infinities.[7]
For around 100 million years, from about 635 to 542 million years ago, the first large multicellular organisms emerged on Earth. Biologists call this period the Ediacaran Garden—a time when, around the globe, a rich variety of strange creatures spent their lives attached to the ocean floor, where they fed, reproduced and died without doing very much in between. There were a few tiny slugs and worms toward the end of this period, but most of the creatures, such as the flat, frond-like, quilted Dickinsonia, were unlike any plants or animals living today.
Then, quite suddenly by geological standards, most of these creatures disappeared. Between 530 and 520 million years ago, they were replaced by a remarkable proliferation of animals who lived quite differently. These animals started to move, to have brains and eyes, to seek out prey and avoid predators. Some of the creatures in the fossil record seem fantastic—like Anomolocaris, a three-foot-long insectlike predator, and Opabinia, with its five eyes and trunk-like proboscis ending in a grasping claw. But they included the ancestors of all current species of animals, from insects, crustaceans and mollusks to the earliest vertebrates, the creatures who eventually turned into us... [8,9,10,11]
In this context, I think it helps to replace the term "minimize free energy" with "minimize unbound energy", making it clear that there is a constant push and pull between order and chaos. (chaos being where everything is unbound - free - and order where everything is bound - notfree).
Not that this helps with answering the fundamental "why" question, of course.
posted by Old'n'Busted at 7:45 AM on September 26, 2019 [3 favorites]
Not that this helps with answering the fundamental "why" question, of course.
posted by Old'n'Busted at 7:45 AM on September 26, 2019 [3 favorites]
All of this makes perfect sense to me and there's nothing unusual about it at all.
I wonder if Friston has made the connection to Buddhism yet: all life is suffering (free energy).
Beings act to reduce suffering, and unenlightened ones do so in ways that minimise short-term suffering that simply benefits the self, while enlightened ones minimise long-term suffering and benefit more beings.
The math for this fitness test is left as an exercise for the reader.
posted by seanmpuckett at 7:47 AM on September 26, 2019 [7 favorites]
I wonder if Friston has made the connection to Buddhism yet: all life is suffering (free energy).
Beings act to reduce suffering, and unenlightened ones do so in ways that minimise short-term suffering that simply benefits the self, while enlightened ones minimise long-term suffering and benefit more beings.
The math for this fitness test is left as an exercise for the reader.
posted by seanmpuckett at 7:47 AM on September 26, 2019 [7 favorites]
Oh boy, Friston. I've got so many mixed feelings. And the lead article, which I found far too hagiographic, does not help. The fact that he is highly prolific and a phenomenally terrible writer and presenter gives people the impression that he's some genius who's single-handedly revolutionizing AI, whereas mostly what he is doing is mathematical philosophy that unifies a number of topics in AI.
To someone with training in fundamental AI research, the Free Energy principle is not especially complicated. As far as I understand, it essentially combines variational Bayesian inference with recent work on modelling action selection as an inference problem to paint a general picture of cognition with just a couple of equations. The devil however is in the vast number of details. And I can say as someone who works on these topics that the Free Energy principle in no way helps me do my work. If I want to better understand how to solve variational inference problems, or how to solve action as inference problems, I consult the literature on these subjects.
What I do appreciate in Friston's (and his students' and collaborators') work is that his vast knowledge of neuroscience allows him to draw a large number of connections between these mathematical concepts and the structure of the brain. And in my own dives through his work, and in my brief conversations with him, I don't get the feeling he's being immodest, and he's mostly clear that all he's doing is unifying existing work, as opposed to making fundamental contributions to the theory of AI. At the same time, I can't help but view him as partially culpable for the cult of personality that has formed around him and distorted the whole conversation.
I mean, this quote comes at the end of the Wired article:
“I often reflect on the jokes that people make about me—sometimes maliciously, sometimes very amusingly—that I can’t communicate. And I think: I didn’t write it for you. I wrote it for me.”
Thanks asshole, you also published it in Nature.
posted by Alex404 at 7:57 AM on September 26, 2019 [55 favorites]
To someone with training in fundamental AI research, the Free Energy principle is not especially complicated. As far as I understand, it essentially combines variational Bayesian inference with recent work on modelling action selection as an inference problem to paint a general picture of cognition with just a couple of equations. The devil however is in the vast number of details. And I can say as someone who works on these topics that the Free Energy principle in no way helps me do my work. If I want to better understand how to solve variational inference problems, or how to solve action as inference problems, I consult the literature on these subjects.
What I do appreciate in Friston's (and his students' and collaborators') work is that his vast knowledge of neuroscience allows him to draw a large number of connections between these mathematical concepts and the structure of the brain. And in my own dives through his work, and in my brief conversations with him, I don't get the feeling he's being immodest, and he's mostly clear that all he's doing is unifying existing work, as opposed to making fundamental contributions to the theory of AI. At the same time, I can't help but view him as partially culpable for the cult of personality that has formed around him and distorted the whole conversation.
I mean, this quote comes at the end of the Wired article:
“I often reflect on the jokes that people make about me—sometimes maliciously, sometimes very amusingly—that I can’t communicate. And I think: I didn’t write it for you. I wrote it for me.”
Thanks asshole, you also published it in Nature.
posted by Alex404 at 7:57 AM on September 26, 2019 [55 favorites]
seanmpuckett, I was going to make the same comment. I'm in no way current on any real AI or consciousness research. I've just been pondering about it for about 30 years now as one of the more interesting things to sit and think about.
I'm not sure I've ever heard of Friston but get the same sense of not really anything terribly new here. I'll have to go and read some of his stuff and see if I can grok it.
Great post kliuless!
posted by zengargoyle at 8:13 AM on September 26, 2019 [2 favorites]
I'm not sure I've ever heard of Friston but get the same sense of not really anything terribly new here. I'll have to go and read some of his stuff and see if I can grok it.
Great post kliuless!
posted by zengargoyle at 8:13 AM on September 26, 2019 [2 favorites]
To be alive, he says, is to act in ways that reduce the gulf between your expectations and your sensory inputs. Or, in Fristonian terms, it is to minimize free energy.
Life increasing entropy was a principle described by Schrödinger. How does this idea become Fristonian?
posted by They sucked his brains out! at 8:19 AM on September 26, 2019 [3 favorites]
Life increasing entropy was a principle described by Schrödinger. How does this idea become Fristonian?
posted by They sucked his brains out! at 8:19 AM on September 26, 2019 [3 favorites]
because he uses a term widely understood and proven again and again in the strongest possible way..... across billions of experiments and since the mid 1800s in all of physical and biological science to mean one thing .....to mean something else.
grrrr
posted by lalochezia at 8:28 AM on September 26, 2019 [16 favorites]
grrrr
posted by lalochezia at 8:28 AM on September 26, 2019 [16 favorites]
What I do appreciate in Friston's (and his students' and collaborators') work is that his vast knowledge of neuroscience allows him to draw a large number of connections between these mathematical concepts and the structure of the brain.
Except the connections they draw between free energy/active inference and actual brain structures are extremely tenuous. In some ways it is like the 'Bayesian brain' all over again: a mathematical theory that describes an optimal way to integrate information, the assumption that the brain must be doing something like that, and then correlating fMRI signals with abstract quantities in order to find, for example, the 'bayesian surprise' region of the brain.
Don't get me wrong, I like free energy, and I think it's a valuable counterpoint to neuroeconomics research that tries to assign 'reward value' to every goddamned aspect of neural processing, but it really hasn't been cashed out in terms of what's inside your head. Part of the problem is Friston and how opaque he can be, but another huge part is that empirical neuroscientists look at a few equations and get The Fear, which causes them to run back to ideas that make them feel safe.
posted by logicpunk at 8:33 AM on September 26, 2019 [7 favorites]
Except the connections they draw between free energy/active inference and actual brain structures are extremely tenuous. In some ways it is like the 'Bayesian brain' all over again: a mathematical theory that describes an optimal way to integrate information, the assumption that the brain must be doing something like that, and then correlating fMRI signals with abstract quantities in order to find, for example, the 'bayesian surprise' region of the brain.
Don't get me wrong, I like free energy, and I think it's a valuable counterpoint to neuroeconomics research that tries to assign 'reward value' to every goddamned aspect of neural processing, but it really hasn't been cashed out in terms of what's inside your head. Part of the problem is Friston and how opaque he can be, but another huge part is that empirical neuroscientists look at a few equations and get The Fear, which causes them to run back to ideas that make them feel safe.
posted by logicpunk at 8:33 AM on September 26, 2019 [7 favorites]
because he uses a term widely understood and proven again and again in the strongest possible way..... across billions of experiments and since the mid 1800s in all of physical and biological science to mean one thing .....to mean something else.
Which is the better of the two usual possibilities when somebody says "free energy" and contextualizes it by talking about thermodynamics so he's not really doing himself a favor with the naming here.
posted by atoxyl at 8:46 AM on September 26, 2019 [1 favorite]
Which is the better of the two usual possibilities when somebody says "free energy" and contextualizes it by talking about thermodynamics so he's not really doing himself a favor with the naming here.
posted by atoxyl at 8:46 AM on September 26, 2019 [1 favorite]
Facebook's Latest Purchase Gets Inside Users' Heads—Literally - "The social media company acquires CTRL-Labs, a 'brain-machine-interface' startup that lets users control devices by tapping signals off a wristband."[2,3]
CTRL-Labs has devised a wristband that uses electrodes to measure your forearm muscles' activity + accelerometer to measure arm's motion. Why is this being spun as "getting inside users heads"? There is no neural interface. It's outrageous that this misreporting is being perpetuated by so many sources. Also, some of the exec team are apparently real assholes so it's a shame some of their betters in this space are being overlooked, but I guess that's silicon valley for you.
posted by shaademaan at 8:55 AM on September 26, 2019 [3 favorites]
CTRL-Labs has devised a wristband that uses electrodes to measure your forearm muscles' activity + accelerometer to measure arm's motion. Why is this being spun as "getting inside users heads"? There is no neural interface. It's outrageous that this misreporting is being perpetuated by so many sources. Also, some of the exec team are apparently real assholes so it's a shame some of their betters in this space are being overlooked, but I guess that's silicon valley for you.
posted by shaademaan at 8:55 AM on September 26, 2019 [3 favorites]
consciousness seems cumulative, and emergent.
I've long been puzzled by the idea that consciousness seems emergent. I mean, it is literally everywhere we can possibly observe it. It seems entirely consistent with the data that this is because it is everywhere.
posted by howfar at 9:37 AM on September 26, 2019
I've long been puzzled by the idea that consciousness seems emergent. I mean, it is literally everywhere we can possibly observe it. It seems entirely consistent with the data that this is because it is everywhere.
posted by howfar at 9:37 AM on September 26, 2019
Oh god, spare us from "great man" hagiographies of scientific progress. Friston is smart and all, but come on. Comparing his h-index to Einstein's? That's idiotic, comparing h-indices between fields is meaningless enough, let alone the dramatically different publishing cultures between today and a century ago.
I've read Friston's papers. He's a terrible writer and chooses mathematical notation that seems almost deliberately obfuscatory. The actual content of his free energy principle ideas are not that confusing, he's just incapable of or uninterested in communicating them clearly even to people who do have the mathematical background to evaluate his arguments.
At its core, his free energy ideas seem to equivocate between a sort of thermodynamic tautology when talking about things like artificial and biological neural networks on the one hand and incoherent nonsense when talking about organisms, evolution, and behavior on the other. To the extent that they are true, they are not novel, and to the extent that they are novel, they are not true. That's a bit of an overstatement but I'm prepared to argue it.
posted by biogeo at 9:40 AM on September 26, 2019 [22 favorites]
I've read Friston's papers. He's a terrible writer and chooses mathematical notation that seems almost deliberately obfuscatory. The actual content of his free energy principle ideas are not that confusing, he's just incapable of or uninterested in communicating them clearly even to people who do have the mathematical background to evaluate his arguments.
At its core, his free energy ideas seem to equivocate between a sort of thermodynamic tautology when talking about things like artificial and biological neural networks on the one hand and incoherent nonsense when talking about organisms, evolution, and behavior on the other. To the extent that they are true, they are not novel, and to the extent that they are novel, they are not true. That's a bit of an overstatement but I'm prepared to argue it.
posted by biogeo at 9:40 AM on September 26, 2019 [22 favorites]
I haven't read any these articles, but the excerpt about infinities in mathematics is bullshit on the level of New Age quantum woo. For example, it is very easy to find examples of countably infinite sets with "an infinite number of points between any two points", the most famous example being the rational numbers. It is likewise easy to find examples of uncountable sets with two points with no points between them, for example, if you take all the real numbers greater than or equal to 1, or less than or equal to 0.
posted by J.K. Seazer at 9:55 AM on September 26, 2019 [14 favorites]
posted by J.K. Seazer at 9:55 AM on September 26, 2019 [14 favorites]
Friston's approach seems like another way to say that humans reduce risk for self-preservation, which is its own reward obviously, but we also increase risk by seeking reward when considering the game is about starvation. We perhaps forget that predators and prey played a major role in biological development, and how our human response with tools is ideally suited to both offense and defense (and this sets the stage for our cultural and mental problems too). For example, we likely learned to throw rocks because it confused adversaries as to our hidden position, or drove prey into traps. We made tall spears to defend against pouncing ambush from above, and to attack on the run. We likely donned skins because it both camouflaged us and kept is warm while hunting. We built shelters to hide in ambush, and to defend against attack. Our imaginative brains likely developed the way they did, in contrast to most animals, because of an ability to make sense of minimally partial information as either danger or reward approached, such as seeing only a glimpse of movement through dense leaves, hearing only a twig snap, then determining if it is friend or foe, prey or predator. We conflated offense and defense and our human advantage was due to moderating both risk and reward. Our affinity with cats and dogs is no accident if we consider that they also consciously moderate both attack and flight, and therefore understand us through our mutual freedom in nature.
posted by Brian B. at 10:21 AM on September 26, 2019 [1 favorite]
posted by Brian B. at 10:21 AM on September 26, 2019 [1 favorite]
As far as AI, he has come up with an algorithm. It's clever, and a refreshing redirection to those for whom the answer to everything is more sorting and pattern-matching. It's self-calibration, which isn't (imho) that much of a breakthrough, but which is an interesting approach.
The free energy and Markov blankets are pure hoo-ha.
posted by sensate at 1:11 PM on September 26, 2019
The free energy and Markov blankets are pure hoo-ha.
posted by sensate at 1:11 PM on September 26, 2019
So is this roughly the same as that time that Wolfram wrote that huge book on cellular automata claiming them to be the grand unified solution to everything, and everyone just used the book as a doorstop?
posted by scruss at 1:29 PM on September 26, 2019 [6 favorites]
posted by scruss at 1:29 PM on September 26, 2019 [6 favorites]
I was ready to angel shit all over this due to Wired's *yawn* hagiographic, narrative style...*yawn*
But now I see wiki and will have to dig a bit more. Fits into some of my interests.
One idea I have (though not directly related) is how we perceive patterns in random noise. If we are all quantum fluctuations, what if the order we perceive isn't really order but illusion. When we see things around we're really perceiving static, but - I guess this gets to a platonic view - we have the preformed "Ideal" already bound to consciousness, and when these patterns in the chaos emerge, we see them as objects. Picture pareidolia, but instead of patterns of wood looking like faces it's the static noise of the universe looking like... well. whatever it is we see/interact with.
I don't think that really is a viable stance, though perhaps it could be fit into this concept...
Another concept that's more theological in nature is one where entropy and sin are tied together. Death is of course the result of "sin" but "sin" is a "choice" between a given order and a different order. Choice can only exist within systems that offer dualities or options.
There's some sort of theodicy where god is bound by entropy, that is to say, in order for us to be "free" to make a so-called "sinful" choice, entropy must necessarily exist. Otherwise we're just in static/block time.
Once entropy comes into play the concept of choice, sin and death all result.
I feel like this could lay the foundation of a theology of sin as entropy and will.
Shopenhauer and Mainlander, eat your hearts out.
posted by symbioid at 1:46 PM on September 26, 2019 [1 favorite]
But now I see wiki and will have to dig a bit more. Fits into some of my interests.
One idea I have (though not directly related) is how we perceive patterns in random noise. If we are all quantum fluctuations, what if the order we perceive isn't really order but illusion. When we see things around we're really perceiving static, but - I guess this gets to a platonic view - we have the preformed "Ideal" already bound to consciousness, and when these patterns in the chaos emerge, we see them as objects. Picture pareidolia, but instead of patterns of wood looking like faces it's the static noise of the universe looking like... well. whatever it is we see/interact with.
I don't think that really is a viable stance, though perhaps it could be fit into this concept...
Another concept that's more theological in nature is one where entropy and sin are tied together. Death is of course the result of "sin" but "sin" is a "choice" between a given order and a different order. Choice can only exist within systems that offer dualities or options.
There's some sort of theodicy where god is bound by entropy, that is to say, in order for us to be "free" to make a so-called "sinful" choice, entropy must necessarily exist. Otherwise we're just in static/block time.
Once entropy comes into play the concept of choice, sin and death all result.
I feel like this could lay the foundation of a theology of sin as entropy and will.
Shopenhauer and Mainlander, eat your hearts out.
posted by symbioid at 1:46 PM on September 26, 2019 [1 favorite]
Also - looking forward to "New Kind of Science 2: Electric Free-Enerloo"
posted by symbioid at 1:48 PM on September 26, 2019 [1 favorite]
posted by symbioid at 1:48 PM on September 26, 2019 [1 favorite]
My understanding of this general idea is as follows:
Bayes' rule gives you a sort of normative ideal of what your beliefs "should" be, given your priors and after observing the current state of the world. This is not realizable in practice, typically, so you have to approximate it -- essentially by tuning some knobs to make the beliefs your robot's brain can actually represent as close as possible to what they should be.
The standard way to do this is called variational inference, and proceeds by maximizing a quantity called the "evidence lower bound" (ELBO for short); or, alternately, you can flip a negative sign and call it "minimizing free energy", which sounds more profound. There are various schemes to tune the knobs to push this quantity in the right direction, which takes your beliefs closer to the ideal belief even if they don't quite get all the way there.
Additionally, some people realized that when your agent doesn't just observe but actually acts, you can still squeeze things into this framework by just hooking some of the same knobs up to the robot's arms and legs, such that as you tune the knobs by the same variational inference procedure, the robot will both change its beliefs to accord with the world, but also change the position of its arms and legs to bring the world closer to its beliefs.
Would be interested to know to what extent this is right or if I'm missing some additional stuff.
posted by vogon_poet at 7:26 PM on September 26, 2019 [2 favorites]
Bayes' rule gives you a sort of normative ideal of what your beliefs "should" be, given your priors and after observing the current state of the world. This is not realizable in practice, typically, so you have to approximate it -- essentially by tuning some knobs to make the beliefs your robot's brain can actually represent as close as possible to what they should be.
The standard way to do this is called variational inference, and proceeds by maximizing a quantity called the "evidence lower bound" (ELBO for short); or, alternately, you can flip a negative sign and call it "minimizing free energy", which sounds more profound. There are various schemes to tune the knobs to push this quantity in the right direction, which takes your beliefs closer to the ideal belief even if they don't quite get all the way there.
Additionally, some people realized that when your agent doesn't just observe but actually acts, you can still squeeze things into this framework by just hooking some of the same knobs up to the robot's arms and legs, such that as you tune the knobs by the same variational inference procedure, the robot will both change its beliefs to accord with the world, but also change the position of its arms and legs to bring the world closer to its beliefs.
Would be interested to know to what extent this is right or if I'm missing some additional stuff.
posted by vogon_poet at 7:26 PM on September 26, 2019 [2 favorites]
It seems entirely consistent with the data that this is because it is everywhere.
Some modern thinkers say 'Could be'.
(Article title: “Ultimate Mystery of the Universe” –Human Consciousness: ‘We’re Like Neanderthals Trying to Understand Astronomy’)
Another capable thinker who thought on these lines: quantum physicist John Archibald Wheeler. "Wheeler originated the notion of a “participatory,” conscious universe, a cosmos in which all of us are embedded as co-creators..."
Deep divers hint: Also worth note: Michael Polanyi ( (IIRC in 'Tacit Dimension') pointed out that the organization of matter by lifeforms is anti-entropic. (See: enthalpy. MP was a Royal Society member; his son is a Nobel-winning chemist)
posted by Twang at 8:05 PM on September 26, 2019 [2 favorites]
Some modern thinkers say 'Could be'.
(Article title: “Ultimate Mystery of the Universe” –Human Consciousness: ‘We’re Like Neanderthals Trying to Understand Astronomy’)
Another capable thinker who thought on these lines: quantum physicist John Archibald Wheeler. "Wheeler originated the notion of a “participatory,” conscious universe, a cosmos in which all of us are embedded as co-creators..."
Deep divers hint: Also worth note: Michael Polanyi ( (IIRC in 'Tacit Dimension') pointed out that the organization of matter by lifeforms is anti-entropic. (See: enthalpy. MP was a Royal Society member; his son is a Nobel-winning chemist)
posted by Twang at 8:05 PM on September 26, 2019 [2 favorites]
It's really a shame that the word "pseudoscience" has been used to denote paranormal and alternative medicine type beliefs......it's perhaps more aptly used for speculative metaphysics with some equations and physics-y concepts glazed over it.
posted by thelonius at 8:12 PM on September 26, 2019 [1 favorite]
posted by thelonius at 8:12 PM on September 26, 2019 [1 favorite]
So can anyone point to a nice 'explain it to me like I'm five' account of the free energy principle please? I am deeply suspicious of anything that can't be explained clearly.
posted by leibniz at 9:52 PM on September 26, 2019
posted by leibniz at 9:52 PM on September 26, 2019
Anyone who has published an average of 50 academic articles a year for the last 20 years must have produced their own fair share of angel shit.
posted by louigi at 7:51 PM on September 27, 2019
posted by louigi at 7:51 PM on September 27, 2019
So can anyone point to a nice 'explain it to me like I'm five' account of the free energy principle please?
It is a carefully calculated ratio of midichlorians to quatloos.
posted by They sucked his brains out! at 8:03 AM on September 28, 2019
It is a carefully calculated ratio of midichlorians to quatloos.
posted by They sucked his brains out! at 8:03 AM on September 28, 2019
That last, "Cambrian Explosion" link is awful.
Dickinsonia has been shown to have used cholestrosteroids - it was related to present day animals.
Also from the Ediacaran was Cloudina - a very common shell-boring predator.
I could go on.
Also, there is no Royal Society of Fellows - it's the Royal Society.
posted by thatwhichfalls at 8:17 AM on September 28, 2019
Dickinsonia has been shown to have used cholestrosteroids - it was related to present day animals.
Also from the Ediacaran was Cloudina - a very common shell-boring predator.
I could go on.
Also, there is no Royal Society of Fellows - it's the Royal Society.
posted by thatwhichfalls at 8:17 AM on September 28, 2019
So can anyone point to a nice 'explain it to me like I'm five' account of the free energy principle please?
So, You are walking around on an infinite plain leaving a line behind you as you walk around. You bump into somebody else and immediately fall in love with them, the Other. But the two of you part after this random encounter. You give chase and try to meet the Other again. But the Other keeps moving. Eventually You manage to bump into the Other again.
Both You and the Other are leaving trails. If you were to take a colored pencil and fill in the area between the trails since that first time you met and now.... That colored in area is free-energy.
The better You can predict the Other leads towards an ever decreasing area of free-energy.
Now take this into the Other is Your perception of the world and Your love is survival.
The ELI5 starts getting hairy here, because the same principle of reducing that colored area goes all multi-dimensional and can be applied to almost everything.
posted by zengargoyle at 4:10 PM on September 28, 2019 [1 favorite]
So, You are walking around on an infinite plain leaving a line behind you as you walk around. You bump into somebody else and immediately fall in love with them, the Other. But the two of you part after this random encounter. You give chase and try to meet the Other again. But the Other keeps moving. Eventually You manage to bump into the Other again.
Both You and the Other are leaving trails. If you were to take a colored pencil and fill in the area between the trails since that first time you met and now.... That colored in area is free-energy.
The better You can predict the Other leads towards an ever decreasing area of free-energy.
Now take this into the Other is Your perception of the world and Your love is survival.
The ELI5 starts getting hairy here, because the same principle of reducing that colored area goes all multi-dimensional and can be applied to almost everything.
posted by zengargoyle at 4:10 PM on September 28, 2019 [1 favorite]
As for myself... the real issue is usually impedance mismatch that sucks up free-energy. AT&T Archives: Similiarities of Wave Behavior (Bonus Edition). It's the tuning of information transfer across the various fields and interactions that make up the fabric. :P
posted by zengargoyle at 4:22 PM on September 28, 2019
posted by zengargoyle at 4:22 PM on September 28, 2019
An AI Pioneer Wants His Algorithms to Understand the 'Why' - "Deep learning is good at finding patterns in reams of data, but can't explain how they're connected. Turing Award winner Yoshua Bengio wants to change that."
At his research lab, Bengio is working on a version of deep learning capable of recognizing simple cause-and-effect relationships. He and colleagues recently posted a research paper outlining the approach. They used a dataset that maps causal relationships between real-world phenomena, such as smoking and lung cancer, in terms of probabilities. They also generated synthetic datasets of causal relationships.posted by kliuless at 10:23 AM on October 13, 2019
The algorithm in the paper essentially forms a hypothesis about which variables are causally related, and then tests how changes to different variables fit the theory. The fact that smoking is not only related to cancer but actually causes it, for instance, should still be apparent even if cancer is correlated with other factors, such as hospital visits...
Judea Pearl, who won the Turing Award in 2011 for his work on causal reasoning, says he is impressed with Bengio’s ideas, although he has not studied them closely. A recent book co-authored by Pearl, The Book of Why: The New Science of Cause and Effect, makes the case that AI will be fundamentally limited without some sort of causal reasoning ability.[1,2]
« Older Indelicate | City Of Death at 40 Newer »
This thread has been archived and is closed to new comments
It would just be a delightfully typical thing in this weird universe to end up with an true general AI on accident, and when we ask it how it works, that it doesn't know, either, any better than we do about our own consciousness.
I have feelings, I have thoughts, I can identify them and "weigh" them and contemplate them. I can measure them in my brain with fMRI machines, and probe them by exploring my response to different stimuli, and alter them at a coarse level with chemical inputs into my system, but where do *I* come from? Hell if I know. Could be that true AI, should it occur any time soon, will feel the same, perhaps just a lot faster.
posted by tclark at 7:31 AM on September 26, 2019 [6 favorites]