Slashdot Mirror
Physicist Proposes New Way To Think About Intelligence
An anonymous reader writes "A single equation grounded in basic physics principles could describe intelligence and stimulate new insights in fields as diverse as finance and robotics, according to new research, reports Inside Science. Recent work in cosmology has suggested that universes that produce more entropy (or disorder) over their lifetimes tend to have more favorable properties for the existence of intelligent beings such as ourselves. A new study (pdf) in the journal Physical Review Letters led by Harvard and MIT physicist Alex Wissner-Gross suggests that this tentative connection between entropy production and intelligence may in fact go far deeper. In the new study, Dr. Wissner-Gross shows that remarkably sophisticated human-like "cognitive" behaviors such as upright walking, tool use, and even social cooperation (video) spontaneously result from a newly identified thermodynamic process that maximizes entropy production over periods of time much shorter than universe lifetimes, suggesting a potential cosmology-inspired path towards general artificial intelligence."
How was the weather?
http://xkcd.com/793/
Interesting idea. http://techcrunch.com/2013/04/14/nes-robot/
That guy took basically a random generator and 'picked' good results to build on. However the input is basically chaos.
Humans are soulless meat computers. Intelligence is just a byproduct of electrical signals in a random mesh of electrons. Everything that can be discovered has already been discovered.
... I burn stuff. Now I can feel smarter about it. Win!
Intelligence was invented by man, as a way to make them seem better then other animals in the world.
Then we further classified it down so we can rank people.
So it isn't surprising if we want to find intelligent life outside of earth, then we need to change the rules again, as well we need to change the rules of what intelligence is by the fact we have created technology that emulates or exceeds us in many areas we use to classify intelligence.
Intelligence is a man made measurement, I expect it will always be in flux. However you shouldn't dismiss or automatically accept as good ideas just because someone number that was granted by a fluctuating scale.
If something is so important that you feel the need to post it on the internet... It probably isn't that important.
"much shorter than universe lifetimes"
Good, because my attempt that took much longer than universe lifetimes was going a bit too slow.
http://www.alexwg.org/publications/PhysRevLett_110-168702.pdf
This looks eerily like a physicist who has just opened a biology textbook and is now restating the idea that 'intelligence' is the product of an evolutionary selection process because it's a uniquely powerful solution to the class of problems that certain ecological niches pose and is now attempting to add equations....
Is there something that I'm missing, aside from the 'being alive means grabbing enough energy to keep your entropy below background levels' and the 'we suspect biological intelligence of having evolved because it provides a fitness advantage in certain ecological niches' elements?
Intelligence, the ability to delve into the past and reach into the future, in order to craft the present and manipulate the probability of eventualities. The greater the ability the greater the intellect, the power of choice.
Chaos - everything, everywhere, everywhen
It appears to me that the algorithm is trying to maintain entropy or disorder, or at least keep open as many pathways to various states of entropy as possible. In the physics simulations, such as balancing and using tools, this essentially means that it is simply trying to maximize potential energy (in the form of stored energy due to gravity or repulsive fields - gravity in the balancing examples, and repulsive fields in the "tools" example).
While this can be construed as "intelligence" in these very specific cases, I don't think it is nearly as generalized or multipurpose as the author makes it out to be.
Better known as 318230.
I am always amused when I see these posts on intelligence as if it is an singular thing. There are many types of intelligence from scientific (i.e. medicine, mathematics) to creative (i.e. literature, music, sculpting) to social (i.e. business, politics) and athletic (i.e. sports, recreation). No one person can be good at all of these types of intelligence, but collectively we can. Or put another way. We all saw Watson win at chess and Jeopardy, I don’t think would do so well playing Texas hold’em against some tournament champions.
"Allowing the rod to fall would drastically reduce the number of remaining future histories, or, in other words, lower the entropy of the cart-and-rod system."
increase the entropy.
"Keeping the rod upright maximizes the entropy. It maintains all future histories that can begin from that state, including those that require the cart to let the rod fall."
minimizes the entropy.
You can tell this is a physicist's paper. It lacks spherical cows, but only because the toy models were set up in 2D. So, instead, we get a crow, chimpanzee, or elephant approximated by circular disks.
I skimmed through the article. The idea as entropy as a driving factor for intelligence is certainly novel (to my knowledge), I haven't even met it in science fiction stories ! But, while interesting in his small test set, I really wonder about the author's extrapolations. Intelligence and free will seem much more complex than a thermodynamic optimisation. Perhaps, just perhaps that his idea is part of the very first steps from matter towards life and intelligence but much more research needs to be done.
For a learning process there needs to be a machinery though. Brain and body. Or a computer. I.e. something that can possibly balance a stick or walk on two legs, either real or simulated. So there is no intelligence out of thin air. The described max. entropy principle seems to apply to learning process, though (children do break things apart, no?).
Organisms capable of learning develop because of evolution benefit compared to those that have no intelligence and do not need to learn.
Just like any other physical trait.
Try nonexistent. The authors have ZERO background on intelligence. Basically they are proposing intelligent design....
Not pictured in the article: cognitive scientists the world over laughing at him over a cup of coffee.
The universe developed intelligence as a way of making entropy wind down faster ... which will destroy all intelligence ... which is a tragedy because the winding down was necessary to create us ... and the universe WANTED TO SEE US SUFFER.
this is a load of horse shit invented just so nerds can do something trivial and think the are doing something monumental and amazing instead
I agree with your statement. However, as I wrote, it could (maybe!) be that thermodynamics work in a way that improves the conditions to go from dead matter to life and intelligence; compared to pure chaos.
I'm maintaining the maximum number of possible outcomes for the day, in harmony with the laws of nature. :)
..don't panic
The paper has nothing to do with the "design" and/or evolution of intelligence. It proposes a general mechanism by which "intelligent" brains may be able to "figure out" how to perform a wide variety of tasks (by maximizing future available states based on a simplified internal world model). Plain old evolutionary selective pressures would favor critters with brains good at carrying out this type of cognition.
Why don't other animals have it? The answer is it just can't compete, in an evolutionary sense, with other phenotypes (like instinct). Or, put more simply: "He who hesitates is lunch". Evolution can certainly be modeled as a system that maintains entropy, but I just don't see this abstraction being all that useful in explaining intelligence.
Since when is Sheldon Cooper allowed to post online?
Here's a review of this paper by a researcher who actually works in the field of AI and cognitive psychology:
Interdisciplinitis: Do entropic forces cause adaptive behavior?
Few choice quotes:
and after he explains what the paper's about and how utterly empty it is, he offers some advice to authors:
And here's a relevant SMBC:
http://www.smbc-comics.com/?id=2556
He's just repeating the old adage - best demonstrated by Gary Larson - that nature abhors a vacuum. :)
This is just a restatement of the second law of thermodynamics with some thoughts about how that constrains life/evolution. Nothing to see here folks, move along.
The comments referring to Gary Larson catch the biased nature of the thoughts. Nothing to see here folks, move along.
Lots of entropy here, more than most! Does that mean that I am super intelligent ?
I don't have the background to judge the novelty of this approach or not. But the quote from the Sante Fe institute fellow would imply that the chaos/complexity folks find it interesting.
This paper formalizes something that until now we viewed as common sense.
Intelligent agent:
- is creative
- thinks outside of the box
- comes up with solutions that are unexpected
- is novel
- surprises you with their behavior
Dumb agent:
- follows rigid rules
- thinks only within rigid set rule-set (thinks inside the box)
- comes up with obvious solutions
- is repetitive
- is very predictable and repetitive in their behavior
Intelligent agent traits involve more entropy than the dumb agent. If you have two adaptive paths, one that will maximize the entropy and a second that doesn't, the one that will maximize entropy will produce a more intelligent agent (maybe not in the next generation, but eventually). High entropy = open mind. Low entropy = closed mind.
This has interesting implications for every day life. If you choose a low entropy career (assembly line worker), your intelligence will suffer, compared to a choice of a high entropy career (researcher).
Low entropy environments are predictable, high entropy environments are more unpredictable. I would say that unpredictability is a good way to judge the level of intelligence. Take 2 agents. The one who can successfully predict the actions of the other is more intelligent. I can predict most of the actions of my 5 yo daughter (for now), that makes me more intelligent. She can not predict my actions to the same degree (yet).
Deeper philosophical question is the level of unpredictability. If someone is so unpredictable that their actions are crazy ... are they really crazy? Or are they simply a misunderstood genius? Maybe their ideas will be understood only by later generations? What is the difference between complexity and randomness? After all I can make an agent that has very high entropy (make it jump around randomly) and yet it will not be as intelligent as an agent who is complex (but doesn't have the same level of entropy). High entropy isn't the only thing that makes someone intelligent. High entropy has to be bounded by adaptive behavior, otherwise it's useless. A mad scientist may have very high entropy (discovers new form of energy generation) but very poor adaptive behavior (is jobless/homeless because of poor hygiene and social skills). Some things like tying shoes and brushing teeth have to be somewhat low entropy (repetitive) just to satisfy the criteria of basic day to day survival.
.
We live in a society defined by division of labor. The physicist figured that out, as have many video game addicts.
When P-man walks he gets to think about his theories more, he gets necessary exercise, and he gets his chore done in about the same amount of time. And he simply isn't interested in most of the stuff that we rush around doing. He doesn't particularly want or need a cell phone, and for sure not a tablet. TV is low bandwidth, high noise -- easily filtered out with the convenient OFF button. Shopping is a once-a-week thing that someone else does...no need to duplicate effort. Same with laundry, with those two large machines doing most of the work.
It is called the simple life. And it kind of rocks.
I come here for the love
In any case both require clear definition, but it would appear that this paper applies to "life" rather than the more restrictive "intelligence".
IMO "intelligence" is primarily the ability to refer to things outside the here and now, the property that linguists call displacement. See Derek Bickerton's "Adam's Tongue" for details.
Actually, the opposite: "intelligence" functions by seeking to maximize entropy.
Don't you mean that intelligence "functions by seeking to maximize the entropic gradient"?
Set your phasers on "funky"!
the article is self contradictory — it says "It actually self-determines what its own objective is," said Wissner-Gross. "This [artificial intelligence] does not require the explicit specification of a goal".
this is not true, because it then goes on to say, "trying to capture as many future histories as possible".
so there IS a goal — it maximizes the number of future states — exactly the same way a negaMax search can maximize the mobility paramater in a chess engine search.
in other words, this is a lot of hype — defining intelligence as maximization of finding states of less entropy (i.e. maximal future states), and running a classic negaSearch on that basis is what is going on here.
its a novel way to go about things, but redefining the terms doesnt actually make anything new in the sensation way this article claims.
The premise of the claim is that procrastination is the ultimate goal of intelligence, with procrastination defined as keeping open the widest range of possible options by avoiding all actions that would decisively limit that range.
This would seem, even on the surface, to ignore the many situations where intelligent life must take the narrow path, sacrificing procrastination to the pursuit of a single goal. Once through a narrow path we may find a wide vista of prospects again before us. But without taking such narrow paths at significant times, by always hesitating at the crossroads for as long as possible, we may find ourselves with Robert Johnson, sinking down.
Also, the claim that the natural goal of choice is to maximize future choice is entirely circular. Like saying the goal of walking is to maximize future walking, the goal of eating to maximize future eating, there's something to it, but it's not quite true. Also, a great deal of research shows that people strive to avoid choice, for the most part.
"with their freedom lost all virtue lose" - Milton
Compare to Terrence Deacon's Incomplete Nature, which: "meticulously traces the emergence of this special causal capacity from simple thermodynamics to self-organizing dynamics to living and mental dynamics" (Amazon).
(Deacon's book is good, though has been criticized as drawing heavily from prior work: "This work has attracted controversy, as reviewers[2] have suggested that many of the ideas in it were first published by Alicia Juarrero in Dynamics of Action (1999, MIT Press) and by Evan Thompson in Mind in Life (2007, Belknap Press and Harvard University Press) yet these works were not cited or referenced by Deacon." (Wikipedia))
Or compare to Stuart Kauffman's Origins of Order, which Deacon cites (and it seems the two are in communication). Kauffman's notion is that there are implicit geometries to energetic forms which in the situation of excess total energy can locally channel a system towards structure and shape that bias, and perhaps belie, the notions of random variation and natural selection being the primary drivers for the creation of structures in living beings.
Neat to see this coming to the east coast/MIT.
First of all, the paper is steeped in jargon. Phrases such as (2nd para) "characterized by the formalism of" instead of "described by" obfuscate the meaning and confuse the reader.
Count the number of uses of the verb "to be" (is, are, "to be", were, &c). It's everywhere! Nothing runs or changes, everything "is running" or "is changing". Passive voice removes the actor in a paper describing - largely - actions.
Useless words and phrases litter the landscape, such as "To better understand", "for concreteness", "as a simple means", "to the best of our knowledge". Cushiony adverb goodness pads the document to the required length. "Explicitly propose" (instead of just "propose"), "dynamically revealed information" (as opposed to that other kind of revealed information), "remarkably sophisticated behaviours" (as opposed to the pedestrian kind, I guess).
This paper is all kinds of awesome! It should be the touchstone for Stanford's "Writing in the Sciences" online course.
My first impression (it's really dense!) is that the author conflates maximum entropy with the agent goal. It's not always the case that the goal is the maximally entropic state. This is likely true when actors must cooperate (as shown in the paper), but when actors compete the individual goal may not be maximum entropy.
For example, consider competing for mates. Instead of choosing mates based on competitive merit, should an individual limit their offspring in order to give everyone in society a chance to reproduce? The non-cooperative goal isn't for maximal entropy.
It also appears to describe intelligence as an evolutionary process seeking a function minimum over multiple-parameter phase space. While this might solve physical puzzles such as walking or throwing a ball, I'm not convinced that chess can be solved in this way. The search space is too big for an evolutionary solution.
Still, I may be misapprehending the point of the article (it's really dense!). Read the paper and make your own assessment.
Perhaps this could be looked at in a completely different way: That it reinforces the hypothesis that our universe is actually an AI-driven simulation.
Intelligence, as postulated for betas, is literally the 'ghost in the machine'. Here's the problem. Semantics CANNOT arise from syntax.
Betas are currently told one of two lies. Either 'intelligence' (by which we mean a conscious entity able to process 'meaning') is a wibbly-wobbly magical outcome of quantum mechanics and the 'uncertainty' principle, or it is an outcome of 'system complexity'.
Here's the problem. Only maths can be the fundamental tool of which scientific models are based. All 'maths' runs on a 'Turing Complete' computer by definition. A 'Turing Complete' computer NEVER EVER requires any concept of semantics to process the state transitions that make up the calculations by which any possible expression of maths requires.
Put another way, our clockwork universe can never produce intelligence, and never needs such a concept to 'work'. No understanding of the origin of intelligence can arise from the study of the clockwork universe (which is actually really the study of state transitions on a Turing Complete computer).
Let me make this even easier to comprehend. Statistics is a field of maths. It tells us, for example, the likelihood of throwing all sixes with a given number of dice. At some point, as the number of dice (n) increases, such an outcome becomes vanishingly unlikely, even given the 'age' of the Universe. You are aware that you could place any number of dice in the six uppermost position. What most of you fail to comprehend is that such an act is no different from rolling the dice according to the rules of the clockwork universe. You are therefore able to create an outcome statistically unlikely to the point of being completely impossible. You mind, by definition, does NOT run on a Turing Complete computer.
If you follow this line of argument, you will naturally conclude there is some 'meta' maths or 'super' maths that cannot run on a Turing Complete computer. You are WRONG. Turing and Gödel gave us the final pieces to understand maths at the completely abstract level, and eliminated for all time the idea of higher levels of 'meta' maths. By doing so, they proved once and for all that consciousness does not, will not and cannot arise from the mechanisms of a clockwork universe. Maths cannot explain the working of the conscious mind. And this means neither can science. The fact that you can apply statistical analysis to the decisions made by groups of Humans does not change this fact.
Do make this clear, true randomness cannot be modelled on a Turing Complete computer (and such a thing would break the causal concept that underpins physics), but obviously you can apply statistical models to the output of though imagined events of true randomness. That you can conceive of applying maths in the form of statistical analysis to the output of a mythical truly random process neither explains such a process nor brings it into creation.
Our minds exist, so true randomness can exist too, but both will exist outside the realm of what we call 'science' (which is actually state transitions on a Turing Complete computer, by definition). Alpha scientists accept there is the clockwork universe of science and maths, and then there is, as a fully separate concept, conscious intelligence creating and processing meaning. To an alpha, it is self-evident that concepts of 'self' are higher than concepts arising from observation of the clockwork universe. Betas, on the other hand, are trained that the concept of 'self' is LOWER than the concepts arising from scientific observations. Paradoxically, evil alphas also ensure betas then get a contradictory message from the indoctrination given by whichever organised religion they are forced to associate with.
Anyone who feels like a single entity needs to consider that much of what we are is a distributed system. Very small parts have tasks, such as shown on this movie: https://www.youtube.com/watch?v=OWUmXx5V_wE . THis is a white blood cell (Neutrophil type) chasing down a little bad thing.. and gobbling it up. Sure, this can be attributed to Chemotaxis, but, the Neutrophil has intent. This seems unbeleivable. To get even crazier.. look into molecular motors.https://www.youtube.com/watch?v=-7AQVbrmzFw ! Each of us is an amalgam, a colony of amazing things all going on at one time! The smart part of us - the brain, is the same part that does stupid things like smoking that takes its toll on so many little parts of us.. but.. the brain decides that its in its own best interest to buy the WHOLE CARTON because its cheaper that way. Oh, the humanities!
Time for a new Political party in the US (or two!) One is off the rails Other cant pony up a leader.
If so, that's all the hard scientific proof we need.
Hafta digest the pdf before I can say more than that. Ideas like this are not exactly new, though, at least to a Pchem zombie.
I laughed out loud at the 'beef tensors' !
Non-Linux Penguins ?
Here is my summary of the paper, from the authors website
Entropy is usually defined as a function of the macroscopic state of a system at a given time. If we assume that the system evolves so as to move in the direction of maximum entropy at any time, then this defines some dynamics. What the authors propose is a foreward looking dynamic where the system moves in the direction that maximizes entropy and some future point. This automatically builds in forward looking (i.e. intelligent) behaviour into the system.
They are able to demonstrate various kinds of "intelligent" behavior arising from this simple and general heuristic. The comparisons with human and animal behavior are in my opinion ridiculous. The value of the paper is in demonstrating that a wide variety of forward looking optimizations can be accomplished with a simple rule.
Another piece of the Skynet puzzle falls into place.
One of my favorite comics ever. Seriously though, I like when physicists step out of their area and play or dabble in other areas. Probably refreshing for them too. I think there isn't enough interdisciplinary projects out there for scientists.
You can make the examples work if you modify the problem. For instance, define that the box is cold, the room is hot and the food is hot. Then, putting the food in the box increases entropy. This is in fact analogous to eating, even in physical terms; the box that eats the food is an energy sink, just like a living being.
-1?! This is insightful! All this blathering on about "universes," when we don't even know if more than one exists! This is just theoretical math that proves nothing about anything, but people (especially those who must publish) talk as if it were so.
"Those who consume the bulk of goods are those who make them. We must never forget this secret of our prosperity."
What "states" are "available" is merely a matter of perspective. Most actions humans take are designed toward particular outcomes, i.e. they are intended to reduce the number of possible states to those which are desired. This all raises the question of what a "state" is, and begs the question of whether we can even know how many there are.
"Those who consume the bulk of goods are those who make them. We must never forget this secret of our prosperity."
Way to go Slashdot: mod down those whom you disagree with. Who's the bigot now? Censorship sucks.
"Those who consume the bulk of goods are those who make them. We must never forget this secret of our prosperity."
Indeed.
"Those who consume the bulk of goods are those who make them. We must never forget this secret of our prosperity."
Mod up.
"Those who consume the bulk of goods are those who make them. We must never forget this secret of our prosperity."
"States" as a vague philosophical term are indeed poorly defined and unquantifiable. However, as a technical physics term in physics, there are very precise definitions for "counting states" (where "states" mean, e.g., positions and momenta of the components in a system). The paper shows that a hypothetical critter basing its decisions on how to move on maximization of available states in this precise physics sense will end up "solving" various problem-solving tasks. This hints that some brain function that guides action according to approximate estimates of "physics" accessible-state-maximization might be a component in producing observed "problem solving" behaviors in critters (likely including humans).
Another way of saying this, I think, is that an intelligent creature will seek to maximise the amount of power it has over its environment
It is worth noting that this seems to be exactly what Spinoza calls conatus. In Spinoza thinking, conatus, and not reason, is driving human being behavior
I think it's more reasonable to suggest that there are brain functions which are capable of observing and predicting and operating to produce certain physical actions, rather than estimating conceptual, statistical outcomes.
"Those who consume the bulk of goods are those who make them. We must never forget this secret of our prosperity."
Think what you will; this question won't be resolved without a lot more careful scientific work. However, I think you're confusing yourself by adding the unnecessary requirement that brains need a "conceptual" view to estimate outcomes. I don't think anyone is arguing that the tool-use-capable brains of crows are not only capable of "conceptually" grokking freshman calculus, but also path integral formalisms. However, implementing heuristics approximately equivalent to path integral statistical formalisms isn't out-of-the-question for brain functions. In fact, "massively parallel" computations of the type necessary to (approximately) evaluate path integrals appear to be what the heavily-interlinked networks of zillions of neurons in a brain are good at. "Observing and predicting and operating to produce certain physical actions" might also be a lot harder than you think, given the difficulty of getting supercomputers to perform many of the most basic functions of a retarded squirrel (much less a crow or human).
"we have explicitly proposed a novel physical connection between adaptive behavior and entropy maximization"
But this is simply Darwin's theory of evolution. Their examples just show that there are more possible futures for behaviours that cooperate or stand upright etc. If high fitness relies on some unknown future state, then it is more likely in the case with more possible futures. Hence things evolve generally to 'keep their options open' and 'not close too many doors' and allow more events to be possible.
The Basketball Jones theory of intelligenge:
My brain is like a seive.
I pour knowledge in.
I squeeze the handle a few times.
All the useful stuff comes out the bottom.
All the hard, useless chunks are left inside.