So the more government and the more taxes and the more regulations and inflation the worse the economy is and the less productive people are the more they have to work to satisfy even basic demands.
Hilariously wrong! Here's a chart of US labor productivity and worker compensation. Note how productivity is steadily climbing? US workers today are more efficient/productive than ever in history. Now, look at worker compensation: it perfectly tracks productivity gains up to ~1980, then completely flatlines once Reagan-era "trickle-up" economics took hold. Since we started down the path towards historically low top tax rates and deregulation, the working class hasn't seen a (inflation-adjusted) penny of gain for all their unceasing productivity increases.
OK, here you go: a CNN report with chart of productivity and inflation-adjusted wages. Note how hourly compensation perfectly tracks steady productivity gains up to ~1980, then completely flatlines thanks to Regan era "trickle-up" policies (continuing into the present day) while productivity continues on the same upwards trend.
It's been well over a century since carpets needed to be handmade. The working class did receive the benefits of mechanization for about the first three quarters of the 20th century (including machine-made carpets and cloth) --- however, in the last couple decades of the century, the trend where increasing worker productivity also meant increasing wages/benefits came to a halt. For the last several decades, the American working class has continued to become increasingly productive, but has seen (inflation-adjusted) wages stagnate as all the benefits accrue to a tiny wealthy elite. Improved mechanization no longer means the working class gets more/better stuff for the same work; it means the working class loses jobs and wages, so they're struggling to afford even cheap Wal*Mart crap.
The masses bought into the propaganda narrative that growing working-class prosperity mid-20th-century was the result of capitalism, instead of counter-capitalist workers' movements (unionization, fights for minimum wages and improved working conditions). So, by the Regan era, advances for the working class were brought to a halt (even as the overall economy grew, the amount going to the masses stagnated while all the gains in productivity were given to the rich), and now thrown into full reverse (so the working class is seeing their remaining sliver of the economy trickle away into the pockets of the rich). Total economic productivity has continued to grow plenty to support a continuing trend of decreased work with higher standards of living, but the overwhelming majority of gains are captured by the top 0.01% instead of being distributed to the populace.
If we started handing out the paycuts to the top capitalist class instead, who pocket the savings whenever they replace a worker with a robot, then the working class could receive the benefits of mechanization (same quality of life for less hours of work) instead of just the downsides.
However, when you consider how that GDP is distributed, the US is far worse in terms of return to the nation's citizenry per CO2 produced. The US middle and lower class is *losing* net worth --- the economy for the vast majority of US citizens has been *shrinking*, with all the benefits of polluting accruing to a tiny top elite. In China, that lower GDP per ton CO2 is at least benefiting significant swathes of the Chinese citizenry, who are capturing some economic benefits in return.
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).
because this implies that the crow's brain is capable of reason, even on a subconscious level (isn't the crow's brain entirely subconscious?)
You might be being a bit species chauvinistic. We (I) don't understand consciousness well enough to answer the "conscious/subconscious" question for crows, but "birdbrains" seem to be capable of quite a bit of what was once considered "exclusively human" types of cognition. Why is "reason" --- if by "reason" you mean ability to approximate solutions to mathematical systems of equations --- the sole province of human or "conscious" thought? Simulated "neural networks" demonstrate the flexibility to heuristically approximate a large variety of systems. Why, in particular, would approximating the outcomes of toy models of state availability be particularly out-of-reach?
Ever seen a dog catching stuff thrown by their owner, leaping and twisting through the air to perfectly intercept that tennis ball with its teeth? Critter brains seem pretty good at (approximately) integrating kinematic differential equations. Conceptualizing the accessible range of motional degrees of freedom in a physical system might be a reasonable "extension" of the "kinematic intelligence" (approximate world physics model) that animals (including humans) already exhibit. It's heuristic approximations on a subconscious level --- your dog or human friend doesn't have to "know" calculus well enough to pass an AP exam, but can sure think up approximations faster than a typical college student can solve equations on paper with all that fancy symbolic logic. Any solution that answers complex questions like "how does a critter figure out tool use" is probably going to be a bit complex and clunky (the brain is a complex, clunky tangle of neurons), but "maximize available kinematic states" is at least a simpler and more straightforward general mechanism than many cognitive models for "solving" a variety of problems.
"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).
Yes, an "estimation of future world state accessibility" model is a heck of a lot more complicated than a head-towards-yummy-smells model. Observations show, however, that crows appear pretty smart --- they're able to solve a lot of problems that, e.g., a flatworm with a simple "follow-the-chemical-gradient-towards-yumminess" model, could not. When a crow is observed to move away from a yummy (inaccessible) morsel in order to come back with a twig to fish it out, obviously something much more complex than stuff-yummies-in-beak instinct is going on.
Indeed, this is a "poor alternative to investigating how the crow's brain actually works" --- but with our current level of understanding/technology, we're nowhere near having good "bottom-up" models from neuron interactions to "intelligent" behavior. Like many problems in science, it's useful to "work from both ends and meet in the middle": some people work on better understanding low-level neuron functions; others develop high-level effective theories that, not relying on specific knowledge of how the underlying parts work, still reproduce observed behaviors. Once a scientific field "matures," we can link up the micro- and macro-scale views; cognitive science, however, is still in a rather "primitive" phase, so it's handy to toss out even rudimentary theories for how things like "problem solving behavior" might be generated.
I'm not an author of this paper, and I have no particularly special insight into how right the theory is. However, I think you're trying to stretch the model to apply to the wrong class of issues. The entropy-maximization isn't about complex, "long-term" planning for "future states" on a how-to-survive-to-next-year, or even how-to-survive-to-next-hour, time scale. It's a model for generating short-term actions (e.g. "move towards the stick and grab it; push the stick towards the trapped food ) that can often produce useful results, not explicitly calculating long-term survival plans. However, chances of long-term survival can be greatly enhanced by particular short-term decision strategies, that don't (and can't, since predicting the distant future is basically impossible due to lack of relevant information) require analyzing long-term outcomes.
Mistakes and misunderstandings are "accounted for" by assuming they just happen all the time, but fortunately usually aren't fatal --- the simple "maximize available states" model doesn't break catastrophically if the crow's calculation is a bit off from from what a physicist with a computer would calculate. Equivalently: I bet, in the toy model, that you could introduce significant numerical differences between the physics "estimated" by the model crow and the "real" world physics, and still get similar results.
Nobody said anything about entropy being "the process at work in the crow's brain." What was said is that there may be processes in the crow's brain producing results close to a particular calculation of entropy for simplified model worlds. A process for estimating entropy is not the same as a process which is entropy. Brains seem to be reasonably good at estimating some "real-world" things, so an entropy-estimator is not an a priori unreasonable thing for a brain to have (especially if it confers survival benefits from problem-solving skills).
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.
Nice try with the sophistry here, but... nope. You're begging the question by assuming that a Turing Machine can't exhibit self-organizing behavior to produce some combinations of symbols (e.g. 6666......6666) with greater probability than others. It's darn easy to write a simple computer program that generates 66666.... as the output; in a big "clockwork universe" executing zillions of "subroutines," a simple digit-repeater program is wholly unexceptional. You want to prove your mind is something special? Then lay out a sequence of dice that, when interpreted as a base-6 data representation, has the MD5 hash "66666.....66666". That would be a compelling hard-for-turing-machine-to-produce result.
Yeah, the "logical syllogisms" approach is somewhat of a straw-man extreme of outdated thinking (from much earlier only-humans-can-really-think, and-they-think-like-they-think-they-do era). However, it's still a wide-open question how much "cognition"/problem-solving requires a few general principles, or a big complicated bundle of "specialized" skills. This paper supports a rather "minimalist" approach, where several apparently "advanced" problem-solving skills "pop out" of a rather generic, physics-motivated heuristic model. More accurate "crow models" will very likely need more specific crow "biology and psychology" --- but how much? Also, can you produce "lifelike artificial intelligence" without a lot of fine-tuned "species-specific" baggage, or do you need a big chunk of specialized instincts in addition to first-principles general mechanisms for "intelligent" behavior?
And if the crow instinctually understood undergraduate level stat mech, it would know setting the stick on fire would provide even more state possibilities. The point is that, in a simple "toy model" of the world including a small number of objects that don't come into pieces, accessible state maximization produces "useful" results. Presumably, whatever mental model the crow has to understand/approximate/predict how the universe works is rather simplified compared to an overly-accurate model that calculates entropy down to nuclear excited states. Note, we actually do observe critters exhibiting not only "tool-using" behavior, but also "tool-making" behavior where they break off bits and pieces from one object to create more useful manipulators --- behavior which you probably could model with basically the same formalism.
The PRL journal article is not "pop lit," and precisely defines the "causal path entropy S_c" in a technically correct manner. Admittedly, loose use of the word "entropy" is causing a lot of confusion among the less-scientifically-literate folks posting to Slashdot, so your formulation of "maximizes possible future states of local environment" is indeed better for avoiding confusion in non-technical discussion. Nevertheless, "maximize entropy" (in the formal sense described in the article) is not an incorrect statement; one just needs to be aware that they're not talking about the same entropy of chemical states that a chemist would worry about.
This model posits that an entropy-maximizing motivation (regardless of any food-eating motivation) would allow the crow to "figure out" how to dislodge the food from its nook (without consciously setting the goal of "dislodge food from nook"). Once the food is dislodged, some other motivation (like instincts to grab yummy-smelling stuff with your beak) would be needed to model/explain what completes the getting-food-in-stomach cycle. However, in addition to a grab-food-with-beak ability/instinct (which is of no help when the food is out of reach), an entropy-maximizing ability/instinct might be evolutionarily advantageous since it turns out to be helpful for generating "problem solving" actions (in a wide variety of situations) where more direct "move towards food and cram it in your face" methods aren't particularly helpful. The crow may not need a direct "mental concept" of "get food in my stomach" at all --- instead, it has a variety of behaviors that combine to achieve the result of "food in stomach" and all other behaviors needed to survive and reproduce.
String theory stands in a "philosophical gray area" because it's (currently) too bloody complicated to calculate predictions for things we can currently measure, and the predictions string theory can make produce effects only visible at far too high energy scales to reach with current experimental technology. Also, string theories have a zillion parameters, so you might be able to match them to basically any observed process.
The paper's mechanism doesn't have these problems: it predicts behavior accessible to laboratory measurement (how critters move and act), starting from an extremely simple model with only three tunable parameters (effective "temperatures" for the critter and the world, and the "planning ahead" time interval). Making a model that replicated observable behavior is really the best science can do --- Newtonian gravity didn't fundamentally "explain" gravity, it just posited a force with a particular mathematical form that matched observational data.
True, this theory is in a very "immature" early phase, with gigantic gaps and uncertainties in how (and if) this links up to "microscopic scale" biological neural systems. Nonetheless, it's a starting point, and absolutely offers falsifiability: for all cases where critters don't move/act according to this model, the model is wrong (unlike, e.g., "intelligent design," where you can always say "that's how the designer made it"). Neither is the model particularly biologically implausible: state availability heuristics integrating over potential paths is the type of massively-parallel approximating calculation that brain neural networks could be good at. As "bottom-up" understanding of neural systems advances, the theory will also become falsifiable on "what biological systems actually do" grounds.
Working "inward" from "both sides" of a problem (bottom-up microscopic dynamics and top-down macroscopic effective approximations) is an extremely common component of scientific practice.
The interesting thing about changing the motive from "hunger" to "maximizing states" is that (a) the same "maximizing states" motivation works to explain a variety of behaviors, instead of needing a distinct motive for each, and (b) "maximizing states" simultaneously provides a "motive" and a theoretical mechanism for achieving that motive --- "I'm hungry" doesn't directly help with solving the food-trapped-in-hole problem, while the simple "maximize states" motive (precisely mathematically formulated) actually generated "problem-solving" behavior.
For this to work, it is indeed still necessary for the crow to have a mechanism for estimating how its controllable bodily motions impact the range of accessible states. The crow needs to "be a physicist" --- i.e. have an internal mental model of how the world works, which it can use to estimate the state accessibility impact of actions. Smart critters (humans included) do seem to have a subconscious "physics sense" about the world (which, e.g., allows you to catch a ball tossed to you), so it's not completely bonkers to assume that critters can "do the math" enough to calculate heuristics for more or less "constrained" outcomes. To me, this at least seems less bonkers than (now mostly outdated) approaches that tend towards assuming something that looks like formal syllogistic symbolic logic generates "smart" behaviors.
How is this paper not a scientific approach to empirical data? We have empirical observations of a wide range of animal/human behaviors. The authors propose a toy mathematical model that reproduces key features of several interesting observed behaviors. This is perfectly good science, just like saying "hey, an F=G*m_1*m_2/r^2 force between massive objects recreates the observed motions of the heavenly bodies" --- a predictive, testable mathematical model that can be compared with measurements of the motions and behaviors of actual critters to see how well it works.
Thanks for the additional references. Like the paper's authors, I'm a physicist, and personally ignorant of the status of cutting-edge cognitive science research, so I don't make any claims about whether this research is particularly novel or useful in the cognitive science field. But it was at least novel and interesting to me, and cool to see a "physicist's approach" with a simple implemented mathematical model generating "complex, intelligent" results.
Yes, the model "intelligence" does have the one goal of "future history maximization." The interesting thing is that this one particular goal can produce a variety of behaviors --- instead of requiring each behavior to be motivated by its own specific and non-generalizable goal. Instead of needing a brain with specific goals for "walk upright," "get tool to augment manipulative abilities," "cooperate with other to solve problem," plus a heap of other mechanisms to achieve said goals, the simple entropy-maximization principle generates a wide variety of apparently "complex" and "intelligent" behaviors (both selecting the "goal" and implementing actions to achieve the goal).
The critter is assumed to have a certain capacity for expending energy to do work on their environment (parametrized by their "temperature" T_c) --- needing to expend energy is not a barrier. With the stick balanced up, you can quickly and easily swing the stick to many other combinations of position and velocity. When the stick is dangling down, it takes more time rocking back and forth to "swing it up" into many positions. If the critter was very strong (high T_c, able to exert much greater forces than gravity on the stick), then it wouldn't care so much about where the stick was (since it could swing it into any state with approximately equal ease with or against gravity). However, in the model example, the stick is somewhat "heavy" compared to the forces the cart can exert, so it has to take gravity into account and carefully coordinate its motions to gently swing the stick into place (and prefers a position where gravity will work to help, rather than hinder, its future range of possibilities for the stick).
The claim of the paper isn't that "entropy maximization" is the sole motivating factor for all behaviors. For example, after their toy model critter succeeds in knocking the "food" out of the hole using the "tool," some other guiding mechanism probably takes over to make it eat the tasty food (which decreases the accessible degrees of freedom in their simple model compared to keeping the food and tool nearby to toss about). So, indeed, there are plenty of actions that require different behavioral models to explain (e.g. instincts to stuff yummy-smelling objects into your face). However, the general "entropy maximization" model elegantly reproduces several kinds of "intelligent" cognitive behaviors without requiring immensely clunky and overly-specific mechanisms.
Slashdot Mirror
So the more government and the more taxes and the more regulations and inflation the worse the economy is and the less productive people are the more they have to work to satisfy even basic demands.
Hilariously wrong! Here's a chart of US labor productivity and worker compensation. Note how productivity is steadily climbing? US workers today are more efficient/productive than ever in history. Now, look at worker compensation: it perfectly tracks productivity gains up to ~1980, then completely flatlines once Reagan-era "trickle-up" economics took hold. Since we started down the path towards historically low top tax rates and deregulation, the working class hasn't seen a (inflation-adjusted) penny of gain for all their unceasing productivity increases.
OK, here you go: a CNN report with chart of productivity and inflation-adjusted wages. Note how hourly compensation perfectly tracks steady productivity gains up to ~1980, then completely flatlines thanks to Regan era "trickle-up" policies (continuing into the present day) while productivity continues on the same upwards trend.
It's been well over a century since carpets needed to be handmade. The working class did receive the benefits of mechanization for about the first three quarters of the 20th century (including machine-made carpets and cloth) --- however, in the last couple decades of the century, the trend where increasing worker productivity also meant increasing wages/benefits came to a halt. For the last several decades, the American working class has continued to become increasingly productive, but has seen (inflation-adjusted) wages stagnate as all the benefits accrue to a tiny wealthy elite. Improved mechanization no longer means the working class gets more/better stuff for the same work; it means the working class loses jobs and wages, so they're struggling to afford even cheap Wal*Mart crap.
The masses bought into the propaganda narrative that growing working-class prosperity mid-20th-century was the result of capitalism, instead of counter-capitalist workers' movements (unionization, fights for minimum wages and improved working conditions). So, by the Regan era, advances for the working class were brought to a halt (even as the overall economy grew, the amount going to the masses stagnated while all the gains in productivity were given to the rich), and now thrown into full reverse (so the working class is seeing their remaining sliver of the economy trickle away into the pockets of the rich). Total economic productivity has continued to grow plenty to support a continuing trend of decreased work with higher standards of living, but the overwhelming majority of gains are captured by the top 0.01% instead of being distributed to the populace.
If we started handing out the paycuts to the top capitalist class instead, who pocket the savings whenever they replace a worker with a robot, then the working class could receive the benefits of mechanization (same quality of life for less hours of work) instead of just the downsides.
However, when you consider how that GDP is distributed, the US is far worse in terms of return to the nation's citizenry per CO2 produced. The US middle and lower class is *losing* net worth --- the economy for the vast majority of US citizens has been *shrinking*, with all the benefits of polluting accruing to a tiny top elite. In China, that lower GDP per ton CO2 is at least benefiting significant swathes of the Chinese citizenry, who are capturing some economic benefits in return.
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).
because this implies that the crow's brain is capable of reason, even on a subconscious level (isn't the crow's brain entirely subconscious?)
You might be being a bit species chauvinistic. We (I) don't understand consciousness well enough to answer the "conscious/subconscious" question for crows, but "birdbrains" seem to be capable of quite a bit of what was once considered "exclusively human" types of cognition. Why is "reason" --- if by "reason" you mean ability to approximate solutions to mathematical systems of equations --- the sole province of human or "conscious" thought? Simulated "neural networks" demonstrate the flexibility to heuristically approximate a large variety of systems. Why, in particular, would approximating the outcomes of toy models of state availability be particularly out-of-reach?
Ever seen a dog catching stuff thrown by their owner, leaping and twisting through the air to perfectly intercept that tennis ball with its teeth? Critter brains seem pretty good at (approximately) integrating kinematic differential equations. Conceptualizing the accessible range of motional degrees of freedom in a physical system might be a reasonable "extension" of the "kinematic intelligence" (approximate world physics model) that animals (including humans) already exhibit. It's heuristic approximations on a subconscious level --- your dog or human friend doesn't have to "know" calculus well enough to pass an AP exam, but can sure think up approximations faster than a typical college student can solve equations on paper with all that fancy symbolic logic. Any solution that answers complex questions like "how does a critter figure out tool use" is probably going to be a bit complex and clunky (the brain is a complex, clunky tangle of neurons), but "maximize available kinematic states" is at least a simpler and more straightforward general mechanism than many cognitive models for "solving" a variety of problems.
"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).
Yes, an "estimation of future world state accessibility" model is a heck of a lot more complicated than a head-towards-yummy-smells model. Observations show, however, that crows appear pretty smart --- they're able to solve a lot of problems that, e.g., a flatworm with a simple "follow-the-chemical-gradient-towards-yumminess" model, could not. When a crow is observed to move away from a yummy (inaccessible) morsel in order to come back with a twig to fish it out, obviously something much more complex than stuff-yummies-in-beak instinct is going on.
Indeed, this is a "poor alternative to investigating how the crow's brain actually works" --- but with our current level of understanding/technology, we're nowhere near having good "bottom-up" models from neuron interactions to "intelligent" behavior. Like many problems in science, it's useful to "work from both ends and meet in the middle": some people work on better understanding low-level neuron functions; others develop high-level effective theories that, not relying on specific knowledge of how the underlying parts work, still reproduce observed behaviors. Once a scientific field "matures," we can link up the micro- and macro-scale views; cognitive science, however, is still in a rather "primitive" phase, so it's handy to toss out even rudimentary theories for how things like "problem solving behavior" might be generated.
I'm not an author of this paper, and I have no particularly special insight into how right the theory is. However, I think you're trying to stretch the model to apply to the wrong class of issues. The entropy-maximization isn't about complex, "long-term" planning for "future states" on a how-to-survive-to-next-year, or even how-to-survive-to-next-hour, time scale. It's a model for generating short-term actions (e.g. "move towards the stick and grab it; push the stick towards the trapped food
) that can often produce useful results, not explicitly calculating long-term survival plans. However, chances of long-term survival can be greatly enhanced by particular short-term decision strategies, that don't (and can't, since predicting the distant future is basically impossible due to lack of relevant information) require analyzing long-term outcomes.
Mistakes and misunderstandings are "accounted for" by assuming they just happen all the time, but fortunately usually aren't fatal --- the simple "maximize available states" model doesn't break catastrophically if the crow's calculation is a bit off from from what a physicist with a computer would calculate. Equivalently: I bet, in the toy model, that you could introduce significant numerical differences between the physics "estimated" by the model crow and the "real" world physics, and still get similar results.
Nobody said anything about entropy being "the process at work in the crow's brain." What was said is that there may be processes in the crow's brain producing results close to a particular calculation of entropy for simplified model worlds. A process for estimating entropy is not the same as a process which is entropy. Brains seem to be reasonably good at estimating some "real-world" things, so an entropy-estimator is not an a priori unreasonable thing for a brain to have (especially if it confers survival benefits from problem-solving skills).
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.
Nice try with the sophistry here, but... nope. You're begging the question by assuming that a Turing Machine can't exhibit self-organizing behavior to produce some combinations of symbols (e.g. 6666......6666) with greater probability than others. It's darn easy to write a simple computer program that generates 66666.... as the output; in a big "clockwork universe" executing zillions of "subroutines," a simple digit-repeater program is wholly unexceptional. You want to prove your mind is something special? Then lay out a sequence of dice that, when interpreted as a base-6 data representation, has the MD5 hash "66666.....66666". That would be a compelling hard-for-turing-machine-to-produce result.
Yeah, the "logical syllogisms" approach is somewhat of a straw-man extreme of outdated thinking (from much earlier only-humans-can-really-think, and-they-think-like-they-think-they-do era). However, it's still a wide-open question how much "cognition"/problem-solving requires a few general principles, or a big complicated bundle of "specialized" skills. This paper supports a rather "minimalist" approach, where several apparently "advanced" problem-solving skills "pop out" of a rather generic, physics-motivated heuristic model. More accurate "crow models" will very likely need more specific crow "biology and psychology" --- but how much? Also, can you produce "lifelike artificial intelligence" without a lot of fine-tuned "species-specific" baggage, or do you need a big chunk of specialized instincts in addition to first-principles general mechanisms for "intelligent" behavior?
And if the crow instinctually understood undergraduate level stat mech, it would know setting the stick on fire would provide even more state possibilities. The point is that, in a simple "toy model" of the world including a small number of objects that don't come into pieces, accessible state maximization produces "useful" results. Presumably, whatever mental model the crow has to understand/approximate/predict how the universe works is rather simplified compared to an overly-accurate model that calculates entropy down to nuclear excited states. Note, we actually do observe critters exhibiting not only "tool-using" behavior, but also "tool-making" behavior where they break off bits and pieces from one object to create more useful manipulators --- behavior which you probably could model with basically the same formalism.
The PRL journal article is not "pop lit," and precisely defines the "causal path entropy S_c" in a technically correct manner. Admittedly, loose use of the word "entropy" is causing a lot of confusion among the less-scientifically-literate folks posting to Slashdot, so your formulation of "maximizes possible future states of local environment" is indeed better for avoiding confusion in non-technical discussion. Nevertheless, "maximize entropy" (in the formal sense described in the article) is not an incorrect statement; one just needs to be aware that they're not talking about the same entropy of chemical states that a chemist would worry about.
This model posits that an entropy-maximizing motivation (regardless of any food-eating motivation) would allow the crow to "figure out" how to dislodge the food from its nook (without consciously setting the goal of "dislodge food from nook"). Once the food is dislodged, some other motivation (like instincts to grab yummy-smelling stuff with your beak) would be needed to model/explain what completes the getting-food-in-stomach cycle. However, in addition to a grab-food-with-beak ability/instinct (which is of no help when the food is out of reach), an entropy-maximizing ability/instinct might be evolutionarily advantageous since it turns out to be helpful for generating "problem solving" actions (in a wide variety of situations) where more direct "move towards food and cram it in your face" methods aren't particularly helpful. The crow may not need a direct "mental concept" of "get food in my stomach" at all --- instead, it has a variety of behaviors that combine to achieve the result of "food in stomach" and all other behaviors needed to survive and reproduce.
String theory stands in a "philosophical gray area" because it's (currently) too bloody complicated to calculate predictions for things we can currently measure, and the predictions string theory can make produce effects only visible at far too high energy scales to reach with current experimental technology. Also, string theories have a zillion parameters, so you might be able to match them to basically any observed process.
The paper's mechanism doesn't have these problems: it predicts behavior accessible to laboratory measurement (how critters move and act), starting from an extremely simple model with only three tunable parameters (effective "temperatures" for the critter and the world, and the "planning ahead" time interval). Making a model that replicated observable behavior is really the best science can do --- Newtonian gravity didn't fundamentally "explain" gravity, it just posited a force with a particular mathematical form that matched observational data.
True, this theory is in a very "immature" early phase, with gigantic gaps and uncertainties in how (and if) this links up to "microscopic scale" biological neural systems. Nonetheless, it's a starting point, and absolutely offers falsifiability: for all cases where critters don't move/act according to this model, the model is wrong (unlike, e.g., "intelligent design," where you can always say "that's how the designer made it"). Neither is the model particularly biologically implausible: state availability heuristics integrating over potential paths is the type of massively-parallel approximating calculation that brain neural networks could be good at. As "bottom-up" understanding of neural systems advances, the theory will also become falsifiable on "what biological systems actually do" grounds.
Working "inward" from "both sides" of a problem (bottom-up microscopic dynamics and top-down macroscopic effective approximations) is an extremely common component of scientific practice.
The interesting thing about changing the motive from "hunger" to "maximizing states" is that (a) the same "maximizing states" motivation works to explain a variety of behaviors, instead of needing a distinct motive for each, and (b) "maximizing states" simultaneously provides a "motive" and a theoretical mechanism for achieving that motive --- "I'm hungry" doesn't directly help with solving the food-trapped-in-hole problem, while the simple "maximize states" motive (precisely mathematically formulated) actually generated "problem-solving" behavior.
For this to work, it is indeed still necessary for the crow to have a mechanism for estimating how its controllable bodily motions impact the range of accessible states. The crow needs to "be a physicist" --- i.e. have an internal mental model of how the world works, which it can use to estimate the state accessibility impact of actions. Smart critters (humans included) do seem to have a subconscious "physics sense" about the world (which, e.g., allows you to catch a ball tossed to you), so it's not completely bonkers to assume that critters can "do the math" enough to calculate heuristics for more or less "constrained" outcomes. To me, this at least seems less bonkers than (now mostly outdated) approaches that tend towards assuming something that looks like formal syllogistic symbolic logic generates "smart" behaviors.
How is this paper not a scientific approach to empirical data? We have empirical observations of a wide range of animal/human behaviors. The authors propose a toy mathematical model that reproduces key features of several interesting observed behaviors. This is perfectly good science, just like saying "hey, an F=G*m_1*m_2/r^2 force between massive objects recreates the observed motions of the heavenly bodies" --- a predictive, testable mathematical model that can be compared with measurements of the motions and behaviors of actual critters to see how well it works.
Thanks for the additional references. Like the paper's authors, I'm a physicist, and personally ignorant of the status of cutting-edge cognitive science research, so I don't make any claims about whether this research is particularly novel or useful in the cognitive science field. But it was at least novel and interesting to me, and cool to see a "physicist's approach" with a simple implemented mathematical model generating "complex, intelligent" results.
Yes, the model "intelligence" does have the one goal of "future history maximization." The interesting thing is that this one particular goal can produce a variety of behaviors --- instead of requiring each behavior to be motivated by its own specific and non-generalizable goal. Instead of needing a brain with specific goals for "walk upright," "get tool to augment manipulative abilities," "cooperate with other to solve problem," plus a heap of other mechanisms to achieve said goals, the simple entropy-maximization principle generates a wide variety of apparently "complex" and "intelligent" behaviors (both selecting the "goal" and implementing actions to achieve the goal).
The critter is assumed to have a certain capacity for expending energy to do work on their environment (parametrized by their "temperature" T_c) --- needing to expend energy is not a barrier. With the stick balanced up, you can quickly and easily swing the stick to many other combinations of position and velocity. When the stick is dangling down, it takes more time rocking back and forth to "swing it up" into many positions. If the critter was very strong (high T_c, able to exert much greater forces than gravity on the stick), then it wouldn't care so much about where the stick was (since it could swing it into any state with approximately equal ease with or against gravity). However, in the model example, the stick is somewhat "heavy" compared to the forces the cart can exert, so it has to take gravity into account and carefully coordinate its motions to gently swing the stick into place (and prefers a position where gravity will work to help, rather than hinder, its future range of possibilities for the stick).
The claim of the paper isn't that "entropy maximization" is the sole motivating factor for all behaviors. For example, after their toy model critter succeeds in knocking the "food" out of the hole using the "tool," some other guiding mechanism probably takes over to make it eat the tasty food (which decreases the accessible degrees of freedom in their simple model compared to keeping the food and tool nearby to toss about). So, indeed, there are plenty of actions that require different behavioral models to explain (e.g. instincts to stuff yummy-smelling objects into your face). However, the general "entropy maximization" model elegantly reproduces several kinds of "intelligent" cognitive behaviors without requiring immensely clunky and overly-specific mechanisms.