Kevin said: "I'm just saying: if you use evolutionary principals in the same way nature does you are more likely to end up with an artificial person than an artificial intelligence - and their behavior can be a bit unpredictable.
"

This is an interesting bias. What is it about evolution that you think includes things not human, and excludes other things done or created by us humans? This distinction is just plain wrong. If you think it isn't, please show me the line in the sand that delineates natural from un-natural, the line that that delineates the results of evolution from the the results of human thinking or human tinkering.

I think I was unclear: you can use evolutionary/genetic methods for all sorts of things, you just need to be careful about what the system goals are. Given the obvious limitations of humans, I wouldn't want to use a system with the same goals that produced humans.

However, evolution is a "brute force" method that may or may not work and has mixed results. Humans design tools to handle tasks in a much more directed way. I don't recall ever delineating the approaches as natural or otherwise, and you can mix them if it suits.

Here is what I know: If it looks like what a system does, any system, anything it does, is not evolutionary or "natural", then you can be absolutely sure you don't understand evolution or the causal mechanics of the system in focus.

Not sure what point you are trying to make.

The tendency to see humans and the product of humanity as artificial is an common mistake. The mind is so easily black-box-ed… especially when the machine used to look at the mind is the ragged edge, the nosebleed hight, of abstraction that we call consciousness.

http://www.merriam-we...

I'm an engineer, the mind is just another machine to be taken apart and understood - then rebuilt better.

I am further dismayed by your existential separation of the words "creativity" and "intelligence". I don't have time or the energy to go back to the basics at this moment, but every process is creative and all systems are to some extent intelligent. This discussion, any discussion about AI, is meaningless unless it is at heart a discussion about systems at least as complex as human intelligence and at least as creative. Assumed also is a level of autonomy equivalent to that enjoyed by human minds.

I think you might be overreaching.

Maybe your interests fall towards so-called "knowledge" or "expert" systems, towards database aids as are currently bantered about as "semantic web" or "collective intelligence". If so, you must be frustrated by hard-AI as is discussed in this meetup. Maybe you think true AI is impossible or that the most effective path from here to AI is through mundane and pedestrian systems (expert, semantic, etc.). What motivates your interest in AI?

Randall Reetz

Nope, I'm a hard-AI guy. However I would be more convinced AI was possible with current hardware if I was seeing a few expert systems and autonomous robots. Given the difficulty of getting a robot car to drive around, I'd say there is still a lot of work to do.

I'm interested in how to program massively parallel applications - AI seems to be one of those, and it includes aspects of real-world modeling that I'm interested in too.

Good. OK.

You too quickly moved past my point that everything is evolutionary. And then you restated in different words the same mistake I was rebutting. You contend that human minds solve problems with less brute force than does evolution. Upon which measurements do you base your assertions? You ask that AI prove itself through the development of autonomous intelligence based on "current hardware". Meaning, you require that a chip do a significant part of what the human mind does despite the fact that the chip is about a billion times less complex. Should I instead ask that you prove that you demonstrate your ability to perform acts of intelligence a billion times your baseline performance? Let's be fair here.

Let us also remember the causal chains upon which systems and their abilities rest. For instance it isn't the automobile that makes individual transportation possible. What makes this level of freedom over geography possible is automobile manufacturing plants and the long chain of materials and process delivery upon which they depend. Add in a global network of roads and a gas refinery and delivery system and you have all the makings for the kind of magical invisible infrastructure that allows for the kind of embarrassing and arrogant hubris that allows a person to falsely point to the car itself and declare it a revolution. Same goes for computer chips (its the fabs that make computing). Same goes for software (it is the compilers and the operating systems). Same goes for human problem solving (absolutely dependent on the long evolutionary history of the universe AND the fact that each individual thought is itself the product of an evolutionary (least energy) process.

I hear nothing in any of your criticisms or attitude that would support your assertion that you are a "hard AI guy". You want intelligence without the kind of fuzzy answer space that is the domain of complex systems.

Randall Reetz

Good. OK.

...You contend that human minds solve problems with less brute force than does evolution. Upon which measurements do you base your assertions?

Just observation.

You ask that AI prove itself through the development of autonomous intelligence based on "current hardware". Meaning, you require that a chip do a significant part of what the human mind does despite the fact that the chip is about a billion times less complex. Should I instead ask that you prove that you demonstrate your ability to perform acts of intelligence a billion times your baseline performance? Let's be fair here.

I'm not asking AI to prove anything, I am merely observing that humans are not very good at designing certain kinds of machines: if you can't build a dumb robot you aren't going to be able to build a smart one either.
The pieces that go to make up a human brain (the neurons) appear to be fairly simple in themselves, as such chips at a similar complexity level (combined in large numbers) should be programmable to perform similar tasks, however at the moment we don't have the appropriate programming tools.

Let us also remember the causal chains upon which systems and their abilities rest. For instance it isn't the automobile that makes individual transportation possible. What makes this level of freedom over geography possible is automobile manufacturing plants and the long chain of materials and process delivery upon which they depend. Add in a global network of roads and a gas refinery and delivery system and you have all the makings for the kind of magical invisible infrastructure that allows for the kind of embarrassing and arrogant hubris that allows a person to falsely point to the car itself and declare it a revolution. Same goes for computer chips (its the fabs that make computing). Same goes for software (it is the compilers and the operating systems). Same goes for human problem solving (absolutely dependent on the long evolutionary history of the universe AND the fact that each individual thought is itself the product of an evolutionary (least energy) process.

I think you are digressing somewhat...

I hear nothing in any of your criticisms or attitude that would support your assertion that you are a "hard AI guy". You want intelligence without the kind of fuzzy answer space that is the domain of complex systems.

Randall Reetz

All you need to be a "hard AI guy" is to believe that intelligence is just algorithmic, and that given the right hardware and software an AI can be built - or maybe just the right software if you aren't in a hurry.

My parallel C++ project provides a means for describing neural networks efficiently, but I think it needs a meta language for describing higher level algorithms like intelligent thought, vision or sound processing. At the moment I suspect that the insight to design the specific meta language is beyond my abilities, however I might be up to the task of building a language that has genetic/evolutionary aspects that will get there with some persuasion.

Kevin,

I don't know where to begin, or even whether or not I should. I really don't think it is fair that you pretend to be interested in the science of intelligence (or anything else for that matter) and then base huge areas of your assertions on statements like: "Just observation."

That your entire scientific philosophy on intelligence is to be defended with a two word answer like that? I am supposed to take anything you say after that seriously?

Based on what you have written here, it would be reasonable to conclude that all of your assertions are based on the notion that intelligence is somehow qualitatively different as a process than all of the other processes in nature. Not quantitatively, but qualitatively. If that is so, then you are going to have to point to some line in time before which there was nothing like intelligence and some point in time after which intelligence existed. Furthermore, because you are saying that intelligence is not an evolutionary process, you are going to have to show how this new ether, this new ichor, this new ectoplasmic substance, this essence of intelligence is somehow not the same as evolution, does not start as more energy dense configurations and become through time, more energy diffuse, like everything else.

Kevin, I doubt you really believe this. I doubt that you really believe that intelligence is extra-evolutionary.

This magic or "algorithmic" substance or structure that you assign to intelligence and not to everything else... what is it? How does the mind accomplish this exclusion of evolution and simultaneously manufacture change via this new and non-evolutionary other stuff?

But mainly, what I see you doing is epidemic among human thinking. It is as if one could say of a professional tennis player that their ability was independent of all of thousands of hours they had played and been coached. It would be like looking at strand of human DNA and saying that its structure and all of its resulting behaviors were independent of the billions of years of evolution that shaped it and all of the ancillary bio-machinery with which it interacts.

When you observe a person (or a computer) asked "What is 2+2?", answer "4", you can not assert that because that entity's answer was constructed out of sight, that the process of deriving the answer was not evolutionary. That is akin to asking the earth how to build an oxygen-rich atmosphere, going away for a billion years, and returning to see plants and assuming that no process led to the development of those oxygen producing plants.

If you are going to make such bold and un-explained assertions, the burden is on you to at the very least posit a guess as to how intelligence is different from the process that constructs all other complexities in this universe. If "Just observation." was sufficient an answer to questions about complex processes, then it would be well within rational thought to believe that people on TV can levitate, disappear, stick their arms through solid concrete and steel, make elephants and cars disappear, and all manor of other physics-busting activities. Everyone would be understandably under-whelmed if a TV observer came to the conclusion that TV actors are subject to a whole different form of physics.

An algorithm, independent of its elegance or computational scope, was developed by some process before it was executed on logic or in the human brain. But even its execution, subject to scrutiny at a sufficiently fine resolution, will reveal all of the attributes and aspects of an evolutionary system: energy input, work of some sort, a result or answer, some amount of waste energy that has been degraded in the process of execution, each of the actions necessary within the clockwork that is execution can only happen by the laws of evolution, and causality, and, neither the computer doing the calculation, nor its environment, is not the same afterwards.

Please either take the time to show me that you have a well worked out position backing up your assertions or don't bother extending this conversation. If you believe as it seems you do, that algorithms are qualitatively different and external from evolution, and if this belief is as central to your AI cosmology as you assert it is, then it would seem that the very least you should attempt to make a public guess as to what this "algorithm" thing is and how it is fundamentally different from the evolutionary process of which everything else is built.

Thank you,

Randall Reetz

Kevin,

I don't know where to begin, or even whether or not I should. I really don't think it is fair that you pretend to be interested in the science of intelligence (or anything else for that matter) and then base huge areas of your assertions on statements like: "Just observation."

That your entire scientific philosophy on intelligence is to be defended with a two word answer like that? I am supposed to take anything you say after that seriously?

Observation: nature/evolution appears to have taken millions/billions of years to produce intelligent animals, humans have come up with the iPod in a matter of hundreds from the discovery of electricity.

Darwin's theory of evolution came from observation. Most hard science starts with observing phenomena and then trying to explain them.

Based on what you have written here, it would be reasonable to conclude that all of your assertions are based on the notion that intelligence is somehow qualitatively different as a process than all of the other processes in nature. Not quantitatively, but qualitatively. If that is so, then you are going to have to point to some line in time before which there was nothing like intelligence and some point in time after which intelligence existed. Furthermore, because you are saying that intelligence is not an evolutionary process, you are going to have to show how this new ether, this new ichor, this new ectoplasmic substance, this essence of intelligence is somehow not the same as evolution, does not start as more energy dense configurations and become through time, more energy diffuse, like everything else.

Kevin, I doubt you really believe this. I doubt that you really believe that intelligence is extra-evolutionary.

You are reading stuff into my words that isn't there. Obviously (by observation) evolution can give you an intelligent animal.

The problem is that we are (as yet) unable to disassemble the intelligent machine and discern its workings.

This magic or "algorithmic" substance or structure that you assign to intelligence and not to everything else... what is it? How does the mind accomplish this exclusion of evolution and simultaneously manufacture change via this new and non-evolutionary other stuff?

But mainly, what I see you doing is epidemic among human thinking. It is as if one could say of a professional tennis player that their ability was independent of all of thousands of hours they had played and been coached. It would be like looking at strand of human DNA and saying that its structure and all of its resulting behaviors were independent of the billions of years of evolution that shaped it and all of the ancillary bio-machinery with which it interacts.

I think that takes back to an earlier point that the physical structure humans is programmed at the genetic level, and learning adds the extra skills.

Being good at sports requires particular physical traits, if you don't have them learning won't help. The same would apply to a robot, i.e. if the robot can't hold a tennis racket it isn't going to be able to play regardless of its IQ.

...

Please either take the time to show me that you have a well worked out position backing up your assertions or don't bother extending this conversation. If you believe as it seems you do, that algorithms are qualitatively different and external from evolution, and if this belief is as central to your AI cosmology as you assert it is, then it would seem that the very least you should attempt to make a public guess as to what this "algorithm" thing is and how it is fundamentally different from the evolutionary process of which everything else is built.

Thank you,

Randall Reetz

The algorithms in the brain work on at least two levels: what the machine supports directly (because that's the way the hardware has evolved) e.g. walking, and learned algorithms on top of that e.g. 2+2=4.

Obviously evolution has created the entire machine, but the machine creates its own view of the world and rules about it.

As a believer in "hard AI" I'm assuming the higher level algorithms are separable from the machine level. I.e. I could build myself a pseudo-human brain with all the neurons and all the right connections and it will behave the same as a real one and/or if I can find the right abstraction of the base machine I can just work with the higher level parts.

If you are claiming that the higher level thought processes are intrinsically tied to the biology then you really aren't a believer in "hard AI".

Another assumption I make is that the IQ of an AI is going to depend on the hardware it runs on. I would guess that to a large extent it will be rule-based and the rules will vary depending on the abilities/needs of the hardware. As such I would expect the first (commercial) AIs to be domain-expert helper systems - e.g. Monica's language processing system, and motion/process control systems. As better hardware/software becomes available the IQ should improve and there may be no specific inflection point (singularity) in the process.

Kev.

Kevin, this discussion will go around in circles forever if you refuse to talk to my assertion that learning IS evolution.

The fact that attributes competing for resources in a mind are abstractions and not eyes and limbs (or the Genes that pattern them) is meaningless to the over arching scheme.

Nature (physics) doesn't discriminate. All systems are thermodynamic at base. Neither the chunks being aggregated nor the forces effecting their aggregation are ever different enough to make one evolving system "natural" and one "cultural" or "learning". These labels are simply identity conventions and say nothing of actual causal defining differences.

What makes thinking faster than DNA based genetic evolution is strictly a factor of the efficiency afforded within the brain (density, ease of aggregation, flat(er) energy topology, redundant expendable parts and associations, etc.). None of these factors do anything for thinking that isn't happening in genetic evolution. In fact, thinking is the evolutionary result of, well, of the pressures that cause evolution. That is what evolution does, it finds new and more effective ways of evolving.

The way you talk of intelligence, as a Pandora's box, it is akin to turning the key in your car's ignition and concluding that the thing drives around all day from the energy of that original twist of wrist.

What amazes me about all of these discussions about AI is the apparent inability of humans to imagine a means of evolution as much more effective as memetic evolution is when compared to genetic evolution.

The reason I hammer this evolution theme is this: To build a new structure one must understand the concept of structure. One can hack together a shack without really understanding much about the factors effecting structure. But if one intends to build an economically viable and safe building half a mile high, or a thousand feet below the sea, or in orbit, then one had better have a great deal of knowledge about structure and construction. Ignorance of the most important causal factors effecting the aggregation and behavior of a system will limit the complexity of the projects possible. If we don't get that learning and computation and thinking and intelligence are evolutionary processes at base, we well be forever stuck building successful simple projects and spectacular catastrophes when we build towards the complexity that is intelligence.

Randall Reetz

If structure exists, it is there as a result of evolution. Period. The attributes the substrate under examination don't matter a wit. Biology, a mountain, ocean currents, stars, planets, galaxies, atoms, molecules, ideas, learning, computing, etc. None of it either exists or operates by any other mechanism than evolution. The oft observed distinction between that process which shapes the beak of a finch and that process that allows us to learn to play tennis... well it is an illusion.

An idea is every bit as physical as the brain from which it arises. No special physics arbitrate learning, thinking, computation. Evolution is evolution. It is the one and only process. The fact that we pay special attention to the product's of evolution that we deem "progress" does in no way exclude all other products of change… disorder, chaos, the steady (universal) degradation of energy and structure.

Until we get this we will fail at attempts to build systems as complex as abstraction machines need to be.

Standing in the way of our understanding of thinking as a process, like a hundred foot high wall of stone, is the unfortunate fact that the conscious sensory experience of thinking is so fundamentally limited. Cartoon.

Randall Reetz

Not sure I agree entirely with that. Evolution by itself is fairly goalless, other than the "survival of the fittest". Humans are (as far as we can tell) a product of evolution, however human artifacts are mostly created with a design intent. Designs may "evolve" over time and compete for survival in the market place, the artifacts themselves don't (as yet). You can decide that these are the same processes, but is it useful to do so?

Your argument that we have to understand the underlying process of evolution and its ubiquity is clearly wrong since we got this far without understanding it. The laws of physics and chemistry do not evolve, and most of human achievement so far has not needed other understanding. Humans build very complex systems by individually specializing in different things and then combining the results to achieve more than any individual alone, and given the continuation of Moore's law I'd say that process has not run out of steam yet, and the ability to build the large and complex systems for AI is probably here already. Incremental improvement in existing processes (evolution) will continue, with occasional mutations, extinctions and restarts.

Personally I suspect the wall is nearer 9ft and some jumping is required.

Kev.

Kevin,

Please show me a process that isn't evolution at base. I think you have this hundred year old notion of evolution that is bio-centric, sexual-centric, entity-centric, individual-centric, species-centric, etc... and where oh where do you get this "survival of the fittest" statement?

Not only is is "useful" to "decide" that all process is evolution, it is the only possible explanation, and "use"less to attempt a project as complex as the "design" of sentience from any other perspective.

All of your arguments are built on seemingly innocent and smallish falsehoods ("I suspect... some jumping is required"). In others posts, your innocent mistake is what you simply call "thinking", or "what humans do", or "design".

Yet these black-box secret-sauce ingredients represent the essence of the machinery you claim to be interested in building. How are you going to do this outside of an understanding of the evolutionary process, of why evolution happens? Outside of a profound understanding of what evolution does and more importantly, why it does it?

If you are stuck on the word "fitness", lets use it. What do you think "fitness" ultimately means? Meaning, there is an obvious direction to change. What accounts for the fact that there was once just the strong force and quantum gas, and now there are things like you and me debating the causal influences at play in the universe?

Oh, maybe I understand. Maybe you think sexual selection and reproduction are the only means by which evolution happens. Wow, that wouldn't be very evolutionary. A system that moves towards complexity handling capacity, but only so long as there are organisms that swim or run around and have sex? I guess if you think that way, you know the mechanism that biology uses, and by knowing that, you don't really know what evolution is at all. If you believe that evolution is the domain of blindness, you really haven't thought about what an organism is. If you think organisms exist purely to survive as individuals, you really can't explain why complexity handling increases over time. If you can't explain that, you will never be able to build a thinking machine.

Why do abstractions exist? What is an abstraction? Is anything, not an abstraction? What does evolution have to do with the building of abstractions? Why expend the energy to build a map? If you do expend that energy, what kinds of maps will most effect the future of maps? Is any configuration not a map? What is the goal of map making? Said another way, which maps built today will effect the shape of the most maps in the future? What can you say about the flow of energy with regard to this hierarchy or range of map influence? Why does energy flow matter to a universe? I am giving you all of the clues. All of them.

Randall Reetz

Kevin,

Please show me a process that isn't evolution at base.

The motions of the stars have little to do with evolution.

I think you have this hundred year old notion of evolution that is bio-centric, sexual-centric, entity-centric, individual-centric, species-centric, etc... and where oh where do you get this "survival of the fittest" statement?

I thought "survival of the fittest" was Darwin himself (but according to wikipedia it's a contemporary: Herbert Spencer), and I have no problems applying evolution in multiple spheres. However, at the moment evolution (genetic programming) doesn't seem to be doing much for software engineering or AI.

Not only is it "useful" to "decide" that all process is evolution, it is the only possible explanation, and "use"less to attempt a project as complex as the "design" of sentience from any other perspective.

In what way useful, and even if it is: what has that to do with sentience?

All of your arguments are built on seemingly innocent and smallish falsehoods ("I suspect... some jumping is required"). In others posts, your innocent mistake is what you simply call "thinking", or "what humans do", or "design".

A suspicion is neither fact nor falsehood, and the height of a metaphorical wall is certainly open to debate. My argument may not be rigorous, but neither is yours.

Yet these black-box secret-sauce ingredients represent the essence of the machinery you claim to be interested in building. How are you going to do this outside of an understanding of the evolutionary process, of why evolution happens? Outside of a profound understanding of what evolution does and more importantly, why it does it?

Evolutionary methods (genetic programming) are a short-cut means where understanding is an obstacle. I.e. given the right goals an evolutionary system can produce results where reductionist methods don't, understanding is not required (that's the point).

If you are stuck on the word "fitness", lets use it. What do you think "fitness" ultimately means? Meaning, there is an obvious direction to change. What accounts for the fact that there was once just the strong force and quantum gas, and now there are things like you and me debating the causal influences at play in the universe?

I'll ask my AI that one later.

Oh, maybe I understand. Maybe you think sexual selection and reproduction are the only means by which evolution happens. Wow, that wouldn't be very evolutionary. A system that moves towards complexity handling capacity, but only so long as there are organisms that swim or run around and have sex? I guess if you think that way, you know the mechanism that biology uses, and by knowing that, you don't really know what evolution is at all. If you believe that evolution is the domain of blindness, you really haven't thought about what an organism is. If you think organisms exist purely to survive as individuals, you really can't explain why complexity handling increases over time. If you can't explain that, you will never be able to build a thinking machine.

Evolution produces things like "thinking machines" without the need for explanation or understanding, requiring a full understanding of evolution in order to use it seems a little self-contradictory.

Why do abstractions exist? What is an abstraction? Is anything, not an abstraction? What does evolution have to do with the building of abstractions?

An abstraction is simply a reduction of distinct cases to a single model for (say) a particular class of problem. A good abstraction can be reapplied in different areas, e.g. object-oriented programming is a fairly useful abstraction. Evolution is itself an abstraction: a model for incremental improvement through random choices.

Why expend the energy to build a map? If you do expend that energy, what kinds of maps will most effect the future of maps? Is any configuration not a map? What is the goal of map making? Said another way, which maps built today will effect the shape of the most maps in the future? What can you say about the flow of energy with regard to this hierarchy or range of map influence? Why does energy flow matter to a universe? I am giving you all of the clues. All of them.

Randall Reetz

You seem lost.

Entropy: go with the flow, time and tide...

Kev.