This isn't really true though and greatly understates how amazing this demo, and current AI actually is.

Saying the agents are obeying a set of human defined rules / freedoms / constraints and objective functions would lead one to imagine something more like video game AI.

Typically video game AI works on a set of weighted decisions and actions, where the weights, decisions and actions are defined by the developer; a more complex variation of:

if my health is low, move towards the health pack,
otherwise, move towards the opponent

In this demo, no such rules exist. It's not given any weights (health), rules (if health is low), nor any instructions (move towards health pack). I guess you could apply neural networks to traditional game AI to determine the weights for decision making (which are typically hard coded by the developer); but that would be far less interesting than what's actually happening here.

Instead, the agent is given a set of inputs, a set of available outputs, and a goal.

4 Inputs:
- Position of the agent itself
- Position and type (other agent, box, ramp) of objects within a limited forward facing conical view
- Position (but not type) of objects within a small radius around the agent
- Reward: Whether they are doing a good job or not

Note the agent is given no information about each type of object, or what they mean, or how they behave. You may as well call them A, B, C rather than agent, box, ramp.

3 Outputs:
- Move
- Grab
- Lock

Again, the agent knows nothing about what these mean, only that they can enable and disable each at any time. A good analogy is someone giving you a game controller for a game you've never played. The controller has a stick and two buttons and you figure out what they do by using them. It'd be accurate to call the outputs: stick, A, B rather than move, grab, lock.

Goal:
- Do a good job.

The goal is simply for the reward input to be maximised. A good analogy is saying 'good girl' or giving a treat to a dog that you are training when they do the right thing. It's up to the dog to figure out what it is that they're doing that's good.

The reward is entirely separate from the agent, and agent behaviour can be completely changed just by changing when the reward is given. The demo is about hide and seek, where the agents are rewarded for not being seen / seeing their opponent (and not leaving the play area). The agents also succeeded at other games, where the only difference to the agent was when the reward was given.

It isn't really different from physically building the same play space, dropping some rats in it, and rewarding them with cheese when they are hidden from their opponents - except rats are unlikely to figure out how to maximise their reward in such a 'complex' game.

Given this description of how the AI actually works, the fact they came up with complex strategies like blocking doors, ramp surfing, taking the ramp to stop their opponents from ramp surfing, and just the general cooperation with other agents, without any code describing any of those things - is pretty amazing.

You can find out more about how the agents were trained, and other exercises they performed here:

https://openai.com/blog/emergent-tool-use/

Neural Network Prototyping On the Go has been added as a related post - related requested by ChaosEngine on that post.

I think one of the key things is that it was filling in the eyes with eyes. It was using completely different color eyes even and it knew where they needed to go. Content Aware only uses what's in the image, so it would just fill in that area with flesh and random bits of hair and mouth. This seems to pull from a neural network database thingymajigger.

Your video, Neural Networks for Character Control, has made it into the Top 15 New Videos listing. Congratulations on your achievement. For your contribution you have been awarded 1 Power Point.

For the time being, I don't think that's a concern; with all of the incredible progress being shown in using neural networks to replicate (and yes, improve on) human skills, I'm not aware of any real advances in any field related to giving AI any sort of actual will.

What's more concerning to me is how this sort of technology will be put to use to categorize and control people. I'm not talking about shadowy cabals, or even evil corporations--I'm talking about the unintentional consequences of being able to accurately reduce people to a set of metrics and predictable behaviors, and how that may push culture to a bland algorithmic mush rather than the chaotic but vibrant human mess it's always been. Time will tell, I guess.

That's a shame, I thought it was a quite interesting explanation of all those Google deep dream images.

There was another computerphile video just on the neural networks used, linked in the description, but I didn't watch it.

The most basic idea is reasonably straightforward - the neural networks are being used to classify images, so there is a low level categoriser for low level things like edges and corners, and then a higher level one that looks for how edges and corners are arranged to make, say, ears... and then a top level one to look for how ears and noses are arranged to make cats.

The complicated bit is that then they run the device backwards, so instead of using it to assign a probability that something is an ear, they actually put the ear in the image even though it wasn't really there to start with.

Since I'm not really saying anything that they didn't, I'm assuming that didn't help?

Well said,

It could be that the very existence of the future of evolution is already "installed" genetically in human brains. Taking this substance might be exposing these neural networks that seem like our technological transcendence into the Universe.

The Status Quo would not want anyone to know and would deny it exists, and as I have found by talking about it - people will straight up make fun of you. Until they do it, and then they apologize profusely!

>> ^gwiz665:

It's dead mate.


works for me. Maybe check your computers gel packs to make sure the neural networks are not burned out.

feeling, memories are registered in the brain, not necessarily stored. the entire body is a neural network that we are only beginning to understand....
the whole idea of re-animating dead flesh seems a bit far fetched. it seems to push the limits of the existing dna. cloning with fresh dna...maybe. re-animating frankenmom, unlikely.

>> ^Almanildo:
Is there any reason to doubt his awareness now? I can't see any.

Scientific reductionism is unscientific.

As far as we can tell, my consciousness exists because of the unique configuration of everything at the time of my birth. There's no way we can scientifically eliminate factors and test to see if I would still be the consciousness that was brought into existence. If we ask what factors were necessary to bring me into existence, we have to say "everything."

If we conceive it is possible to artificially create consciousness, then we have some crazy scenarios to imagine. For example, "pulsing" a consciousness on and off in rapid succession and giving it no memory so each entry to the world is its first. Is it the same consciousness? Is there any suffering? Is there any change to the state of a consciousness without the necessary cues of memory and senses? Is there no additional energy cost to producing a million consciousnesses inside a neural network versus just one?

What is consciousness if it is completely textureless and featureless, depending entirely on its host brain for all its function? The answer "an illusion" has to be disqualified because this "illusion" is at the foundation of all we know and experience. There are illusions in our conscious experience, but if it were entirely an illusion there would be no reason for anyone to be experiencing it. The term "illusion" itself requires a viewer to be experiencing it, who presumably would be conscious as well.

I have no doubt science will inform us more on this subject, but if I had to predict it, I would say that reductionism just won't do. Consciousness must be some kind of energy, and follow similar laws of other energies, and not just be an infinitely reproducable phenomenon. If it is infinitely reproducable, then everything in existence has a consciousness value and ours is just higher or more concentrated. One of those two.

I'll try to approach this from another angle.

I know that I am 'aware'. That is, not only do I behave as though I was a concious being with self-awareness, I am indeed aware of what's going on. I don't know, however, whether any of you guys are. You certainly behave like it, but there is no way to know for sure.

However, there is no reason to believe otherwise. From an objective point of view, there is nothing about any other human being that is fundamentally different from me, therefore I have no reason to believe that any other human being is not 'aware', just like me.

Now, here's a thought experiment: Make a brain in a vat. Take a live person, somehow extract his brain from his body without killing the poor guy, and place it in a vat. Attach artificial sensory organs and life-support systems, such that all the information that flowed between the brain and the body now flows between the brain and an artificial machine. In the traditional philosophical experiment, the person shouldn't notice any difference, but that's not really important for my purposes. The important thing is that he can experience the outside world.

Is there any reason to doubt his awareness now? I can't see any.

Now continue to mess with our poor victim. Replace his prefrontal cortex with an equivalent ANN (Artificial Neural Network), made out of electronics. Whether or not that's possible, here's the next step:
Abandon the vat and simulate the entire configuration on a computer.

Now we have an ensemble of electronics instead of an ensemble of neurons. Still, the guy's behaviour is essentially the same. The objective facts can be correlated with the facts about the original person. He still has a prefrontal cortex, it just exists in computer memory instead of in a real configuration of neurons.

I still can't find any reason to doubt whether this person is aware. This seems to demonstrate that it's not the 'stuff' you're made of that's important; it's the abstract configurations between the functional units, no matter how the units themselves are manifested in reality. And that seems to argue against any important ontological difference between people and other things.

Sorry it took so long to respond, I had a busy weekend.

They are not simple probabilistic events, and they are operating off the same basic principles, that does not mean that systems do not have qualities which their component parts lack.

Does a piston have the capacity to convert petrol into kinetic energy? Does an internal combustion engine have this capacity? Which part of the engine imbues it with this power?

Systems are qualitatively different from their component parts, and some sets of systems, such as systems which decide, are qualitatively different from systems which don't

I'm going to need a definition of "decide" I suppose. It seems like you are dancing around these squishy intuitive concepts instead of having a specific physical distinction to point out. The amoeboid is composed of a lipid bilayer membrane riddled with intricate protein micro-machines that detect changes in the environment, and behaviorally compensate. To discount the intricacy of the mechanisms of genetic expression and chemical signaling that exist even in the simplest of eukaryotic organism is foolish IMHO. Many of the modern models of genetic expression, and compensation for environmental factors look strikingly similar to the connectionist network models of the brain. The computations are similar in the abstract.


You are anthropomorphizing the mold, it does move, this motion increases its chances of finding food, it survives/reproduces. It in no way displays evidence of doing any of this "in order" to accomplish some goal. If you want to suggest that evolution, as a system, displays intelligence, by selecting molds which move in certain ways, I would be willing to acknowledge that intelligence, not a consciousness, but an intelligence.

Well, more likely I'm moldopomorphizing us. What goals do we have that are ultimately distinct from survival, reproduction, and the general continuity of our species? Even something as seemingly unrelated as making music, or art could be cast as some sort of mating ritual. When you somehow separate our behavior from the rest of life on Earth it's as though you want to draw a barrier between us and them. You want to somehow separate us from the natural order. I hate to break it to you, but it just isn't so. We are just demonstrate the spatial heterogeneity of the second law of thermodynamics.


Why is context necessary for experience? What do you experience in infinitesimal time? Why should we posit some sort of experience which is entirely distinct from the type we claim to have?

I experience the moment. In fact, that's all I'm ever experiencing, although my sensation of it may run a little behind. I never experience my memory, I merely compare my experience to memory. Further, what I'm suggesting is not entirely distinct from any experience we claim to have. Some autistic individuals, for instance, report an extremely chaotic existence, in which causal models can't be formed as sensory modalities are not unified in the same way as ours. They are experienced as independent inputs, not reflective of a coherent physical world. Still, they experience it.

Physical laws are not obeyed, they are enforced. electron movements are completely deterministic, like billiard balls, they roll down hill, they don't decide if/when to do so.

Things can not be enforced without an enforcer. Further, as you've conceded the determinism of our brains, again, how are we not passively allowing the laws of nature to push us around? What exactly are we deciding?


I don't believe that you are claiming that electrons have tiny field sensors which feed into a neural network which analyzes them for patterns and then attributes meaning to them by comparing them to earlier similar sensation patterns. Perhaps you can state this more clearly.

No, I believe that by some other physical mechanism, likely involving quarks and particle physics that I admittedly have a poor understanding of, the electron receives information from not immediately proximal locations, and physically displaces itself to a location with more desirable properties given its current energy state. I don't see how that's different than cuddling up to a warm fire.


You seem to be positing that the structure of the universe is not topological, but that it is instead the consequence of 10^80 atoms all working on concert to decide what the laws of the universe are at this moment. If this is your thesis I am inclined to ask on what basis you think it is even vaguely likely that they would came to a consensus, such as they must to allow the functioning of a universe like ours.

Something like that , although I still don't like the word decide. I don't necessarily think they do come to a consensus. It's just that, as with an attractor network, or similar guaranteed convergence dynamical systems, certain macroscopic states are just more likely than others, despite chaos at the subordinate level. The reason I'd rather drop the word decide is because I don't necessarily want to open the door to something like free will. To cast it in a "God" metaphor, I imagine more of an omniscient God, than an omnipotent God.


Please provide some basis to believe that there is a phenomenal experience.

I can't other than to refer you to what I presume you to have. I could suggest focussing on your breathing, or what have you. I can point you towards literature showing that people that claim to focus on their consciousness can perform physical feats not previous considered possible (for instance monks rewriting the books on the physical tolerance of the human body to cold). Otherwise, I can't. I will say this, however, I take it to be the atomic element of inductive reason. The natural "laws" you are taking as primary are secondary. There is a simple reason for this as Alfred North Whitehead pointed out. If suddenly we were to observe all bits of matter floating away from one another, and were to confirm we were not hallucinating, and perhaps have the experience corroborated by our colleagues, it would not be the experience which was wrong, it would be the laws of nature. Experience has primacy. Matter is merely the logical consequence of applying induction to our particular set of shared experiences.


And that will persist as long as we are not talking about anything. You say "X exists". I say "What is X?". You say "You can't disprove X". And here we are talking about nothing.

I told you, in the best english I can, what X is. It's the qualia of phenomenal experience. Now I can't provide you with direct evidence for it, but I can tell you that nearly everyone I talk to has some sense of what I mean.


You must be using an alternate form of the word "believe". How can someone believe something, and simultaneously be completely unwilling to assert that it is a fact?

I take the Bayesian sense of the word. All probabilities are subjective degrees of belief. I adopt this degree of belief based on anecdotal experience and generalizations therein. None of this would be accepted as evidence by any reviewer, nor should it, and thus I wouldn't want to risk my credibility by asserting it as fact. I can believe some hypotheses to be more likely than others on the basis of no evidence, and in fact do all the time. That's how I, and all other scientists, decide what experiment to run next. I should not, however, expect you to believe me a priori, as you may operate on different axioms, and draw from different anecdotal experience. Thus, I would not feel compelled to assert my beliefs as fact, other than in so far as they are, in fact, my beliefs.

>> ^bmacs27: How are our actions not "probabilistic events?" The amoeba is operating off the same basic principals.

They are not simple probabilistic events, and they are operating off the same basic principles, that does not mean that systems do not have qualities which their component parts lack.

Does a piston have the capacity to convert petrol into kinetic energy? Does an internal combustion engine have this capacity? Which part of the engine imbues it with this power?

Systems are qualitatively different from their component parts, and some sets of systems, such as systems which decide, are qualitatively different from systems which don't

It's moving matter in order to seek out food, and even flexing its pseudopods along the shortest path between food sources in proportion to their delivery frequency.

You are anthropomorphizing the mold, it does move, this motion increases its chances of finding food, it survives/reproduces. It in no way displays evidence of doing any of this "in order" to accomplish some goal. If you want to suggest that evolution, as a system, displays intelligence, by selecting molds which move in certain ways, I would be willing to acknowledge that intelligence, not a consciousness, but an intelligence.

Why is memory necessary for experience?

Why is context necessary for experience? What do you experience in infinitesimal time? Why should we posit some sort of experience which is entirely distinct from the type we claim to have?

Electrons are "comparing" electric fields when they settle into a state, otherwise they couldn't obey their physical laws.

Physical laws are not obeyed, they are enforced. electron movements are completely deterministic, like billiard balls, they roll down hill, they don't decide if/when to do so.

As far as I'm concerned an electron is sensing an electrical field in the same way I am sensing visual band EM.

I don't believe that you are claiming that electrons have tiny field sensors which feed into a neural network which analyzes them for patterns and then attributes meaning to them by comparing them to earlier similar sensation patterns. Perhaps you can state this more clearly.

I just believe that consciousness is a fundamental property of matter.

You seem to be positing that the structure of the universe is not topological, but that it is instead the consequence of 10^80 atoms all working on concert to decide what the laws of the universe are at this moment. If this is your thesis I am inclined to ask on what basis you think it is even vaguely likely that they would came to a consensus, such as they must to allow the functioning of a universe like ours.

It's the sheer fact that there is a phenomenal experience, not the particular nature of those phenomena.

Please provide some basis to believe that there is a phenomenal experience.

You've presented me no evidence that I should only expect phenomenal experience in a complex organism, as you have no test for phenomenal experience.

And that will persist as long as we are not talking about anything. You say "X exists". I say "What is X?". You say "You can't disprove X". And here we are talking about nothing.

"I believe that P(X) > P(!X)". Something you shouldn't really care to contest,

You must be using an alternate form of the word "believe". How can someone believe something, and simultaneously be completely unwilling to assert that it is a fact?

>> ^flavioribeiro:
My point is that current AI doesn't scale. One can't get the neural network used for image recognition and propose that human level AI is just a matter of using more nodes, because it isn't. And I have yet to see a proposal for useful (i.e., robust and expressive) knowledge representation that don't have exponential requirements (or worse).


The idea of a computer being better than a person at everything is a while off, but that doesn't need to happen to cause social and economic problems. People aren't hired because they are good at everything, they are hired because they are very good at specific things, and that is something that computers will be able to replace. Price/performance will be the main decider of what jobs remain on the human end.

Either way, I don't forsee the number of available jobs keeping up with the number of people, especially as medicine continues to advance. I can't think of a reason not to expect more widespread unemployment even if computers stay exactly where they are today.


The singularity you mention is a very interesting subject in itself, but that is talking about a time well beyond the advances which will further limit the number of jobs that require people.

>> ^Psychologic:
One of Kurzweil's inventions was a computer program that taught itself how to distinguish between pictures of dogs and cats. No one programmed it with any description of either. This is built into a camera that he has demonstrated in public. You can take a picture of the page of a book and it will read it to you as well.

Both fine examples of pattern recognition, and featuring no automated reasoning whatsoever. The current state of AI was well predicted in the 70's and 80's, because most ideas used today were actually conceived back then (or even before). They just weren't implemented because the hardware wasn't fast enough.

My point is that current AI doesn't scale. One can't get the neural network used for image recognition and propose that human level AI is just a matter of using more nodes, because it isn't. And I have yet to see a proposal for useful (i.e., robust and expressive) knowledge representation that don't have exponential requirements (or worse).

There has always been a huge amount of handwaving in the AI scene. The bleeding edge of today's AI is basically an improvement of what began in the 70's (for example, regarding detailed specialist systems for medical diagnostics, or theorem provers for assisting professional mathematicians). The human-level part remains very unclear, and academic researchers are very conscious of how little is known. However, this doesn't stop "futurists" from making wild predictions about the singularity and what not.