What can we learn from AI-driven imagination?
Here is one of the latest demos of AI creating photo-realistic images based on very simplistic input. I find this pretty incredible. A lot of this technology is based on our understanding of how human brain works (a.k.a. neural networks), although the current state of technology is very far from replicating the inner structure of the brain.
Seeing videos like this makes my brain work because it erases the boundaries of what I knew only human mind was capable of. What conclusions can we draw from it?
Personally, I think that this technology replicates the 'mindless' parts of the human brain. In other words, imagination is a complex computational process, but it doesn't contain experience in itself. Experience is created when the result of computation is presented to our mind. Thus, we can imagine something and experience it internally, or we can delegate imagination to the AI technology but keep experience to ourselves.
Humanity has made a remarkable progress in creating mindless brains.
Seeing videos like this makes my brain work because it erases the boundaries of what I knew only human mind was capable of. What conclusions can we draw from it?
Personally, I think that this technology replicates the 'mindless' parts of the human brain. In other words, imagination is a complex computational process, but it doesn't contain experience in itself. Experience is created when the result of computation is presented to our mind. Thus, we can imagine something and experience it internally, or we can delegate imagination to the AI technology but keep experience to ourselves.
Humanity has made a remarkable progress in creating mindless brains.
Comments (22)
The problem I see with all of it, old fogey that I am, is that it becomes impossible to distinguish image from reality. Reality is, apart from anything else, painful. It is bloody, it is treacherous, it changes continually, and the pain that accompanies it is real. Whereas in the VR and AI worlds, there is no possibility of real pain, only simulation, and the difference is fundamental, but apparently not discernable to a great many people.
I'm reminded of Baudrillards' simulcra:
Quoting pfirefry
Right - it is something that is characteristic of beings. I think it's great you make that observation.
There is nothing computed in the brain. The brain, contrary to computers, offers an active medium for the physical world to project a physical state in, by means of the body that lies in between. Litterally every physical process has a potential analogue on the neuron network, which can run around autonomously or resonate with a physical process.
This is just programmed virtual reality. No hyper reality is involved.
Done and did.
Then the dream is the hyper hyper reality.
Evidence?
I think you have to create a brain for that. Which we can't. An amazing piece of AI nevertheless!
I'm not sure if this is the same thing you are talking about.
We've been talking in recent weeks about art in different threads. One thing that struck me is that a lot of art is expressed within tight constraints of the rules and technical difficulties of the discipline and tradition. Working within those constraints requires subtlety and awareness. It provides a frame and structure. Seems like something similar probably takes place in life in general. Life without constraints is something other than human, for better or worse. I'm retired and I know I have to be aware of the attractiveness of sitting around doing nothing but reading and participating in intellectual video games like the forum.
I'm saying “replicates <…> brain” because the fundamental mathematical model was influenced by how brain neurons work (each neuron receives multiple inputs and then emits own signal). "Replicates" also refers to humans using their brains to imagine complex things based on simple cues. Although we aren't always able to create vivid pictures in our minds, we're still capable of drawing detailed paintings or unconsciously filling gaps in our vision (blind spots).
By 'mindless' I mostly mean not displaying the faculty of will. As a program, it just does what it's being designed to do. If it was considered mindful, we would have to deal with an ethical issue, but that's not the current state of affairs. Although imagination is a faculty of mind, it doesn't seem to create a mind by itself.
Come to think of it, AlphaGo was the first computer Go program to beat the world's best professional Go players just about five years ago. It used a deep neural network to form an 'intuition' about the best moves for a given board and the winning percentages of those moves. By 'intuition' I mean that it didn't receive instructions about how to evaluate moves, but instead it observed a lot of Go matches and learned to predict what moves a human would do on a given board or how likely they are to win. Then it used a Monte Carlo tree search algorithm to find the most optimal move.
Are we allowed to say that AlphaGo had the faculty of will under the constraints of the game of Go? Perhaps we are, but it still seems far-fetched to say that AlphaGo had a mind. Nonetheless, I'm not opposed to the idea that some day combining mindless AIs with different faculties will result in a mindful system. If so, then it's just a matter of time.
It appears like you're drawing a significant distinction between computers and the physical world. Programs don't fly and brains don't compute. I'm not aways following, but at least I can see a pattern.
Quoting Raymond
How is it that every physical process has an analogue on the neuron network? Are you referring the neuron network of the brain, or the artificial neural network, or both?
That would indeed be interesting! Let's keep the artist.
1. Partial Cartesian : Yuval Noah Harari calls them "imagined orders" e.g. religion, culture, governments, capitalism, etc.
2. Full Cartesian: The universe, as a whole, is a simulacrum.
What we have on our hands is the possibility of a simulacrum within a simulacrum.
Pain has been, allegedly, the gold standard for reality checks. However, there's no necessity for pain at all. Does a plane going down - the cacophony of beeps, flashing warning lights and all - feel pain? You might need to unpack pain to get a handle on what I mean.
I suppose with so much money being poured into pain relief & analgesics selling like hot cakes, what we truly, really want is to transform reality (pain-ridden) into a simulation (pain-free). Heaven could be an anticipation of virtual reality (some aspects of it).
There's more...
I'm off topic.
The brain. There is no program stored in the brain. There are no commands, executed on the rythm of a (computer)clock, that tells the electric pulses on our neural network how to run on the circuit (by applied voltages). The pulses run autonomously, stimulated by our senses, and these follow the lightning like axons, to encounter more or less resistance at the synapses (a broader synapse offers less resistance, ie the connection is strengthened). A parallel bundle runs rather on its own on the neuron substrate, given direction by the connection strengths. Every physical situation we encounter there is a corresponding path of collective parallel neuron pulses, depending on how our body is situated relative to the process. Connection strengths are enforced by being in more or less the same situation often. That's memory forming. If you find yourself in a similar situation, bang! Recognition. The process falls in the path, so to speak. No program involved. If you calculate possible paths, the number is astro astronomical: 10exp(10exp30)! More or less... A 1 followed by 10exp30 zeros... Damned, that's a lot!
That is a good point. In most of the modern video games it's not possible to lose. If you die, you just start from where you were a minute ago. But I know a few games in which people invest hundreds of hours and there is a real possibility to lose all progress in a matter of a moment. These games are not hugely popular, but the people who play them achieve unbelievable results. It is the possibility to lose everything that adds weight to the time that they spend in those games. Losing progress can be painful. The fear of death is the fear of losing years of progress.
I think you mean hell... otherwise :up:
The principle behind it is very simple - an algorithm compares any given image with set patterns and shapes that are saved in a database (filling this database with information is the "machine learning" part). In this case, the program compares the shapes you draw with a database of shapes that are equally as rudimentary as the ones you draw. This database of shapes in turn are linked to a database with real life references, which are used by an algorithm to fill out the exact shapes you drew.
It's cool as a tool - but I don't find the principle behind it particularly impressive. If anything, it showcases how simple function can lead to impressive results.
I don't doubt that the way the brain works has similarities to neural networks constructed by humans. I was questioning the specificity of your comment:
Quoting pfirefry
Seems to me; and no, I don't have evidence; that it would model the whole brain. I can't tell if you say "mindless" because you want to leave a door open for consciousness to be something else.
One thing that I know about these systems is that they model the brain mathematically rather than biologically. The biological inspiration is how neurons connect to each other, but the mathematical model is just a matrix A, where an element a[sub]i,j[/sub] is a real number representing the connection between the neuron i in one layer to the neuron j in the adjacent layer. What follows is a mathematical arrangement of neurons into layers that models the whole computation as series of matrix multiplications with a few other simple functions in-between, like the sigmoid function.
That's where the analogy with the brain ends, and we start 'arranging' neurons in the structures that are easy for the computer to work with. We end up with perfectly lined up grids of neurons that are more mechanical than biological. The brain has a complex structure with different parts performing different functions. To my knowledge, the existing artificial neural networks unable to replicate this biological complexity of the brain. This would require orders of magnitude more data points, and the calculations would be drastically slower because they wouldn't line up with the architectures of the modern electronic circuits.
I suspect that as we progress closer towards the possibility of simulating the human brain, we might hit a physical limit of how many computations we perform in a unit of time, rendering it practically impossible to replicate the inner working of the brain in a digital simulation. So instead of digital we would need to go analogous.
Quoting T Clark
Yeah, I think I'm being intentionally vague to avoid trapping myself in wrong terminology. One point that I'm trying to make is that modern AIs don't seem to show the signs of a living creature's mind. I think we would know if we started sensing empathy towards them.
I'm also cautious about turning this in a discussion of [hide]whether a simulated brain is capable of having a subjective experience[/hide].