On 11/01/2018 at 11:48 PM, Awoken said:
If we create a robot to visually respond to the color blue, we are not also creating a robot to experience the visual impression of said color, or for it to experience anything at all, all we are doing is getting the robot to respond to the a particular wave length and then execute some code. Big distinction must be made between these two points, if to you, you see no difference between the two, then this is where you and I diverge.
This is actually what is at the center of the recent A.I. or "Deep Learning" revolution. You are not longer triggering code from specific events.
Rather: A modern neural network is something that is not programmed (in the traditional sense). You supply it with some inputs, and you give it a goal. It then goes on to develop (or program) all the skills that it needs by itself. You are not in control for how it achieves it's goal.
An example (that has not really happened yet): You give a swarm of neural network robots a useful goal like: "Clean the streets".
This is a very simple goal...
Their inputs are: A camera, and some form of radio communication to coordinate with eachother.
You don't tell them what dirt looks like, and you don't supply a communications protocol.
You do grade them by how clean the streets are in the morning and how much fuel they've wasted.
Years pass, and the robots gradually get better.
The thing to be expected, is that they learned what dirty things look like .
However, you find that the communications protocol they are using is very similar to a spoken language. These robots actually learned to speak, and further more invented their own language.
Now it's true, intelligence of such magnitude out of reach of today's neural networks. However the principle is the same: Machines are responding to stimuli in ways which they were never programmed to do.
This is why "Alpha Go" and "Alpha Zero" are so special: Unlike deep blue, and previous A.I: Nobody ever taught them how to play "chess" or "go".
One could claim that this is the start of "creativity" in machines. And putting "consciousness" aside for a minute. This is the big deal with the latest developments. We have machines which can be "creative".
As a side note what I find even more threatening is: Unlike "Alpha Go" which was playing a human, "Alpha Zero" was playing the best computer program that many humans have studied and researched. I don't think there exists a game that is more studied than chess. And the shear amount of computer science skill directed at building chess engines dwarfs any other gaming AI research. And alpha zero beat all of that human effort, without building on top of it. (It learned how to play chess from nothing). Basically it means: There will soon come a time when computers will be able to write better computer programs than humans. Even worse, these programs will be so complex that they will be beyond human understanding (you will not even be able to help debug them...). I don't know about you, but as a "traditional" software dev, this makes me nervous about my job security.