Answer the question
In order to leave comments, you need to log in
What conditions are necessary to create superintelligent systems?
Dear professionals, I am not a programmer, but it became interesting to me under what conditions the creation of artificial subjective systems will become possible.
Subjective systems are those systems that form their own behavior based on their own subjective experience.
Please help me understand if the following conditions are enough to create artificial subjective systems that will not need programming to form their own behavior:
1. Having the ability to choose a set of operating spaces, rewarding functions and goals in such a way that all possible sequences of actions and goals belong to this set?
2. Is it possible to translate external motivation into a sequence of actions to achieve the appropriate state of the system?
3. Is it possible to form your own subjective experience?
What other conditions, in your opinion, are necessary for the creation of such systems?
Answer the question
In order to leave comments, you need to log in
Well, today we have a problem in creating parallel systems with a huge number of connections between them, those are real neural systems in hardware. What is emulated on conventional processors is garbage, what is built through networks is also garbage, everywhere there are bottlenecks that stop the whole process. Today, this TrueNorth
is closest to what is needed , which also does not stand close to our brain yet, and also has a hardcoded neural network.
In short, until we solve the problems of creating a real neural processor with the normal fast operation of the neural network, the required level of parelism and the speed of the neurons themselves, it is too early to dream about it.
At most we can emulate something, but these will only be fakes.
PS There is a second point, and this is a separate big story, you should not focus only on human-like intelligence. For example, the ocean, or parts of it, can collectively be considered a large organism with certain processes, a bunch of organisms, reactions to the external environment.
Or a more interesting example - the Internet, will we be able to notice when the Internet becomes an aggregate meta-intelligence, and we, as its constituent parts, may not notice this (you can call it a new kind of crowd effect, only here the crowd is global).
What conditions are necessary to create superintelligent systems?
You need to first understand the terminology in order to correctly formulate the question.
Subjective systems are systems that have their own evaluation function of external influences and act on its basis. There are many such systems. Bank ratings, autonomous vehicle management, record keeping systems, and so on. In one way or another, they can act as a subject. If they have a parallel learning procedure, they accumulate their subjective experience.
Reasonable systems (we mean that the human mind is the standard) - systems with a functionality equivalent to the human mind, also known as strong AI. The main problem is the complexity of construction. It is necessary to combine and coordinate the set of subsystems that implement parts of the common functionality, as it happened in the process of biological evolution. The project of creating such systems will require more expenditures than the space programs of the Cold War. Ask yourself who and why will pay for this holiday of life?
Super intelligent systems - systems with functionality that exceeds the functionality of the human mind. Any intelligent system on a non-biological basis (for example, based on modern silicon computers) automatically becomes superintelligent due to its physical indicators (more durable and expandable).
Thanks for trying, but I have given the wording of what systems can be called subjective, for definiteness.
One's own evaluative function may well be part of past experience.
In addition, three specific questions were asked and none were answered.
You have something all messy. It is impossible to formulate more in Russian?
I kind of read you, and I don't understand. What do you need?
By subject. Any intelligent system (and not only) must have the following components:
1. Input signals (signals from the outside world of this AI)
2. Output signals (signals of influence on the outside world of this AI)
3. Feedback - any algorithm for changing the behavior of the system to reaching the minimum value of the objective function.
4. An objective function is usually a function that takes the state of the AI world and evaluates it. The goal of any AI (as a control system) is to minimize the objective function.
Simply put, all learning is an interactive process.
Mathematically:
in(t) -- input vector of your AI, depends on time t
out(t) -- output vector, depends on time t
f - objective function
L - learning function
Out - feedback formation function.
params(t) -- some internal parameters of the AI system (ala state), depending on the time t.
Delta = f(in(t), params(t))
out(t+1) = Out(in(t), params(t))
params(t+1) = L(Delta, in(t), params (t))
The goal of any system is to minimize error.
An AI system does not make sense if the goals of its existence are not set.
Didn't find what you were looking for?
Ask your questionAsk a Question
731 491 924 answers to any question