Why “artificial intelligence” and “neural network” are not synonyms and whether Internet may one day begin to figure things out – these questions responsible editor of “Independent newspaper – Science” Andrey Vaganov discusses with the chief of Department of intelligent systems in the humanities of Russian State University for the Humanities, Doctor es Science, Honoured Science Worker of the Russian Federation, Prof. Victor Finn.
Victor Constantinovich, Is there a difference between two nowadays very popular notions of “artificial intelligence” (AI) and “neural network”? Are they synonyms?
My answer will be intricate. Long time ago Francis Bacon defined four kinds of “idols”, hindering knowledge (the science of his time): idols of the theater, idols of the marketplace, idols o the cave and idols of the tribe. Artificial intelligence also has its own idols. In addition to Bacon’s there exists a system of contemporary idols – myths about artificial intelligence.
First of those myths, more general than your question: AI is machine learning. It’s a global myth. There are books entitled just so: “Machine learning – new artificial intelligence”.
Second myth. Solving of problems connected with imitation of human activity within the frame of artificial intelligence is not possible without Big Data.
Third myth is dangerous – it propagates the notion that AI is a sum of special technologies. But for active users of AI methods (including medical men, managers, sociologists, military men, bankers…) would be fruitful to understand that the main product of AI are intelligent systems with problem solver.
And what is dangerous in this?
This myth disorients investors and administrators.
But it was always one of the aims of science, of scientists – to “disorient” people and state structures disposing of financial resources. As Stanislaus Lem remarked, “without doubt, scientists will need at first to “bring up” a whole generation of executives who will agree to get deeply enough into state’s pocket, and besides for the aims so suspiciously akin to traditional science-fiction…”
I think that scientific ethics demands objectivity and estimation of what is possible in its investigations.
Next myth, also dangerous, in my view: is possible quick realization of the ideas of artificial intelligence. Without preparation of specialists, without system of education, without serious enough development of theoretical grounds.
Another myth – that artificial intelligence can be reduced to a set of procedures.
And, at last, a non-trivial myth: human knowledge takes place in the brain. And, consequently, if we will model brain as a mathematical construction, we will solve all problems. If knowledge is in the brain, which we can model, it means that we have reached our goal – create artificial intelligence.
What relation does it have to the problem of artificial intelligence?
From this we have two non-trivial consequences. Firstly, process of acquiring knowledge is social, it is a result of interaction not only with personal system of knowledge, but also with other sources of knowledge.
Secondly, it means that what is now taking place in computer technologies corresponds to new structure of knowing process: parallel computing and computer networks. So it is clear, that this takes place not only in the “brain”, hence it is no neuron network.
By the by, this conclusion helps to clear up yet another myth: that artifitial intelligence is an autonomous system. It is not so. The fact is that the main product of artificial intelligence is an intelligent system which is a partnership-type man-machine system. And from this sprout many things.
First of all it means that is possible an autonomous regime. But it is precisely only one of regimes.
And the second regime is interactive. Man intervenes into the solution of problems, takes part in this process, and so evolves a man-machine system.
Because of this the main task of artificial intelligence is creating partnership-type man-machine systems.
Thus, artificial intelligence is not a neural network and even not machine learning. Nevertheless Internet network may begin to to “figure things out”? In it may emerge its own reflexion?
No, it may not. The network may engender a problem, set up many-many tasks, because it itself is a new reality.
But it, the Network, is always complicating…
And will always set new problems.
In other words, this emerged new world of the Internet cannot perceive man as some “allergen”?
No. it can not happen. But, speaking of Internet, we must understand, that it is indeed a certain autonomous environment. Among other things, there emerged evil. And very devious evil (hacking, etc.).
After some iteration all this comes down to man…
Of course. And there is no escaping. Even Internet “pestilence” is the result of activity of man-machine systems.
The fact is that artificial intelligence itself as a line of research is not still accomplished / fulfilled. Accomplished as a research direction with its own developed conceptual system and methodological foundations.
And cybernetics is not a science. It is a scientific movement. The idea of cybernetics is parallelism between man and machine, idea of control and feedback. Nothing more.
Artificial intellect has no relation to cybernetics?
There is a certain historic continuity. Artificial intelligence and cybernetics are kinsmen. Idea of AI is to a great degree connected with the advent of computers. It seems that we may regard Alan Turing as forefather of artificial intelligence. Although the main Turing’s idea was limited (his famous test).
What is then “artificial intelligence”?
We must define more exactly each word used in this term. “Artificial” means using means using mathematical, algorithmic, computer means. But with “intelligence” all is not simple. It is this that we must make quite exact, and only then it will be possible to speak of the sphere of application and nature of AI as a research direction.
We need to begin with preliminary definition of capabilities that are to be imitated and strengthened. They are not necessarily to be exactly human capabilities. They may be imported from “world three”.
That means that we may even not realize that artificial intelligence had appeared?
Of course. As a result of development of science we may come to the necessity of including into our conception of intellect certain additional capacities.
Which capacities, for example?
I will enumerate them.
In my opinion, physics comes under this definition: it has an abundance of data, but concepts often aren’t clearly cut.
Physicists have mathematical apparatus and experimental base. Historians have not, sociologists have not…
But why? Historians have a huge factological base and many specific procedures of historic investigation…
They are not formalized. There is no language. Physicists have such language. The main problem of the humanities is lack of means of presentation of knowledge and of logical systematization in these sciences. Just because of this they have fell behind in the possibility of constructing theories and realize reasonings. It does not mean that there is no deepness in the humanities, but this is deepness at the expence of content-richness of ideas. We are coming to an important idea. In fact, to knowledge belong to worlds – the world of mental states, which we wish to imitate, and the third world, into which we are invading, wishing to transform knowledge. Consequently, it is necessary to define, what is the thought process? And it is intention (questions, attitudes, imperatives, goals), further – search for premises, relevant to the goal; reasoning and, at last, reflexion.
And what is the knowledge process? It is discovery of knowledge, prediction of knowledge, acceptance of knowledge, integration and growth of knowledge.
Thus, we can say that a system is intelligent, when we may have an increase of knowledge through realization of all of the above. And now we may come to the definition of what artificial intelligence is.
AI is scientific direction dealing with imitation and strengthening of cognitive activity of man through computer systems.
Another important consequence of our definition of AI: systems of artificial intelligence are computer systems applying a certain method.
Lastly, intelligent systems. Such systems have more complex architecture: base of facts – base o knowledge – problem solver –comfortable interface.
The main task is generation of empiric regularities. If we have achieved this goal – it means that we have some acceptable stage of our investigations. It is a strong criterion of demarcation between scientific and non-scientific knowledge in addition to Popper’s criterion of possibility of falsification (refutation) of a theory.
Lowering a little logico-philosophical bar of our conversation, I wish to ask you the following question. Proceeding from definition of AI, given by you earlier, may we say that artificial intelligence only “mimics “, “simulates” natural intellect?
Among 13 capacities enumerated above it’s possible to mark out some that we realize in automatic mode, when after startup program works without human participation. For example, in automatic mode we can recognize essential in data or realize argumentation.
Another product of artificial intelligence beside systems of AI are robots. Intelligent robot is an intelligent system plus sensory unit plus mechatronics. If we take an intelligent system and add to it the sensory unit, we have a cognitive system. It forms its own base of facts as a result of interaction with the environment.
It is essential that intelligent robot may be anthropomorphic. It can act autonomously or partnership-like. But central idea is its having an intelligent system. And the problem of reasoning robot as yet is not solved, although mechatronics and software of such robot may be on fantastically high level. Anthropomorphic robot can already make salto-mortale. But it has no logic of reasoning.
You spoke about intellectual capacities. In one of your texts you said that “it is impossible to replace creativity with machine”… But in year 2016 took place an epochal event – AI defeated man in a game of “go”. Game that has more variations than atoms in the Universe, – 10100 (to compare: the number of possible positions on the chessboard is: in checkers – 1020, in chess – 1060). “Brut-force” calculation (by direct enumeration of variants) in “go” doesn’t work already after two-three moves (in chess after fourth move there is more than 100 thousands possible positions, in “go” – more than 16 billions!). Player’s intuition here is more important than calculation of variants.
In the AI program that won a game of “go” against man there was no intuition! The victory was achieved thanks to the possibility of surveying very big data, where man cannot compete with machine. Somewhat different effect, connected with combinatorics of variants, enabled chess program Deep Blue in its time to outplay Harry Kasparov.
But Deep Blue used another scheme. It sorted out variants very fast.
All this is not so simple. This program previously had examined Kasparov’s parties and knew how to play with him.
Besides, both chess and “go” are not very good examples. Thanks to one simple cause: in these games there are rules. And computer programs are in fact realization of these rules. It is not at all some miracle. Where the task is reducible to combinatorics or to the survey of big data, machine will always defeat man. But revolt of artificial intelligence is a pseudoproblem. Artificial intelligence is completely secure.
Another of your citations: “That supercomputer some time will leave man behind – is fib”. Victor Constantinovich, wherefrom such anthropic chauvinism?
From a very simple cause. This conviction issues from my understanding of intelligence. Machine cannot have intentions, it cannot have intuition or initiative, it has no imagination, it cannot be a creator, but can be his powerful partner.