Artificial inteligence

Čo je to umelá inteligencia?
Niekto navrhol úvodnú definíciu umelej inteligencie (UI, v angličtine AI) ako "UI je pokus prinútiť stroje správať sa ako tie vo filmoch." Toto môže dať istú predstavu o umelej inteligencii, ale vynecháva to najmä jej vedecké aspekty. Žiadna krátka definícia nemôže vystihnúť všetky úlohy a témy UI, preto namiesto definície je lepší iba popis.
UI je pomerne nová vedecká disciplína, vznikla v polovici dvadsiateho storočia. Napriek tomu, že stúpa frekventovanosť jej výskytu v médiách, nie je dostatočne známa. Mnohé vedecké práce sa dokonca medzi širokou verejnosťou nepovažujú za súvisiace s touto témou, aj preto, že sa stali súčasťou softvérového inžinierstva. Preto časť verejnosti považuje tento vedný odbor za zaniknutý.
Niektorí ľudia si podobne, nepoznajúc problém hlbšie, myslia, že čoskoro sa roboty dostanú nad našu úroveň a možno sa ľudia stanú domácimi miláčikmi strojov. Toto však momentálne nehrozí. Stačí si uvedomiť, že pokusy o naozaj dobrú replikáciu zmyslových a motorických schopností sú nie veľmi úspešné, nehovoriac o koordinácii jednotlivých činností a rozumových schopnostiach ako je učenie sa či logické usudzovanie.
Fyzika a chémia študujú hmotu, energiu, sily a rôzne spôsoby ich transformácie. Biológia, geológia, medicína a mnohé iné vedy a inžinierske disciplíny sú založené na štúdiu viac a viac komplexnejších systémov zložených z jednoduchších komponentov. Všetky tieto vedy vyžadujú porozumenie spôsobu, ktorým prírodné i umelé stroje pracujú s hmotou, energiou a ich transformáciami. Ale niektoré stroje, prírodné aj umelé, pracujú s informáciami. Teraz je jasné, že potrebujeme nový druh vedy na štúdium princípov, ktoré určujú, ako sa získavajú vedomosti, volia a dosahujú ciele, komunikuje informácia, dosahuje spolupráca, formujú koncepty, vytvárajú jazyky.
Môžeme to nazývať veda o vedomostiach alebo o inteligencii. O tomto je umelá inteligencia. Nielen o umelých systémoch, ale aj o ľudských a iných žijúcich organizmoch a ich spôsobe manipulácie s informáciami. Bytosti sú informáciami zasiahnuté: šťastné pri pochvale, smutné pri zlých správach, žiarlivé pri pohľade na schopnosti či vlastníctvo druhých, bojazlivé v tme. Preto nie je prekvapujúce, že v uplynulých rokoch skúmanie emócií nabralo aj v UI na dôležitosti. Takto, naproti jej nešťastnému názvu, je o spracúvaní informácie v prirodzených systémoch takisto ako v umelých, a nielen o učení a rozmýšľaní, ale aj o pocitoch.

AI Overlaps with several other disciplines
If we construe AI in this way (as studying how information is acquired, processed, stored, used, etc. in intelligent animals and machines) then it obviously overlaps with several older disciplines, including, for instance, psychology, neuroscience, philosophy, logic, and linguistics. What is new in AI is that the development of computers has given us new ways of investigating the problems. Previously psychologists and brain scientists could only observe and do experiments on existing information processing systems, such as human beings and other animals; and philosophers could only theorise in the abstract about how mind and language ought to work. Now, however, we can go beyond those methods and also "play God", that is we can design new kinds of working systems to demonstrate the implications of our theories and check out whether they can explain the facts they are intended to explain.
That new possibility arises because computers were designed specifically to acquire, store, manipulate and use information, unlike older machines which were designed to transform energy or apply forces to manipulate matter or to produce chemical transformations.
Computers, unlike previous machines, enable us to express our theories about how minds work in the form of working computer programs that enable a machine to do some of the things people do, e.g. to communicate, learn, solve problems, understand input from TV cameras, control artificial limbs, etc.

Some of the things we have learnt
Designing machines with such capabilities has proved far more difficult than many of the early researchers expected. In part that is because many tasks which at first seemed simple turned out to have hidden depths. For instance, seeing is not just a matter of recognising patterns in visual images, but involves making sense of the environment, including understanding all the many ways it can help or hinder us. Similarly, the ability to understand and use a natural language, like English, or French, or Urdu, turned out to be far more complex than some of the early researchers thought. To study and model these complexities, we have had to invent entirely new ways of thinking about the processes involved, including developing new languages in which to express the ideas, such as new sorts of programming languages. (AI researchers often find the languages used by other programmers, e.g. Pascal, C, C++, Java, too restrictive.)

Even "stupid" people have considerable intelligence
This research has helped to reveal a great deal of shallowness in our normal thinking about minds, consciousness, perception, learning, language, and so on. In particular, we now understand that the kinds of people we might normally describe as ``stupid'' are far more sophisticated than any machine we currently know how to design. I know of no robot which could be trusted to clear dishes and cutlery from a dinner table and wash them at the kitchen sink, yet the majority of people can do this, without being specially intelligent. It has proved much easier to design and implement machines which do the sorts of things which we previously thought required special intelligence, like the ability to play chess, do algebra, or perform calculations. These sorts of tasks fit more readily into a computer's mechanisms for manipulating large numbers of precisely defined symbols very rapidly, according to precisely defined rules.
We now understand much better that many commonplace human and animal abilities (e.g. a squirrel leaping among branches of a tree, a bird building its nest, a child listening to a story) involve a very deep kind of intelligence and important and subtle kinds of knowledge, which our theories do not yet accommodate. Likewise animal intelligence includes things like desires, enjoyment, suffering, and various forms of consciousness, all of which play an important role in their information processing, but which we hardly understand as yet.
Many AI researchers are trying to find ways of extending the concepts, theories, mechanisms and models in AI to include all those things. Their work includes trying to find ways of programming computers so that they have the kinds of richness and flexibility required for animal abilities. The design of artificial neural nets, flexible rule interpreters, and various kinds of self-organising software systems are among the approaches being followed.
As mentioned above, some of us are also investigating ways of building AI systems with the sorts of mechanisms involved in motivation, moods and emotions, as well as the more obviously required capabilities like perception, reasoning, problem solving and motor control. (My own work in this area, along with work by colleagues, can be found here.)
It should be clear from all this that insofar as AI includes the study of perception, learning, reasoning, remembering, motivation, emotions, self-awareness, communication, etc. it overlaps with many other disciplines, especially psychology, philosophy and linguistics. But it also overlaps with computer science and software engineering, because it includes the design of new kinds of information processing systems, either to model those in humans or to solve practical problems (e.g. software controlling a robot or factory, or software helping a child to learn about arithmetic).

Brains and computers: AI and neural nets
Some people in AI have been impressed by the fact that the mechanisms of brains are very different in detail from those in computers, even though they may be doing similar sorts of things (storing, transforming, using information). This has led to the investigation of neural nets partly inspired by ideas about how brains work. Some artificial neural nets have developed entirely as practical solutions to engineering problems without much concern for accurate modelling of brain mechanisms. More recent work attempts to move towards more and more accurate models of real neurons, which are incredibly complex and varied.
Some people think that it will never be possible to understand and replicate all the important aspects of brain function unless we replace computers with new kinds of machines, or perhaps build hybrid machines using different technologies. This conclusion is premature. There are two reasons:
We do not yet know what the real potential of computers will turn out to be as we develop new types which are faster and smaller and can be linked together in vast collections of cooperating systems There is much that we do not know about brains: including what they do and how they do it. So we cannot yet say with confidence that there's ANYTHING brains can do which computers will NEVER be able to do, even though there are many things brains can do which existing computers cannot do (and vice versa!).

AI and simulated evolution
Another recent development related to AI is work on simulated evolution. Biological evolution managed to produce an enormous variety of living organisms closely suited to different sets of needs in different environments. By modelling those processes on computers it turns out that we can sometimes get the computers to evolve solutions to problems that we were previously unable to find.
Genetic algorithms (GAs) are increasingly being used both as a research tool and as a means of getting computers to solve practical problems. They use strings of symbols to encode specifications for designs, or solutions to problems, in something like the way biological systems use strings of molecules in DNA. Transforming and recombining portions of strings enables an evolutionary computation to search for good designs or good solutions, in a manner that is partly analogous to biological evolution.
Genetic Programming (GP) extends these ideas by using structures which are better suited to the problem than strings are. For instance, a GP system may directly manipulate tree-like structures representing rules or computer programs.
This work links up with studies in Artificial Life (Alife), which is concerned with simulated evolution of various kinds of artificial organisms, possibly competing or collaborating with one another. Evolutionary techniques may use AI in the systems they evolve. Similarly AI systems may use evolutionary techniques to help with some of the harder problem solving tasks.

Some implications
This new multi-disciplinary field brings together a variety of old disciplines in an entirely new way because we are constantly learning new techniques for building working systems that extend and test ideas and theories synthesised from the different disciplines. This will have increasing practical importance as we continue to develop more and more sophisticated information processing machines performing more and more tasks (at home, at school, in factories, in offices, in hospitals, on the internet...).
This research also helps us deepen our understanding of what we are, how we relate to other kinds of animals and also how we relate to other kinds of machines, including machines of the future, which may become more and more like us. For example, by designing machines with various kinds and degrees of autonomy we can clarify old problems about the nature of free will. Instead of there being one kind of "free will" which you either have or do not have, we find there are many different kinds of freedom, and different people, different animals, and different machines will have different subsets -- different kinds of freedom.
Some of them can then be seen to be more important than others. For instance, it is important to have the freedom to resist external coercion and take decisions and act under the influence of your own desires, preferences, knowledge. It is not so important to have the freedom arbitrarily and randomly to change your own desires, preferences and knowledge. These considerations are as relevant to some intelligent robots as they are to humans and other animals.

AI and Computer Science
How does AI relate to computer science, another new discipline? In part it is like the relationship between physics and mathematics. Mathematics develops many concepts and techniques which physics uses, but the central goal of physics is to understand the world, not to understand those techniques. Likewise computer science (along with mathematics, electronic engineering and software engineering) develops general theories about information processing, and helps to create powerful general tools (e.g. computers, operating systems, and compilers) which are used in AI, but these are not the central focus of AI. The general concepts, techniques and tools produced by computer science are used by AI researchers in the process of studying something else, the kinds of information processing capabilities which we find in many living organisms, and which might also be created in new machines of many kinds. However, just as the history of physics includes many episodes where mathematics was enriched by the work of theoretical physicists, so also has AI had a great deal of influence on the development of computer science. But equally it is having a deep impact on other disciplines with which it is connected, especially philosophy, psychology and linguistics. I think its impact on other disciplines will continue to grow and diversify, including psychiatry, brain studies, biology, and many forms of engineering.