What is artificial intelligence?

Question

1. What is the definition of artificial intelligence?

The question at first glance seems unwilling to search around, but a read on the first links I found on Google left me with more questions than I had before I started studying on the subject.

To Wikipedia gave me some definitions and the Tecmundo that was the first link that the search engine returned to me did not pass me credibility by not citing sources.

2. From a definition like this:

"systems that think like a human"

It would be correct to say that every system that claims 2+2=4 is a system with "artificial intelligence"?

Answer 1

A brief introduction

The first work today recognized as belonging to the area of Artificial Intelligence (AI) was performed by Warren Mcculloch and Walter Pitts in 1943. This work was based on three sources/base subjects:

• knowledge of the basic physiology and function of neurons in the brain (a subject originating in Cognitive Science)
• the formal analysis of propositional logic (a subject originating in Philosophy and Mathematics)
• and the Theory of Computation of Alan Turing (the subject that gave rise to computer science)

But the term "Artificial Intelligence" was only coined in 1956 by the computer scientist John Mccarthy, at a conference that also had the participation of Marvin Minsky (co-founder of MIT and a leading theorist of AI). Mccarthy was very interested in logic, a branch that is intuitively linked to the manipulation and symbolic reasoning of humans, and so had the interest to study how aspects of intelligence (as creativity, self-improvement and use of language, for example) could be reproduced in computer systems (consequently, in a "artificial form").

For more details on the history of AI, see Chapter 1 of the book Artificial Intelligence: A Modern Approach, of Stuart Russell and Peter Norvig. This is the book most used in AI courses in leading universities in the world (and my suggestion to who wants to delve into the subject), and its authors are distinguished - especially Peter Norvig, who is currently Director of Research in Google Inc.

It should be possible to realize that in the early days of AI there was a great concern for the understanding and reproduction of intelligence in humans. However, a first great difficulty is precisely to define what intelligence is. On the one hand, one can say that intelligence is the set of cognitive capabilities of a human being, including communication, perception and abstract thinking. But, on the other hand, other beings without all this capacity are also intelligent.

Chimpanzees are admittedly very intelligent, because although they don’t have all of our cognitive abilities, they’re still capable of make plans and communicate intentions (and in certain respects, it could be said that they are even more intelligent than modern humans hehehehe). Genetic proximity is not the issue here, because crows are also very intelligent, being able to do elaborate plans and learn from trial and error to get yourself adapt to the environment. That is, intelligence involves not only abstract reasoning and planning, but also reactivity and adaptation. A fly can be considered intelligent in the sense that it is able to successfully avoid most of its attacks with the slipper.

In this sense, AI studies eventually evolved into other concepts. The article Elephants Don’t Play Chess (Non-developy Elephants), by Rodney Brooks, which led to Subsunction Architecture widely used in academia and industry robots, they preach that intelligence is something simpler, and whose complexity comes from the various forms of behavior where it is expressed. Thus, in AI - as applied science - the concept ideal of intelligence is rationality: to do the best whenever possible.

Although originally AI had a very strong focus on reproducing human capabilities, it is now believed to have a more applied goal (the discussion of weak and strong AI from our colleague’s reply @mgibsonbr deals precisely with this duality of intentions). Russell and Norvig, in their book, make an interesting comparison:

The objective of Aeronautical Engineering is not to create flying machines that be perfect enough to fool the birds, but create flying machines that do this (fly) in the best possible way. Also the purpose of AI, as an engineering discipline, is not should be creating machines that act so perfectly to deceive humans. But to create machines that are capable of adapt to solve complex problems.

This is not to say that studying AI does not help us to understand our biological capabilities as well. After all, AI involves several areas, including the Processing of Natural Language, to Representation of Knowledge, the Reasoning, the Learning (From Machine), to Vision (by Computer) and the Moving and Manipulating Objects (that is, to have a physical body and to recognize oneself as an agent in the world, something important for the Robotics).

By answering your questions

1. What is the definition of artificial intelligence?

The Articial Intelligence is an area of development that touches the Computer Science in the construction of computational systems capable of solving complex problems and the Engineering in the construction of sensors and actuators necessary for the so-called intelligent agents are fact rational (in the sense that they seek to do their best whenever possible). Therefore, the general architecture of an intelligent system (said an agent, since it exists in an environment, perceives the changes in this environment - yours and that of other agents - and acts/acts to change it in an attempt to achieve its objectives) is as follows:

Engineering enters into the construction of the sensors and actuators of the agent (and also of the environment, in many scenarios), and Computer Science enters into the construction of the decision mechanism (the box with the interrogation there inside).

2. Starting from a definition like this "systems that think like a human", it would be correct to say that every system that states that 2+2 = 4 is a system with "artificial intelligence"?

No, why do the calculation 2 + 2 = 4 does not require "thought", especially reasoning as we humans do. Your calculator is able to do this calculation only with simple logical operations. Focus on this symbol: "2". This can mean anything, or it can mean nothing. What is the difference between "2", "II", "." or two stones in the middle of a crater on Mars? Thought only occurs when someone (or something) gives meaning to a symbol or an operation.

In this respect, a computational system that generates "4" as a response to two inputs of "2" and "2" is merely a program. It was programmed for this and will probably give some error if at least one of the entries is different (".." + 2 gives how much? Probably error of type Mismatch! :) ). An intelligent system seeks to do the best it can. It will not merely give error or "break" (crash), because it will generate a result that can be momentarily bad until it has the possibility to adapt to this new entry (perhaps requesting the correct answer to another agent who knows it, for example).

Answer 2

There is no consensus on the best definition for "Artificial Intelligence" (or "Computational Intelligence", another widely used term). For example, a definition quite simple but that doesn’t say much is:

Artificial Intelligence is the enterprise of building an intelligent artifact.

However, there are controversies in what can be considered "intelligent" or not:

• Intelligent is what a human does well?

• Do other life forms also possess intelligence? (e.g.: an ant is "dumb", but an ant colony behaves surprisingly sophisticated)

• Who aspects of a human being is it necessary for a machine to possess for it to be considered intelligent? (it is necessary, for example, to have vision and hearing, or just to have the ability to reason?)

etc. Finally, there is the question of whether the machine needs to be intelligent in fact or just seem intelligent (see "I.A. Strong" vs. "I.A. Weak"). The "Chinese room" is an interesting mental experiment that illustrates this distinction.

As for a system that states that 2 + 2 = 4, whether it is "intelligent" depends heavily on which premises he works, and that projects it used to arrive at the result. It is a "normal" computer - with a numerical data structure and an addition operation built-in? So I would say that "no", the computer is only running a very specific order that has been programmed in its "essence". On the other hand, if that computer received only the basic axioms of mathematics (set theory, etc.) and was able to "rediscover arithmetic" to the point of not only answering this question but also reaching [correct] conclusions that the author himself did not foresee (such as "the opposite of prime numbers"), then I would say yes that he demonstrated intelligence.