Вы находитесь на странице: 1из 3

Lecture 2 : let´s get started!

Definition of AI

So, let´s see how we can make computers actually think. First of all, let´s define what
Artificial Intelligence actually is. Like in many scientific fields, that is a hard question to
answer. That is no surprise, since “thinking” and “intelligence” themselves are intuitive but
difficult to define words. Let´s see what some authors say about that:

Elaine Rich (AI, 1983) says: AI is the attempt to make computers do things, in which humans
are better. That is not an very exact definition. There is a need to specify what “better”
means, of course. This also limits the definition to the contemporary state of AI. AI
researchers definitely try to make computers better in what they do, no matter how good they
already are in it.

Nilsson (1971) says that AI is the attempt to make computers show behavior, which, when
noticed on human beings, would be considered “intelligent”. Again, there is the problem with
the definition of intelligent behavior. Also, AI researchers struggle with teaching the simplest
of behavior to computers, and only a very few cases actually manage to achieve higher
cognitive functions. With that definition, AI research hasn´t even started yet.

Another definition is given by Ginsberg (AI essentials, 1993, a very good book for
introductionary AI), in which the Turing-Test is being used. That makes AI the attempt of
teaching machines to pass the Turing Test. The Turing Test was conceived by Alan Turing in
the 50s, at the very beginning of computer science. The Turing Test works like this: There are
2 rooms and a computer terminal in each of those. In front of one of these terminals sits a
human, in front of the other one a human or a computer. The human in room A doesn´t know
if there is a human or a computer in room B. The human in room A can ask as many questions
as he wants, from as many topics as he wants. After a while, he has to determine if there is a
human or a computer in room B. If he stated “human” and there actually was a computer, that
computer gets to be called “intelligent”1.
So, what does a computer need to get to be called intelligent? He has to know how to
recognize and compose text, of course, but there are many much more difficult processes in
the background. Google and Wolfram Alpha show how computers can process and

1
Of course, this has a few more rules for systematics sake, to prevent cheating for example, but you get the
point.
understand phrases and text, and compare them to already gathered data. But they can´t
produce new knowledge or form opinions, which are tasks required by a human being. How
these processes actually work is beyond the modern state of cognitive science. AI tries to
decipher these processes, but more so tries to produce intelligent applications for specific
tasks, such as the intelligent text-processor or the face-recognition software at your local high-
security airport. AI is a very diverse field, which includes but is not limited to expert systems,
automatic proving algorithms and search. Expert systems simulate human experts in a field,
such as physicians. The expert systems are informed about all the knowledge an expert should
have in that field, and answer formal questions, such as “Patient A has got symptoms B and
C, what could be his disease and what should be the treatment?”. Automatic theorem proving
does exactly that, it teaches computers how to prove theorems, so that they can prove them
themselves later on. Both of these are by no means easy tasks, especially the second one is
kind of the holy grail of AI, but at the current state far from being actually useful. Search, or
heuristic search (heuristic in greek is finding) is another typical field of AI. With search,
computers search for the best solutions for a problem, which has been properly formalized.

Why are we doing search and boardgames?

Heuristic search is one of the pillars of AI-research and you can find it in most AI-
introduction books. Although it is considered as inelegant, it is very necessary in any AI-field.
For us, it is more interesting because of the simple implementation with python. We will use
mainly search algorithms through trees.
Boardgames on the other hand seem a pretty silly application of research resources. Actually,
Charles Babbage (another father of computer science) envisioned a computer in the 40s
which could play Tic Tac Toe. Shannon und Turing stated their own theories for intelligent
Chessprogramms a few years later. Shannon even postulated an approximation of the number
of possible positions in chess. One of the first computers ever (EDSAC) was reprogrammed to
play tictactoe. The fact that boardgame-programming isn´t obsolete yet id the fact that another
AI milestone was the victory of Deep Blue over the world champion Kasparov in 1997.
Nowadays, Chess programming is no field of research anymore, but Go is a game where
humans are still much better than the best computers.
Lastly, search and boardgame programming produces a large number of techniques used
elsewhere, from AI to economic predictions.
Tree search
Before getting to the first example, let´s have a small overview over trees and graphs. I will be
using both formal and intuitive definitions.

A directed graph is a tuple G=(V,E), where V is a set of “vertices” and E consists of tuples
(v,w) of vertices, where each tuple represents an “edge”. For an edge (v,w) we say v goes to
w.
Practically, you can consider all vertices (singular: vertex) as dots on a paper, and edges as
arrows connecting those dots. The arrowhead in an edge (v,w) points in the direction of w.

A way in G from v0 to vk is an n-tuple (v0,...,vk) where vi is an edge for all i and for each i
(vi,vi+1) is an edge in E.
Practically, a way is a set of arrows that lead form v0 to vk. That means you can start at v0
and follow arrows through G, until you land on vk. W consists of all the vertices you went
through to reach vk.

A tree T=(V,E) is a graph where:


1) T has got exactly one vertex r, which no edge reaches (there is no vertex v in V for
which there exists an edge (v,r) in E). This is the root of the tree.
2) Every other vertex is being reached by exactly one edge, and can be reached through
the root. (vertex v reached through the root means there is a way W=(r,…,v) in T)

Вам также может понравиться