lesson_01.md 3.9 KB
AI - lesson 01
http://home.deib.polimi.it/arrigoni/ArtificialIntelligence.html
Francesco Arrigoni
francesco.arrigoni@polimi.it - 02 23993475
Wed 14:00 - 16:00 send email first
7 October 2015
1. Introduction to Artificial Intelligence
During the course we will see methods to develop intelligent agents. In particular we will see various methods in order of complexity but not necessarily in chronological order.
The algorithms will be presented in pseudocode, no real implementations will be shown, for implementations examples go to aima.cs.berkeley.edu
Relation between AI and other courses:
Relating to Soft Computing, in AI are shown general method for creating intelligent agents, in SC are shown some recent techniques in particular, in Autonomous Agents and MultiAgent Systems are shown methods to connect different intelligent agents. In Machine Learning / Data Mining the topic is how to improve an intelligent agent by making it learn.
There are also some philosophical questions which are faced in Computer Ethics.
Artificial Intellicence
1. Act as humans
Create machines or entities who can act as humans.
Alan Turing in 1950 defined the imitation game, later being known as the turing test.
The original formulation is the following
| Man | Interrogator | Human |
|---|
The interrogator can interact with the two subject using a typewriter The goal of the man is to try to let the interrogator think he is a woman. The role of the woman is to try to let the interrogator know she is really a woman.
Now, let's change the man with a computer, and let him try to act like the man in the precedent case.
The computer is defined intelligent if the success rate is the same to the experiment with Man/Interrogator/Woman.
The most famous variant
Is the one in which a computer can trick an interrogator to think he is human in more than 50% of the case.
This is an operational definiton of intelligence, not an analytical one
The computer is considered as a black box, the internal working of the computer is not relevant, only the external interaction is.
A problem related to this test is that the decision may depend on one interogator or another.
2. Think as humans
The goal of Artificial Intelligence is to build machines that work exactly as a human brain, in other words to create machines that think as humans.
The problem with this approach is that our scientific knowledge is not sufficiend to reproduce a human mind.
3. Think rationally
Artificial Intellicence is building machines that think rationally. Like building machines that think logically and are able to solve problems.
The fact is that human being are not completely rational, our intelligence doesn't stop to proving theorem or modus ponens
4. Act rationally <- the one adopted during the course
To be considered intelligent, a machine must be able to not only think rationally but in general to do the right thing. Act rationally can be seen as a way of maximising the results. This is the most scientifically and engineering oriented definition.
John McCarty, the inventor of Lisp organized in 1956 a summer camp to solve all major computational problems like computer translation, in this occasion the term Artificial Intelligence was created (after the introduction of the imitation game).
Some extraordinary milestones of Artificial Intelligence are:
- Google Car, formerly Stanford University's Stanley
- Deep Blue who managed to win againts the human chess champion
- IBM Watson who won a Jeopardy game against the two biggest champions
Ubiquitous AI are small systems integrated in cars or electrodomestics
Also in videogames AI acts a big role, for example the opponents behaviour.
Experts in Artificial Intelligence are not so required in job market because AI is defined as smart things in marketing.
| agent |
|---|
| Environment |