What is Artificial Intelligence?
no clear definition!
| Humun-like behavior | Rational behavior | |
| Reasoning | Systems that think humanly | Systems that think rationally |
| Acting | Systems that act humanly | Systems that act rationally |
1. Acting Humanly
the Turning test proposed by Alan Turing in 1950 tests wheter a machine acts humanly.
Turing test
A coumpter passes the Turing test if a human interrogator cannot tell wheter the answers from his written questions come from a human or a computer.
/ 기계가 인간과 얼마나 비슷하게 대화할 수 있는지를 기준으로 기계에 지능이 있는지를 판별하고자 하는 시험
The computer would need the following capabilites:
- natural language processing to communicate in natural language;
- knowledge representation to store what it knows or hears;
- automated reasoning to use stored information andto draw new conclusions;
- maching learning to adapt to new circumstances and to detect and explre patterns.
튜링 테스트는 의도적으로 피지컬 인터렉션을 피했는데, 인공지능에겐 그것이 필요하지 않기 때문이다.
Total Turing test
Based on the Turing test and additionally requires that the subject can pyhsically interact and see the other person using video.
The computer would additionally need the following capabilites;
- computer vision to perceive objects;
- robotics to manipulate objects and move them
위의 볼드체 여섯 가지 필드가 인공지능의 핵심 테마이다.
2. Thinking Humanly
To test whether machines think like humans, we need to get "inside" the human mind
/ 이 부분이 사회 윤리적으로 법적 이슈가 있는 부분이다.
- introspection_자아 성찰, 내성 : trying to catch our own thoughts as they go by;
- psychological experimetns_심리적 실험 : observing a person in action;
- brain imaging_뇌 과학 이미지 : observing the brain in action;
Thinking humanly approach
Once one has a sufficiently precise theory of the mind, one could write a computer program of it. If the input-output behavior matches human behavior, some inner workings of the program might correspond to human thinking
이 인간 사고적 접근을 우리는 인식 과학 이라는 새로운 필드에서 다룬다.
This approach is typically followed in cognitive sciences.
3. Thinking Rationally
Rationality
A system is rational if it does the "rigiht thinkg", i.e., has an ideal performance
Logics is often used to mimic rational thinking since it provides correct conclusionts given correct premieses, e.g. :
Socrates is a man; all men are mortal; therfore, socrates is mortal.
(삼단논법)
Difficulties:
- Informal knowledge has to be formalized.
- Computational effoert might be infeasible: Problems with a few hundered facts might exhaust the capabilities of today's computers.
4. Acting Rationally
Agent
An agent is just something that acts
/ Latin: agere means "to do"
Rational Agent
An rational agent is one that acts so as to achieve the best outcome.
/ 최선의 생산을 얻기 위해 행동하는것
- Acting rationally can be based on thinking rationally. However, a fast reaction is sometimes better than thinking longer about the best soulution - ex) removing your hands from a hot objects
- The skills required for the Turing test allow an agent to act rationally - natuaral language processing , knowledge representation, automated reasoning, machie learning, computer vision, robotics
State of the Art
- Automated Driving
- Robotics
- Plannung and Scheduling
- Chess Playing
- Playing Jeopardy
- Speech Recognition
- Machine Translation
- Spam Fighting
- Video Games
Human vs. Machines
| Machine | Human |
| high speed | slower |
| high accuracy | less accurate |
| high forces | less powerful |
| fast feedback | slower feedback |
but:
| Machine | Human |
| less adaptive | hightly adaptive in unforeseen situations |
| typically specific power and sensitivity | large range: power vs. sensitivity |
| typically designed for specific purpose | "Universal" capabilities |
Summary
- Different people approach AI with differend goals in mind. Two important questions to ask are:
Are you concerned with thinking or behvior?
Do you cinsider human-like or rational bevavior?
- For specific tasks such as chess playing, AI can outperform humans similaryly to calculatiors outperforming humans in basic arthmetic.
- AI is a topic of wild and interesting seculations, such as the technological singularity.
댓글