What is AI?
Lately, I had some spare time in hand and decided to study AI. To do that, started to read Artificial Intelligence A Modern Approach from Stuart Russell and Peter Norvig. I have to say that it is a good book to study for intelligence agents and AI. In this AI series, I will try to explain what I learned from this book. Anyway, let’s cut the chase and get started.
What is AI? It’s not that easy to define what AI is. There is no one absolute definition for it. In fact, there are four general approaches to decide about intelligence of the machine. In other words, artificial intelligence.
- Turing Approach (Acting Humanly)
The art of creating machines that perform functions that require intelligence when performed by people.
Kurzweil
Turing proposed a test called “The Imitation Game” to settle the argument about machine intelligence. In the test, there are three rooms. In one room, there is a contestant, in other one there is a machine and in the last one there is a judge. Judge’s job is to decide which one is human and which one is the machine. Judge asks questions to decide. Turing says that if the judge cannot decide which one is human and which one is machine, machine has to be smart/intelligent. In other words, we say that the machine/system acted like a human.
To be able to pass this test, machine needs to have natural language processing, knowledge representation, automated reasoning, and machine learning capabilities.
2. The Cognitive Modeling Approach (Thinking Humanly)
[The automation of] activities that we associate with human thinking, activities such as decision-making, problem solving, learning…
Bellman
This approach says that if a machine can think like a human, it must have intelligence. But, how can we determine how human thinks? In cognitive science there are three methods: through introspection, through psychological experiments, and through brain imaging. “Once we have the theory of mind in hand, it is possible to express this theory as a computer program. If the computer’s input-output behavior matches with human behavior, we can say that some of the program’s mechanisms could also be operating in humans as well.” Since, humans have intelligence, machine might as well.
3. The “Laws of Thought” Approach (Thinking Rationally)
The study of computations that make it possible to perceive, reason, and act.
Winston
Logicians developed a precise notations for statements about all kinds of objects in the world and relations among them. In theory, any solvable problem which can be represented in logical notations can be solved by a machine. Logicist tradition hope to create a machine to solve such problems to create an intelligence machine. This approach is a direct line through mathematics and philosophy to modern AI.
However, there are multiple problems about this approach. Not all intelligent behavior is mediated by logical reasoning. Also, it is extremely hard to encode some uncertain information. Lastly, in practice its intractable.
4. The Rational Agent Approach (Acting Rationally)
This is the actual meat of the book. We will learn together about rational agents in depth with this series. What is rational agent? Let me start by defining what agent is though. Then we will come back to rational agent. An agent operates automatically, stays around for a long time, adapts to situations, creates and pursue goals. I feel like its kind of a human virus. A rational agent is an agent that pursues the best outcome. Given available information tries to maximize the goal achievement. It seems like a pretty.
To do all of these above, a rational agent might use laws of thought as well. In the end, logical reasoning is a good way to solve problems. However, in some situations there is no provable correct thing to do so it may not be the best way, so rational agents takes a different approach. Think about this, as a reflex action. If there is a hot stove in the room and touch it accidentally, you will immediately stop touching the stove. This is a reflex and so fast. Its a lot better than thinking about what is a rational thing to do. Just do the action. Just a reflex. So rational agent takes this reflex action.
There are two clear advantages of rational agents over other approaches. First, its more general than thinking rationally. Its only a way to implement rational agents. There is a lot more thing for rational agents. Also, rational agents is more amenable to scientific development than approaches based on human behaviors.
Hope you guys enjoyed this post. Following posts are on their way. Stay tuned. Peace out.
P.S. I am open for any comment and recommendations. If I missed some points or some parts need more explanation, please let me know.
