«Look, Ma, no hands!»
«If you can’t explain it simple, then you don’t understand it well enough» is what my grandmother tells me when I try to clarify what Internet is. And I surely trust her when she quotes Einstein.
This will be some sort of prelude to a series of articles focused on AI, and the previous quote is my humble motivation to publish it: I wish to master what I’m currently studying and, possibly, generate proper discussion to deepen the topics.
I will start from some issues posed in the book «Artificial Intelligence: a modern approach (3rd edition)» by S. Russell and P. Norwig.
Please note that the book itself will be my first resource, but I will try both to resume and to add my personal opinion. I will analyze concepts and revise algorithms, mostly to prove myself I truly understood them. Also, I hope to have proper discussion with you readers, should it be because I made some dreadful mistake or because you want to add more to my analysis. Thank you in advance for helping me.
Without further ado, let’s see the subjects I will deepen throughout this journey (these cover the first chapters of the book):
- Definitions of intelligence, artificial intelligence, rational agent;
- Rational behave in different environments, and agent-related terms;
- Heuristic search strategies;
- Adversarial search: games and decisions;
- CSP inference and backtracking search;
- Knowledge-based agents and propositional model vs first-order logic;
- Classical planning approaches and algorithms.
I think later on I will investigate many other topics, such as uncertain knowledge and reasoning and making decisions, the theory of learning, but most of all perception, natural language processing and robotics.
So, what is this overly-discussed intelligence?
Here is the answer I came up with, at least regarding human intelligence: it is a mental capability to understand ourselves and the surrounding, and figure out the best response to act accordingly to every stimulus received. It is made of logical reasoning, critical thinking and learning, but also creativity. Moreover, I think that learning from the past and preparing for the future are expression of intelligence.
Consequently, I think (as a human) the most suitable definition for artificial intelligence could be «The automation of activities that we associate with human thinking» (Bellman, 1978). But we can also give a definition concerning behavior: «The study of the design of intelligent agents» (Poole et al., 1998).
The term agent stem from latin agere, and since I have a Language Studies background, I fell for how well the definition suits the term: «to do», but not just «something». We suppose our agent to use rationality: it tries to do the right thing, given what it knows, to achieve the best expected outcome. Which is why making correct inferences is part, but not all, of being a rational agent.
In 1950, Alan Turing proposed a test to provide an operational definition of intelligence. All the following skills allow an agent to act rationally: if a human interrogator cannot tell if the response comes from a person or a computer, then «winner winner chicken dinner» for the computer!
- Natural language processing, to communicate;
- Knowledge representation, to store the infos received;
- Automated reasoning, to use these infos;
- Machine learning, to adapt and detect patterns;
- Computer vision, to perceive the surroundings;
- Robotics, to manipulate objects.
To get a full understanding of what an intelligent agent should do, we need to know how our brain is made and how our mind works. Here, neuroscience and cognitive psychology come to our aid: studying the nervous systems has come to the majestic conclusion that a collection of cells can lead to thought, action and consciousness. Not bad for a gang of neurons, is it?!
I include here a timeline proposed by Bella Wilson of some of the principal milestones in AI, since it is kinda important to have an eye on history too, and I wish to improve this article (and my knowledge) as much as possible.
My final thought brings into play the title of this article (I freaking love circularity, but the quote is also a personal reminder): intellectuals preferred to think that «a machine can never do X», and what do you think AI researchers replied? Well, McCarthy called what followed the «Look, Ma, no hands!» era, so…
Nowadays, AI is integrated in many applications such as robotics, logistics planning, speech recognition and machine translation, game playing, spam fighting, robotics vehicles… And many more are yet to come.
Coming soon with a new article, stay tuned!
