I thought I’ll study something recently and I decided to explore Artificial Intelligence, because it seems to be in the limelight these days. I did a course in Artificial Intelligence when I was in college. At that time no one cared about Artificial Intelligence. Many of my classmates who are promoting it now, didn’t care about it then. I did that course because I had dipped into Roger Penrose’s ‘The Emperor’s New Mind‘ and Douglas Hofstadter’s ‘Godel Escher Bach : An Eternal Golden Braid‘ at that time and was inspired by them and so I was excited by this field. The course wasn’t as interesting as I’d expected and it didn’t capture any of the excitement that I’d felt while reading Penrose’s and Hofstadter’s books. After passing out of college, I went to work in a tech company. The company that I went to work for was one of the biggest companies around but no one there cared about Artificial Intelligence. The company didn’t care about exciting new technology, none of the top management or the managers had any vision on how the technology universe was unfolding and the company just went with current fashion and fads and wanted to make more and more money and improve its profit margins. From a technology perspective it was a junk company. I tried talking to my boss and other people there about exciting new technology, but I was treated like a disrupter who refused to keep his head down and do grunt work and count his dollars. One of my bosses was so upset with me that he even exiled me to the cubicle next to the toilet. So I gave up after a while.
Enough of that sad story. Time to contemplate on Artificial Intelligence now. So I decided to study Artificial Intelligence properly now. I thought I’ll do it the hard way and got a textbook and started reading it. It went well for around 60 pages or so. (The book is around 1200 pages.) After that things got complex, but not with the complexity I liked. I could have ploughed on with it till I reached a place where things got better, but I thought it might be a better idea to read a simpler book on the subject first. I did some research and discovered Melanie Mitchell’s book.
Melanie Mitchell’s book is around 400 pages long. But you don’t feel that because it flies like the breeze. Her writing style is conversational and reading the book is like attending the class of your favourite teacher. Melanie Mitchell has got a Ph.D in Artificial Intelligence, she worked with Douglas Hofstadter while researching for her Ph.D, and she has worked in the field for more than thirty years. She also seems to be a great teacher. All that shows in the book. She wears her learning lightly, she starts from the basics and takes the reader to a reasonably advanced level, and she gives a survey of the field historically and describes all the important happenings today. She also covers all the important developments and technologies that are part of AI. She doesn’t shy away from the important and difficult questions (like can a computer or a software program really think, is a computer sentient), she handles all the tricky questions with aplomb, she separates the hype from the facts, and she states her point of view strongly whenever she disagrees with the hype. Her affection for her field is infectious. The book is brilliant and so is Melanie Mitchell.
As the oft-repeated maxim goes, if you want to read just one book on Artificial Intelligence, this is that book. It is exceptional.
I’m sharing one of my favourite passages from the book. Hope you like it. It is about how computers still cannot think like humans.
Beginning of Quote
“I want to describe one additional question-answering task that is specifically designed to test whether an AI (NLP) system (Artificial Intelligence (Natural Language Processing) system) has genuinely understood what it has ‘read’. Consider the following sentences, each followed by a question :
Sentence 1: ‘The city council refused the demonstrators a permit because they feared violence.’
Question: Who feared violence?
A. The city council
B. The demonstrators
Sentence 2: ‘The city council refused the demonstrators a permit because they advocated violence.’
Question: Who advocated violence?
A. The city council
B. The demonstrators
Sentences 1 and 2 differ by only one word (feared / advocated), but that single word determines the answer to the question. In sentence 1 the pronoun they refers to the city council, and in sentence 2 they refers to the demonstrators. How do we humans know this? We rely on our background knowledge about how society works: we know that demonstrators are the ones with a grievance and that they sometimes advocate or instigate violence at a protest.
Here are a few more examples :
Example 1
Sentence 1: ‘Joe’s uncle can still beat him at tennis, even though he is 30 years older.’
Question: Who is older?
A. Joe
B. Joe’s uncle
Sentence 2: ‘Joe’s uncle can still beat him at tennis, even though he is 30 years younger.’
Question: Who is younger?
A. Joe
B. Joe’s uncle
Example 2
Sentence 1: ‘I poured water from the bottle into the cup until it was full.’
Question: What was full?
A. The bottle
B. The cup
Sentence 2: ‘I poured water from the bottle into the cup until it was empty.’
Question: What was empty?
A. The bottle
B. The cup
Example 3
Sentence 1: ‘The table won’t fit through the doorway because it is too wide.’
Question: What is too wide?
A. The table
B. The doorway
Sentence 2: ‘The table won’t fit through the doorway because it is too narrow.’
Question: What is too narrow?
A. The table
B. The doorway
I’m sure you get the idea: the two sentences in each pair are identical except for one word, but that word changes which thing or person is referenced by pronouns such as they, he or it. To answer the questions correctly, a machine needs to be able not only to process sentences but also to understand them, at least to a point. In general, understanding these sentences requires what we might call commonsense knowledge. For example, an uncle is usually older than his nephew; pouring water from one container to another means that the first container will become empty while the other one becomes full; and if something won’t fit through a space, it is because the thing is too wide rather than too narrow.
These miniature language-understanding tests are called Winograd schemas, named for the pioneering NLP researcher Terry Winograd, who first came up with the idea. The Winograd schemas are designed precisely to be easy for humans but tricky for computers…
Several natural-language processing research groups have experimented with different methods for answering Winograd schema questions. At the time I write this, the program reporting the best performance had about 61 per cent accuracy on a set of about 250 Winograd schemas. This is better than random guessing, which would yield 50 per cent accuracy, but it is still far from presumed human accuracy on this task (100 per cent, if the human is paying attention)…
Maybe that’s a good thing. As Oren Etzioni, director of the Allen Institute for AI, quipped, ‘When AI can’t determine what “it” refers to in a sentence, it’s hard to believe that it will take over the world.’”
End of Quote
Have you read Melanie Mitchell’s book? What do you think about it?