JUST A GLANCE IS NEEDED TO DIAGNOSE CANCER

Yesterday I described the strategy IBM followed to teach their computer to translate texts from one language to another. The technique basically was to tell the computer what a good translation looks like. They taught it also to diagnose cancer – well, at least to distinguish a malignant tumour from a non-malignant one.

How did they do it?

They fed photos of tumours into the computer – 3000 of them. Interestingly, 2800 of the photos were of tumours that were not malignant and 200 of malignant ones, i.e., cancerous ones. The computer has learned what a “not-dangerous” tumour looks like and what a dangerous one looks like.

The interesting fact is the difference in the number of the photos of the bad and the good they used – 2800 good and 200 bad. It looks to me like these scientists are disciples of Don Bosco, whose outlook was always totally positive. He believed in what he called the “Preventive System” : Fill the minds of students with what is good and make this good so much fun that they will not even think of the bad. Fill the days of children so full of fun derived from good activities that they do not even feel tempted to the bad.

That's the best trick of good education whether at school or in the family!

TEACHING THE COMPUTER TO TRANSLATE

I am reading the latest book of Thomas Friedman, released just two weeks ago – THANK YOU FOR BEING LATE. Yesterday I read something about artificial intelligence – a computer that learns from its mistakes as a human does, improving itself with each mistake it makes.

How did IBM teach computers to do translation? They started with translation from English to Spanish. A large group of linguists analyzed the grammar of both languages and the technicians wrote a translation program. It didn’t work.

They got rid of all the linguists. They went the way of statistics. They wrote a sentence and did a good translation of it and a bad one. They fed both into the computer telling it which was the good translation and which the bad translation. They did the same with millions of other sentences, each time giving the computer a good translation and a bad translation.

We now type a sentence into the computer. It makes a literal translation of the sentence and compares it to the millions of good and bad expressions it has stored up in its memory. It rejects what looks similar to the bad expressions. It changes the translated sentence to look more like the good expressions, improving it at each step. All this happens in a fraction of a second and we have the final translation by the time we finish typing our sentence in.

I know this is a very simplified way of explaining this process of millions of operations per second. But, hopefully, it conveys the main idea.

We spent a lot of time, money and energy teaching our children the rules for behavior at work, in school, on the road, in business. What is more important than teaching rules is to provide them with as many models of good behavior as possible – our own good behavior and that of their peers. The importance of our own good conduct and of helping them to get into right youth groups!