Watson Deep Into Artificial Intelligence

Watson deep into artificial intelligence

What is artificial intelligence? Should it be smarter than us? Does it need to learn and does it need to learn from its mistakes? Should it be conscious? I would really like to say artificial intelligence is not about the consciousness of a machine, but it is really about getting help from a machine.

A good test of intelligence is to be able to follow instructions like we do, understand language, like we do and understand what it sees, like we do. Ingensys NEO  literally understands everything in almost any picture. "It's very exciting," said Oriol Vinyals, a research scientist at Google. "I'm sure there are going to be some potential applications coming out of this." It is worth remembering that nobody at Google programmed the software with rules for how to interpret the scenes, its simulated neural networks "learned" by consuming data. You can find more information on how pictures are converted into words on a white paper written by Ali Farhadi.

The Exponential Speed of the AI Cognitive Revolution

The biggest difference between this and the Industrial Revolution in the 1800s is the speed with which the cognitive revolution will come. As in our previous article about the exponential growth of AI, the rate of change adheres to Moore's Law, which means every two years the processing power, and in this case AI understanding, doubles and that is why AI progress is so fast. There is research into Artificial Intelligence going on all around the world, but we have decided to start our tour in New York, where one of the most amazing AIs is from, grew up..err.... was born?

Who is, or what is Watson

About an hour north of Manhattan lives possibly the most famous example of a smart machine. Certainly the one that can beat us humans at our own game (most of the time). Watson is what its creator, IBM, calls a cognitive system. Fill it with unstructured data, and it will analyse it, make connections, derive conclusions and learn from its mistakes. Once it has some data to work with it will then allow you to interrogate it using English-language questions.

Watson reads, it has an unbelievable appetite for information from books, periodicals, blogs and tweets. However Watson not only reads but it actually understands what it is reading and is able to put that vast amount of data in context. Watson learns, with interaction and with feedback it actually gets progressively smarter.

Well, here he is, this is the Watson that won Jeopardy. In total there are ten racks just like this one. That is how much physical hardware it took to beat the best of the best human contestants in 2011. I wonder if we have the driver for that... and does it run Windows 10?

The future is cloudy for Watson

Joking aside, since then, technology has marched on and these days IBM is offering Watson's abilities to the world. Watson now lives in the cloud, although it is categorically not Skynet (so they say.. maybe that is Google?). These days, you can have your very own empty version of Watson. Load it with whatever information you want and it will go to work on it. Like Chef Watson, which is cooking up wacky new recipes, by analysing cookbooks and working out what ingredients are chemically compatible with others, even if they sound like vile combinations.

So we are going to ask Watson something now!
"Watson, What is snow?"
Watson replied " The white stuff in winter, particles of water vapour, vapour spelt incorrectly because I am speaking in English".
Well make what you will of the answer but pretty much it got it right.

The team at Elemental Path labs in New York City are developing a toy dinosaur which uses Watson to answer questions from inquisitive children. Watson is little bit more complicated for that, so it boils it down to what is being said by the toy, and tries to understand what is being said in that one sentence. Children do not ask questions in the same way adults do, and they can be rather abstract. Behind the scenes, the back end of Watson (ooh err!) is actually returning masses and masses of data. Watson then asks, "how did I evaluate that as the right answer for it to respond with? What is the confidence level like?" So I would say there is a 75% chance that I believe that that was the correct answer to the statement. It has the evidence behind it, other facts wrapped around it and it gives the context for it. So you really can drill down on the meaning of what is being said other than just that flat statement. So pretty much this is the answer to your question! This is how you teach an AI, this is how the AI learns and evaluates if it answered the question correctly.

Watson Wins Double Jeopardy

When watching the video, remember that nobody is telling Watson what questions to go for, or how much to bet, and it is listening to the questions, understanding them and answering them. It does help if you know how to play Double Jeopardy or the video does not make sense!

Article Updated May 27th 2022