Watson, as I’m sure that everyone has heard of, is the name of IBMs jeopardy winning computer. normally, there’d be some sort of adjective before computer like “super,” but I don’t know what it would be, perhaps smart is the best word for it? I recently read that IBM was attempting to use this in Oncology, to help doctors treat cancer. This is, for many, perhaps myself included, a scary thought. We all know that computers are capable of doing amazing things, but they are also capable of doing incredibly dumb things. I would almost go so far as to offer a definition of common sense as the thing that computers lack. So I think it is important not to remove humans from the decision making process, but of course nobody is arguing to remove humans from the process. (interestingly, while the best computers have been able to consistently beat the best humans at chess for about 10 years now, which is always the first example given in regards to “oh the humans are losing”  to computers nonsense. What is interesting though is that, in chess, computers almost never make stupid mistakes. However, the best players in chess are “centaurs” or combined human/computer teams, but in this case its the humans, instead of protecting against computer stupidity, providing human creativity).

There’s a story about a teacher trying to teach kids something about knowledge, so she gives her students three numbers: 2,4,8. Then she tells the students that these numbers follow a certain pattern, and that they can give her other sets of numbers, and she will tell her student if it follows the pattern she has in mind or not. (There’s a video of that right here, in case you don’t want to read). The students very quickly notice a pattern, that each number is twice the previous number, and then give her similar sets: 3, 6, 12 or 4, 8, 16. For those of these, the teacher says yes, these fit my pattern.

However, the pattern is just that the numbers are in increasing order, not that they are double the previous number. This pattern necessarily includes all sequences where each number is twice the previous one (for positive numbers). But once we get a certain theory in our heads, its hard to get it out. So what should the students have done? Simple, not only give sequences which they think will yield positive results, but sequences which they think will yield negative results. Now, you may be thinking that this is simply confirmation bias, but you’re wrong. The students were never given any information which contradicted their hypothesis, so it’s not that they were ignoring or minimizing inconvenient information. Furthermore, they weren’t above testing their hypothesis, oh no, they were ready to put their theory to the test. The problem is that they didn’t know where to look, or which questions to ask.

For better or for worse, this is how humans think. They make observations, come up with a theory or pattern, which causes them to make further observations, which leads them to ask certain questions, which may or may not lead them to a decision. This is totally different from how a computer can make these same distinctions. Instead of looking for the “big” evidence which points in one direction, the computer can look at many small items, compare it to some type of database, and can take a bunch of pieces of “small” evidence and turn it into a guess.

The best way that I can describe this is to use the game 20 questions. When humans play it, they typically go through some sort of decision tree, asking a question, then based on the answer pose a new question within that tree. For animal, it might even go through kingdom/phylum etc. (it is a vertebrae? Yes, is it warm-blooded? Yes. Is it a mammal? No. Is it a flightless bird? Yes. Is it an ostrich? No. is it a Penguin? Yes)

Compare that tree of questions to the below, courtesy of 20q.net. (if you’d like to play, for some weird reason select the “think in British” option, it seems to be smarter than the American Version).

1. It is classified as Animal.
2. Can you buy it? No.
3. Can it jump? Yes.
4. Is it ferocious? No.
5. Is it multicoloured? Yes.
6. Is it a type of mineral? No.
7. Is it fluffy? Yes
8. Is it originally from Africa? Sometimes.
9. Does it live in the jungle? No.
10. Is it black? Sometimes.
11. Does it live in trees? No.
12. Would you give it as a gift? No.
13. Does it live in groups (gregarious)? Yes.
14. Does it have a bushy tail? No.
15. Can it be tamed? No.
16. Can it climb? No.
17. Are there many different sorts of it? Yes.

Q18. I am guessing that it is a penguin?

None of those questions seem to follow any sort of pattern (after the initial animal question)- but somehow, taken together, they indicate penguin. None of those seem to hone in on Penguin at all; no question about living in the arctic, swimming, eating fish, or for that matter, even being a bird. Yet based only only those 17 questions (one of which seems like complete nonsense – is it a type of mineral!), the computer somehow knows that I was thinking of Penguin. Note also that most of the questions are completely unlike the children in my story, the computer must be getting a lot of information from the “no” answers.

A long time ago, when 20q first came out on the internet and AOL Instant Messenger was still a thing, used to on, get my friends to play 20 questions with me, only I’d be using 20q. So they’d be given a long string of what they think were nonsense questions, only to be told the correct answer. My friends thought I was doing magic, I suppose I was cheating but it was fun.

Take the question, “is it multicoloured” Penguins are, because they are black and white. But when we hear that, we suddenly think of peacocks, or baboons, or things that are not just multicolored, but extravagantly colored. We infer more from the answer than can be truly stated.

Which all brings me back to our friend Watson. A computer which can help diagnose conditions based on a bunch of diverse, mundane data is something that will almost certainly prove useful. Humans are, on the whole, much more creative than computers, but there are certain circumstances in which computers appear more creative. Considering penguin as an answer to my above set of questions, for instance. It’s not real creativity, it’s just that the computer can consider every animal at once, we are limited in our considerations. While I’m not quite ready for medicine to be taken over by computers entirely, I’m looking forward to seeing Watson help out in the ER.

Leave a comment

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s

%d bloggers like this: