People have been asking why IBM's lies about Watson upset me so much

I wrote a recent post about how IBM is lying about Watson’s thinking abilities in its commercials. While people in AI understand my frustration, many commenters wondered why I was so upset since these are just commercials, which are typically not true anyway.

I have been working in AI, particularly in natural language processing, since the 60’s. In fact, natural language processing is a term I made up since the previous term (computational linguistics) identified the field as part of linguistics which it really wasn’t. Many people besides myself have wondered about what makes people smart and what might make a computer smart. This question is at the heart of AI.

Lately we have had to endure a lot of fear mongering about AI. We are all going to be eaten or something, because of AI. There have been plenty of pronouncements about why we should be very afraid.

I wish AI were sufficiently advanced so that some fear like that were actually warranted. But really, we all know that SIRI can be somewhat useful but isn't smart and isn't a threat to us. We are really far away from such issues.

This is why the Watson ads upset me. They are selling the idea that Watson is intelligent in some way that it most certainly is not.

To see this, forget the TV commercials and take a look at what IBM says about Watson on its own website:

What is Watson?

IBM Watson is a technology platform that uses natural language processing and machine learning to reveal insights from large amounts of unstructured data

Notice the use of the word “reveal.” The obvious question is “reveal to whom?” It is easy to misunderstand IBM here and I believe the deception is intentional. Watson “understands” nothing. It can find words that relate to other words. If doing that on a massive scale is helpful to somebody then who could complain? To put this another way: If a person seeing correlations Watson has uncovered learns something important, I am for it. But Watson does not understand what it has “revealed" because Watson doesn’t “understand” anything about what it is “reading”. It is simply correlating data.

Here is more from IBM’s site:

How Watson answers questions

Watson first needs to learn a new subject before it can answer questions about it
First Watson learns a new subject

All related materials are loaded into Watson, such as Word documents, PDFs and web pages
Questions and answers pairs are added to train Watson on the subject
Watson is automatically updated as new information is published

Again IBM is being deceptive here. Learning is a complex issue. For people, this means more than copying a new fact into one’s memory.

(School tends to use this idea of learning unfortunately. Most of what we know that matters to us is not factual knowledge but pragmatic knowledge — how to do things, how to come up with new ideas, how to execute a plan, how to reason from our experiences.)

IBM is very clearly saying that what they mean by learning is memorizing question and answer pairs. Imagine if people did that. Imagine if the only way we could answer a questions was by looking up the question and seeing if we had a previously memorized answer.

(“What would you like for lunch?” “I have looked into my memory and see that I have previously answered that question in one of ten ways, so I will select one of those answers.”)

Here is IBM again from that same site:

How Watson learns
Watson answers a question

Watson searches millions of documents to find thousands of possible answers
Collects evidence and uses a scoring algorithm to rate the quality of this evidence
Ranks all possible answers based on the score of its supporting evidence

Imagine if people did that. Imagine if we had to search thousands of documents in order to answer a question. We clearly don’t do that. IBM is not doing AI. Watson is not thinking. How people answer questions is still an important AI question. The answer depends upon how human memory is organized and how new experiences are absorbed into memory. AI is still working on that.

It would nice if IBM stopped pretending that they are doing AI or building intelligent machines. If Watson is useful then it will be used. They do not have to damage the field of AI to sell it.

We in AI, have lived through one “AI winter" where all funding for AI was cut because of grandiose claims. It would be nice if IBM was aware of that and tried not to kill off AI or make people apple crazy with fear about a machine that is in no way “intelligent." Being able to search through mountains of data does not define intelligence, nor is it “learning.”