To Become Intelligent, Watson Needs an Ecosystem

On February 16, 2011, an IBM computer named "Watson" crushed its two human competitors in the final night of a three-day Jeopardy battle. Its final score was $77,147 to the humans' $24,000 and $21,600.

But in a Google age where anyone can find the answer to a Jeopardy question in seconds, trivia mastery is attaining buggy-whip status. The most impressive aspect of Watson isn't the trivia database. It's the ability to sort out the convoluted syntax of a question and give a single concise answer.

It might seem that Watson is another incremental step toward computers attaining humanlike intelligence. Just keep making them faster and more complicated, and sooner or later they'll become self-aware.

Actually, no. Artificial intelligence doesn't gain much just by scaling up. What Watson needs is an ecosystem, not more RAM.

I'll explain this by telling the story of a computer my dad showed me when I was ten years old. It played a game called "Hexapawn."

Hexapawn is a simplified version of chess played with only three pawns per player. It's played on a tic-tac-toe board. The object is to get one of the pawns to the other side of the board. Games usually last less than a minute.

When the system lost a game, you "punished" it by eating the jujube that made it lose. After a number of games, you had eaten all the jujubes that led to its losing the game. After that, the system always either won or tied.

Part of an nalysis of Hexapawn.

But I soon saw that there was no intelligence at all in the system. It was just a bunch of boxes, candies, and rules. Once I understood the setup, it became trivial.

Today, after having done a fair bit of coding, I could describe the boxes and candies in programming terms. The boxes were a database. The rules for moving and eating jujubes were procedures.

I now understand that there is no essential difference between a computer and my dad's little pile of matchboxes and jujubes. They are both devices for shuffling around data. They are both computers.

It doesn't matter how impressive they look on the outside. It doesn't matter if they can beat a world chess champion or Jeopardy player. Once you open them up and look at the code, you can see that they are just machines following preset rules.

The clearest evidence of this is that Watson, for all its Jeopardy prowess, couldn't play Wheel of Fortune. Even more to the point, it couldn't want to play Wheel of Fortune.

Many scientists say that the human brain, too, is an information storage and manipulation device. That it's not different in principle from my dad's matchboxes and candies.

To be sure, its scale is much larger. It's got about 100 billion neurons. Each neuron can connect with thousands of others, and those connections store information. The computational neuroscientist Sebastian Seung has estimated that if every neural connection could be stored in a database, it would take up 100 million terabytes of disk storage space. For comparison, Watson has 16 terabytes of storage space.

It would seem that intelligence is at least partly a function of size. That to get creative, innovative behavior, you need to have some minimum number of computational units.

But beyond that, scientists have very few answers. No one really knows how the 100 billion neurons in a human brain can not only learn to play Jeopardy, but can also sulkily refuse to play and ask to be taught chess instead.

Part of the reason for the brain's flexibility is that it had to be flexible to survive. In the distant past, brains that were not flexible quickly became lunch. Brains that could outthink predators survived and reproduced.

But Watson has no predators. All of its energy needs are met. Thus it has no externally imposed reason to do anything other than what it does. More than that, it has no way of creating successors that are different (and potentially better), and therefore no reason to develop a desire to do so.

But Watson's economic context supplies precisely this. IBM, its creator, does have predators (or, rather, competitors.) It does have to expend effort to survive. It does have an incentive to create better computers. The economy in which Watson exists is very Darwinian indeed.

On the level of hardware, Watson is no smarter than a bunch of matchboxes and jujubes. No matter how many terabytes of RAM you add to its successors, they'll never get smart enough to want and ask and wonder.

To get wanting and asking and wondering, you need an ecosystem. It's not that computers will become intelligent. It's that the combined system of humanity and the Internet will become smarter than either one of them alone.

Michael Chorost is the author of World Wide Mind: The Coming Integration of Humanity, Machines, and the Internet, which has just been published by Free Press.

Note: Interested readers can find the details of Hexapawn in Chapter 8 of Martin Gardner's book The Unexpected Hanging, originally published in 1969 (fulltext is here.) Ancient as the book is, I recommend that any kid with an interest in computers read the chapter and make the game. It's an ideal way to get a literal, physical, hands-on feeling for how machines manipulate information.