BOT2001 Report
Freethinking Bots

By Brian Proffitt

"I would dance and be merry, life would be a ding-a-derry,
If I only had a brain."
-The Scarecrow, Wizard of Oz

It is one of the earliest themes in science fiction: machines given the gift of human thought. Within the confines of human drama, such machines are often cast in the role of antagonist, bedeviling the humans in the story until the clever humans outwit the evil machine and pull the plug.

But if you meet Dr. Stephen Thaler, you quickly will dispel any notion that the concept of sentient machines will stay in the realm of fiction for long.

Thaler is the CEO of Imagination Engines, Inc., and creator of DataBots, a technology that uses ordinary Microsoft Excel spreadsheets to create advanced neural networks for adaptive artificial organisms and knowledge agents.

Yes, you read that right. Microsoft Excel spreadsheets.

One of the cores of Thaler's ideas is the neural network, which differs sharply from what he calls "symbolic artificial intelligence" (AI). To Thaler's point of view, almost all current bot and technology is symbolic, since humans are so involved in their creation and their functionality. It is humans that write the bot scripts and provide the rules for the scripts. And, since bots are not creative, they will never deviate from their scripted abilities--no matter how much they are made to resemble sentient action.

On the other hand, neural networks can make their own rules, Thaler explained to the audience at BOT2001. Neural networks can also continuously learn, and can also be creative.

Thaler's discussion takes on the appearance of a highly theoretical talk, discussing topics such as self-training artificial neurons, alert associative centers, and imagination engines. But Thaler has gone beyond the theoretical and into the practical by using built using cell references in a standard Excel spreadsheet.

Each cell in the worksheet is a neural node, with the possibility of millions of nodes in an Excel workbook. These huge neural nets can then be cascaded across multiple machines using TCP/IP technology and the Internet to link them together.

Thaler has already applied this neural network technology to a variety of applications including, he likes to add, the design of a popular toothbrush. To demonstrate the technology for the BOT2001 attendees, Thaler brought his Semantic Parser application.

Semantic parsing is the ability to look at a sentence and judge its components and therefore its meaning based on the sentence itself and the context of the surrounding text. Without that context, even humans can get a little bamboozled.

Thaler's classic example is this sentence: "I saw her duck."

When we look at this, humans might think of three possible interpretations:

• I saw her taking cover.
• I observed her aquatic fowl.
• I cut her aquatic fowl.

Send a scripted parser after this and it might just choose one interpretation, based on the limitations of its script. Thaler's Semantic Parser, linked to a neural network in St. Louis and San Francisco, actually vacillated between the first two interpretations--all without any preprogrammed instructions.

Witnessing a computer making a small, but all too human decision is indeed a stirring event. Thaler has no intentions of stopping with small decisions.

Capitalizing on the stability of both his neural environments and the TCP/IP networks to connect them, Thaler hopes to one day see the launching of a world brain, a sentient, self-aware technology that would likely replace individual bot-agent technology.

"The idea of smaller independent bots may become moot. It may become the world's most powerful search engine," Thaler said.

Ultimately, Thaler believes, it may even guide human thought.