AI Is Far From Just A Dream

§ November 30th, 2015 § Filed under AI § No Comments

In anything we do, we jump to a higher level of performance when we learn to work with patterns rather than merely mastering details. Today, a compelling application of business computing is one that has made this leap.

dcWe see the power of patterns in human performance of tasks, whether the task is driving a car or playing chess.

The driver-education student feels overwhelmed with details; the experienced driver integrates images, sounds and “road feel” to go faster with less conscious effort.

The novice chess player sees individual pieces; the expert sees the board as a unit. Experienced players can memorize board patterns more easily than novices (but only when the patterns are in real games).

Memorizing random layouts is just as difficult for chess experts as it is for inexperienced players, proving that raw memory power is not the difference. The difference is in learning to see a purpose and a direction–in the same way that we learn to recognize a whole printed word rather than having to spell it out.

In similar ways, we can improve the performance of our systems and networks with the aid of pattern-oriented diagnostic tools that are modeled on neural networks. Based on current understanding of brain processes, these tools are abstract nets of relationships, not physical connections.

It’s possible to “hard-wire” a neural net for higher performance, once a task is well-defined, just as it’s possible to replace any software with hardware if the economics make sense. Neural nets are at their best, though, in situations where tasks are continually redefined–an environment like the one that faces most of today’s enterprise IT system builders.

Neural net programming finds unsuspected patterns by analyzing data rather than matching data against predefined rules. Earlier this month, reported on early results from Computer Associates’ neural agent (“neugent”) approach to server fault analysis; on our desktops, IBM’s neural-net virus scanning technology reduces our vulnerability to entire families of malicious code.

Moving beyond IT systems, we can use pattern-oriented monitoring of news and advertising media to assess our companies’ business environments. Public relations agencies, such as Medialink Worldwide, are becoming increasingly technical in their methods of “seeing the board, not the pieces”–as that concept applies to products or industries.

Earlier this month, Medialink announced that it received an exclusive worldwide license to use the InfoTrend technology, a patented technique developed by David Fan at the University of Minnesota. “Analysts ‘train’ the technology to ‘read’ for issues and messages that are important for a particular client or industry,” said the company’s report on InfoTrend analysis.

It’s an effort for IT dinosaurs to expend the vast computing resources involved in pattern-based solutions. We’re used to the idea that limited memory dictates one pass through the data rather than holding it all in memory for in-depth study. Seeing the whole scene is the difference between doing the old stuff faster and doing the new stuff that can make the biggest difference.

Leave a Reply