On Intelligence

Much time and capital is expended toward the goal of developing a machine intelligence that could perform many of the thinking skills currently done by people. The motivation for this expenditure is obvious, especially when one looks at the savings already accomplished through automating physical labor. Human laborers bring a tremendous overhead to every employer. The fewer humans on the payroll, the lower the business overhead in the long run.

Most of the accomplishments in the machine intelligence field have been mere "nibbling around the edges" work. Through the last half of the twentieth century most researchers in machine intelligence had done a poor job of defining what type of intelligence they were trying to replace. The results were abysmal on the whole.

In the book "On Intelligence" Jeff Hawkins, founder of Palm Computing and Handspring, presents his theory of human intelligence, with the aim of developing a machine intelligence patterned after the basic abstract concept of the human neocortex. Go to onintelligence.com to get a closer view of the contents of the book, and to read more about the underlying idea. Hawkins has even founded a new company called "Numenta, Inc.," to pursue the goal of MI.

I will not try to do justice to the ideas in the book in a short summary. I will merely say that anyone with an interest in the underlying bases for human intelligence would probably enjoy looking through the book. Hawkins is insightful and resourceful. He is the type of investigator who will crack the code of the neocortex. It helps to understand basic neuroanatomy and concepts of feedback and informational hierarchy. Those concepts are fairly well explained in the book, however.

My bias is in favor of augmenting human intelligence, rather than putting an emphasis on machine intelligence. It is likely, though, that augmented machine intelligences would facilitate a number of approaches toward augmenting human intelligence. This is not too early to begin thinking about applying safeguards to this technology.

The approach that Hawkins seems to favor will not result in autonomous machine intelligence, not at first. That is a safeguard in itself. An "intelligent" machine without an internal will, does not present a threat unless humans with malevolent intent take control of it. That is true of any number of already perfected technologies.

It would be best to apply the problem of improving human intelligence to the earliest machine intelligences. If Hawkins' approach is valid, early machine intelligences will function very much like human intelligence. Eventually the paths will diverge, but at first there will be strong parallels.