Research into Industry Seminar: 14 October 2009
What Do Machines Think?, Peter Cochrane
All discussion of Artificial Intelligence (AI) has, for decades, been mired by one important and immutable fact; we have no description, definition, quantification or measure of intelligence. We simply cannot describe or say what it is, and any attempt at an accurate comparison of systems, and especially biologically based systems, is mired and generally meaningless. The best we can do is to draw on some general concepts of neuron count and connectedness.
But not understanding something doesn't mean to say we cannot use and exploit it!
This situation has, in no small part, led to some discrediting of AI, which is often seen as not delivering on its early promise, or indeed in the minds of some, any promise! However, such a view is short sighted and missing the essential point that machine-based intelligence now supports the human race and all human life to an extensive degree. And this dependency isn’t just in the manufacturing and aerospace sectors, our species now has a direct and growing need encompassing healthcare, food production, transportation, banking, stock trading and much more.
I think we can safely assume that if we were to switch off all the operational AI systems on the planet, our very civilization would be at great risk, and in some significant manner, it would fail!
In this presentation we examine intelligence from both a silicon (designed) and carbon (evolved) point of view, and move on to our relationship and reaction to systems that are both abstract and anthropomorphic. We then address a most fundamental question; what makes something intelligent? Surprisingly perhaps, sensors and sensory systems appear to play a far more important part than memory and processing power. This contrasts with the popular "game show perception" which sees information recall as a measure of intelligence, which by any scientific measure, it is not!
Using a series of logical delimiters we move on to develop a measure of intelligence fundamentally based on a process originally developed by Frank Drake @ NASA to estimate the number of civilizations in our galaxy. We then hypothesise the final form of a complete descriptor for a generalized intelligence.
Additional support by ...
Leeds Compsoc, The University of Leeds
Computer Society 