Response to Food For Thought...

Thanks John - this has been one of my favourite questions for ages.

One of the big problems with the whole AI 'debate' is the perception of intelligence. When we measure intelligence, we tend to use IQ. But the mental mental of AI goes much much further than this. When we think of the Kurzweilian AI, we think of 'machines as man' - ie, as per Alan Turin's famous test where we can't tell the difference between the responses a computer gives to questions and the responses a human gives.

But this infers an emotional component, not measured (or perhaps even measurable) by IQ. In short, we don't really know what it means to be intelligent. Is someone/thing intelligent because it knows lots of stuff, or has reads lots, or knows more than me about lots of things? Or because they can judge situations rapidly, can put knowledge into context, can see many different angles on the same problem? Or is someone intelligent (in the AI sense) when they know by gut feel the answer to complex problems. Or is someone intelligent when they seemingly act irrationally in the present, but in doing so secures a safe, longer-term future.

In many of these situations, it seems hard to imagine how a machine might act like a human. Philosophically I tend to prefer not to believe that one day machines will think as humans because it takes away the notion of free will and consigns the imagination to a puzzle with a final solution that we haven't solved yet. What it means is that we, as humans, are totally understandable and ultimately programmable, and that given enough time and rational analysis, we will be able to understand 'the brain circuitry'. I prefer to believe there is more to us than that. I don't want to wait for some 'intelligent' scientist to work out how I work.

However, cutting things horizontally, we might find a definition of intelligence that machines can help us with. What I would call 'utility thinking' cries out for machine intervention. Utility thinking at work includes all of the mundane, repetitive, mind-numbing stuff we all have to deal with: expense reporting, finding out who does what for whom, escalating problems, some strategic financial decisions etc. Having machines take care of these things frees us up to spend more time thinking creatively, pondering the unknown, moving up the 'value chain of imaginative thought'. Machines aren't really intelligent in this case, but perhaps they allow us to become more so.

I very much agree with point 2 - if you don't believe we are eminently understandable and in fact that the human spirit has boundary-less possibilities then you will believe that we will still explore and leap into the unknown which means making mistakes. Whatever we get machines to do for us, we will fill the new spare time created with more creative things, discovering new 'knowledge', creating new 'intelligence',  learning from new errors.

So in summary, I think that if we don't really know what intelligence is,  it's going to be difficult to know when we've created an artificial version (how will we know/measure it?!) and if we believe in the boundary less human spirit, we'll always be fallible and make mistakes.

Cliff
www.visitcliff.com