MLU: Why is it important that we understand how our brains function?
SG: In AI terms I think it's important because the brain is the only general-purpose intelligent machine we know of. It could be that there are many different ways to make intelligence, but given that there are an almost infinite number of ways to make a machine that's stupid it's like looking for a needle in a haystack. It seems so obvious to me that we should start our search in a place where we already know that something suitable exists.
In broader terms it's hard to mend something when it breaks unless you understand how it works, and broken brains are very distressing to their owners and those around them. It's also hard to pin down many moral issues (such as when life starts or stops, or which creatures deserve rights) unless you understand what minds are and where they come from. Anyway, how dare we call ourselves Homo Sapiens when we don't even understand how our wisdom arises? It's true that most people use video recorders and computers and electricity without much of a clue how any of it works. That's a travesty. It's disrespectful. But at least somebody somewhere knows these things. When it comes to the brain -- the most important machine in our lives -- we don't even understand its basic principles of operation yet. It's embarrassing.
MLU: Several well-known futurists are predicting that machines will achieve human-level intelligence early in this century. Is AI the last thing humanity need ever invent, as they assert?
SG: I think it's possible that we'll have human-level intelligence some time soon -- very unlikely, but possible. But the futurists' predictions are made on the basis of present trends, and present trends are pretty much meaningless in this instance. I liken it to trying to reach the moon by jumping. Last year you could jump so high; this year you've learned to jump twice as high. If present trends continue, you'll reach the moon in no time. But present trends won't continue -- no matter how hard you try, you can't reach the moon by jumping. It requires a completely different method of propulsion. AI is like that: fifty years ago we had only just started; today we can make computers do several things that humans use intelligence to do. If present trends were to continue, we'd get to human level intelligence in a few decades. But present trends will NOT continue. The methodology and paradigm we have now simply won't work. We've been cheating by doing things the easy way up until now, like someone jumping for the moon. Quoting Moore's Law is meaningless, incidentally: computer power is not the limiting factor. What we need here is a breakthrough -- a completely new way of looking at the problem -- because none of the old ways work. Breakthroughs can't be predicted. You can't say you're 75% of the way towards a breakthrough.
