One way to avoid the inevitable robot uprising

I took an AI class back in undergrad, more than seven years ago now.  We never really discussed ethics, but that's in large part because the field wasn't far enough along.  The trouble is, there may not be that bright a line for when we are.  When an AI can impersonate a human conversationally, passing the Turing test, it will still probably be via the use of a range of heuristics and trickery.  However, the human brain also uses such short cuts.  It seems to me distinctly possible that we still will have highly limited understanding of human consciousness by the time we start to get to real AI.  Without that understanding it will be difficult to say what qualifies as consciousness via software.

So how long do we have.  The Ian Gross of the Global Strategy Institute at CSIS covered one estimate:

But, this article argues human-like computation in robots will be realized within 30 years or less.

Hans Moravec, a robotics specialist at Carnegie Mellon, speculates that a computer would have to be able to compute 100 trillion instructions per second to emulate the human brain. That means that to perform like a human brain a “futuristic” PC would have to be 1,000 times more powerful than the average PC produced in 2008. This bridge between current computer power and that needed in the future may seem large. But, keep in mind that computational power doubled each year in the 1990s. Experts believe the doubling of processing speed every year will continue to grow exponentially well into the next decade.

Living in a world where computers are as smart as and can learn as quickly as humans will change the way people live. According to Moravec, the rise of robots will lead to a “fundamental restructuring of our society” (sounds like our friend Ray Kurzweil, no?). Business practices, scientific advances, and even athletic competitions could all be held without people. Based on the articles, it appears as though human-like robots are truly right around the corner. Whether people in society will be willing to accommodate their integration is another issue.

How to prepare?  It seems to that one logical starting point would be to set a few easy to judge threshholds.  One would be any algorithm that has passed the Turing test or any modification on such an algorithm.  Another could be any learning software on a hardware platform sufficiently powerful to perform 100 trillion instructuions per second and thus emulate the human brain. I don't think such research should be banned, but once those threshhols are hit the designers and programmers should start talking to human subject review boards and bring in all the processes and protections relevant to human experimentation.  I'm not knowlegable or wise enough to suggest what boundaries these boards should lay down, but I feel that such protections should come in at the front end of true AI and not as an afterthought.