Practical Ethics

When will we have proper AI? The literature is full of answers to this question, as confident as they are contradictory. In a talk given at the Singularity Institute in San Francisco, I analyse these prediction from a theoretical standpoint (should we even expect anyone to have good AI predictions at all?) and a practical one (do the predictions made look as if they have good information behind them?). I conclude that we should not put our trust in timeline predictions, but that some philosophical predictions seem surprisingly effective – but that in all cases, we should increase our uncertainties and our error bars. If someone predicts the arrival of AI at some date with great confidence, we have every reason to think they’re completely wrong.

But this doesn’t make our own opinions any better, of course – your gut feeling is as good as any expert’s; which is to say, not any good at all.

Many thanks to the Future of Humanity Institute, the Oxford Martin School, the Singularity Institute, and my co-author Kaj Sotala. More details of the approach can be found online at http://lesswrong.com/lw/e36/ai_timeline_predictions_are_we_getting_better/ or at http://lesswrong.com/lw/e79/ai_timeline_prediction_data/