A little advice -- don't be right too soon.
This is not sour grapes, it's serious advice. 15 months back I suggested here that we should be careful about artificial intelligence. We have to make sure that the models include -- yes, include -- the usual biases we humans have about the world around us. For it's only if the models reflect that real world that they'll ever be of any use in telling us something about that real world. This was a point I've been making for rather longer than that too -- near a decade now.
Now, of course, we're seeing that the new Google AI gives us pictures of black Nazi soldiers as a result of the programming being racially inclusive. The insistence on not regarding our real world has led to the results from the AI producing nonsense results. We're seeing similar things happening in text output, complete gibberish being generated.
Everyone's having to go back and unwind the woke they've built into those models. Well, that's good. Not specifically about woke, but about the models getting back to reflecting reality.
Regarding that “better not make your move too soon” -- if I'd said all this two weeks back, that models have to be as biased as we are, I would now be lauded as being alarmingly prescient. Book contract, TED Talks, life fat in business class. But alas there is only regret in retrospect. Being right over a year back doesn't work quite the same.
Now, the tools we use to aid us in deciding about the world must actually reflect the world as they way it is. The moment we allow the calculating tools to be overcome with political -- or ethical, moral -- desires then those very tools lose their usefulness. Even become grossly dangerous.
Sure, a black Nazi soldier image isn't about to break society. But the whole of the Google engine has been deliberately programmed to reflect that sort of thinking. We're now, only as a result of this embarrassment, seeing the internal documents which show this.
And now to the real warning point here. It is indeed true that there are good ethical and moral reasons for thinking that we humans should be less greedy, less lazy than we are. More attuned to the collective benefit of us all than the personal fortune of the individual. Which is fine, we've systems that try to impose ethics and morality upon us, religion, the law and so on.
But our economic models cannot assume that we are more communal than we are, more concerned with our fellow man than is actually the case. We have to build our descriptions of how the world works from, well, how it does actually work -- with humans as lazy, greedy and uncaring as we actually are. Only once we've built out mental or calculating models correctly can we see where to put in that lever, that crowbar, which then pushes those personal impulses to the collective result that we'd desire.
Which is the same point as the AI engines. Only if they take us as we are, and then calculate, can they produce anything useful about what we do next. The same being true of those economic models. They're only useful if they describe us, warts and all, for only once we've got the truth can we think about how to direct it.
Or, as we might put it, start from the truth then try to make it better. Not assume we're better and try to make that the truth. Political desire beating reality doesn't exactly happen. Reality always beats illusions.
Tim Worstall is a senior fellow at the Adam Smith Institute in London.
