Supercomputer Can Predict the Future

[Dave]

http://www.bbc.co.uk/news/technology-14841018

So far, they have only predicted events that have already happened by feeding it information but the idea brings up a lot of questions in my mind.

I'd first like them to predict something that hasn't happened yet. Until they do that it isn't really up to the claims.

I think that the level of information gathering and its accuracy/validity will be key to this working. We are getting information about current events quicker than ever before, but it comes in so fast that it is often wrong - several people have been incorrectly announced dead on Twitter.

I also can't see how they can predict something as complicated as human actions when they can't predict the Weather with accuracy more than a day or two ahead, and even then still get it wrong.

Also, I'd be worried that some 'disasters' might become self-fulfilling. With the heavy reliance of the stock market trading on prediction software, some of the peaks and troughs have been higher/lower than they otherwise might have been.

 

[Dozmonic]

Very much the sort of thing neural networks in AI are good at doing but, as you say, it is very reliant on being given the correct inputs. Pattern recognition and prediction of "chaotic" algorithms are quite different. Weather can be pretty accurately predicted short term, but over the course of days as small variables can make large changes then the predictions become less accurate. Pattern recognition will only work in instances where something similar has happened in the past. They wouldn't, for example, have been able to predict 9/11 because there were no precursor news stories that'd have led them there. Whereas civil unrest happens over a period of time and is reported a lot, so they could watch for the build up of that.

 

[Metryq]

Weather isn't really "predicted" anymore than this computer "predicts" things like revolutions. Local weather is like a boulder rolling down a hill—"predicting" that it is coming down is no great feat, and the boulder may swing a little to the left or right on its way down based on the irregularities of its surface.

Before weather satellites, we were all standing at the bottom of the hill, and when the boulder arrived, it arrived! Now we have a lookout at the top of a nearby hill who can see the coming boulders further in advance, but the lookouts do not "predict" them.

 

[Vertigo]

I'd say the problem with the retrospective predictions they have done is that even though no information on the actual event is given to the computer, they knew what kind of information to give it. To predict a real future event you would have to know what the event is to be able to feed the right kind of info. I suppose this might be useful if you already suspect something is going to happen but otherwise I don't think it would be much use.

 

[Dave]

I heard about this new Robert Harris book on Radio 4 this morning: The Fear Index. It seems to be about the latter point I made, except I would be much more concerned if we were using similar machines to predict future events rather than stock market fluctuations.

It seems to be a bit of a change for Robert Harris who's books have been historical and concerning WWII or ancient Rome.