Saturday, May 20, 2017

Predicting Me


I'm actually quite important, I think. Some of the world's biggest companies are spending billions on trying to profile and target me. They want to know all about me. 

And now broadcasters seem to care about what I actually want to watch. (Commercially I would want to build the most loved content service that no one wants to watch, since this is commercially optimal - sports channels broadcast 24 x 7 but viewers only want to watch 4 hours of football a week: my viewing habits are so esoteric that I might be difficult to target through content tags and categories: jazz, rugby, Scandi Noir).

Sorry... What do you mean ? They're doing the same to you ? We're all being profiled! But I thoughts I was special! That's why I ticked all those boxes and gave them all my data...

Drat. 

But, of course, my data is more special. I'm richer (well I think I spend more). Or will their profilers see that I actually stay at home, shop at Tescos, cook a lot and really spend very little ? It won't know that I own a few properties outright, so will think I'm poor since financial activity equals wealth, because they measure everything from their side of the equation. Paying mortgages and debts gives them data points. Sitting at home watching Freeview doesn't.

And the fact that I barely have any income (as an entrepreneur espousing salary is the most efficient form of investment) marks me as a pauper, not the paper millionaire I am (and that term means very little these days).

Netflix spends programming time in predicting what I'd like to watch and I, myself, provide spurious data to the likes of Apple Store and TripAdvisor.

Online I research and sometimes buy stuff, then still see the ads six months later: yeah, nice and efficient programatic. Makes loads of money for the likes of Google and makes life easy for the lazy ad agencies.

All of this technology, all of this effort and all the bad ad guys are encouraging by measuring all the wrong things. As Trump might say. Bad data. That data is so bad... Such bad data...

The problem with models is that if you miss one data point, it's futile. All computer models are futile and AI makes them worse, not better, since it generalises where specifics would be preferable.

And that's why those marketing to me are so bad and inefficient, and why these so called 'AI' engines are so bloody dumb and useless in reality: like factories they produce cars in any colour as long as it is black.

Hey you, focus on me, not you and your technology!