What, investors are beginning to think that maybe AI is more hype than substance? A really expensive way of doing not much of anything?

Didn’t this also happen with the dot.com meltdown? A few good ways to productively use it, a lot of ways to wastefully use it.

You would think they would have learned.

The truth is, AI should not stand for Artificial Intelligence, it is in reality Algorithmic Intelligence. Same computer processing, except that it is thousands of computers in a single chip all processing the data at the same time, with each computer doing the same math (a*b+c) but each simultaneous computation uses different input numbers. Great for doing massive simultaneous calculations, but certainly not ‘revolutionary’. Instead of one computer doing one program taking a thousand cycles, a thousand computers doing the same program taking one cycle. What IS revolutionary are the algorithms that are being developed, and the applications these massive calculator arrays are being used for. Some will bear fruit, most will not.

Let’s take a ‘for example’.

Weather predictions for the next day. You start with a given set of data, then run your predictive algorithm a thousand times, each with a slightly different set of possible future data points. One of them is probably the most accurate. The problem is to decide which one. The second problem is that it takes a week to do all the thousand runs. You can have a 95% accurate prediction in one week for what the weather is going to be tomorrow. See the problem?

But with the newest AI chips, you can do the thousand different scenarios at the same time. Also, you can calculate a probability for each scenario at the same time. The scenario with the highest probability is the most likely. Then you check the actual weather, feed it back to the system, and it gets better at calculating the probabilities for the correct scenario. You get tomorrows weather today, and the system gets better and better at calculating the probabilities. This is a useful application.

Second condition. In a lottery, you can use this same server to predict all of the potential combinations of numbers, before the next draw. Except that the probability of any one combination being drawn is, well, still equal. There is just no way the system can get better at selecting the correct random number. The selection of which combination to use is still random, from equal probabilities. You can calculate what all the potential outcomes can be, but this tells you nothing if they are all equally likely. So it tells you nothing. A completely wasteful application.

Yet there are far more companies trying to use ‘AI’ to do the second condition than for the first condition. After investing a fortune in hardware, they still have the same accuracy in decision making as before.