The problem of artificial intelligence is not that "Hollywood" conception, but rather!

The risk associated with the widespread use of artificial intelligence algorithms is far from world-dominating scenarios ... etc.

The real danger is the "normalization" of reliance on algorithms in making decisions, which means the disappearance of the party that can be held responsible when something goes wrong.

Why can't I get a credit card?

"- Sir, we do not know. The regime decided that."

"- Who will bear the millions that we lost in the previous deal?"

"- Sir, we do not know, but it was the regime that suggested it was a profitable deal."

- Why did you bomb the hospital even though it was not a military target?

"We, sir, did not bomb it, the regime did that."

While this risk is general and not related to AI algorithms specifically, the amount of confidence that the non-specialist gives to anything whose description contains the term "artificial intelligence" exceeds that which it gives to other algorithms, which means that it will tend to enable them to make more dynamic decisions. " ususally.

Ironically, traditional algorithms are more trustworthy than AI algorithms - and I mean here specifically those that rely on data consumption to reach decisions - because our understanding of the first type is almost complete, while our understanding of the second type is far from that.

For example, if you are trying to write an algorithm to calculate the shortest distance between two points on a map (the shortest path problem) and find that you are getting a strange result, it is easy to analyze what you wrote, find out where the error occurred, and then fix it. On the contrary, if she works on a convolutional neural network and finds that she insists on classifying a cat as an electrical washer, then within our current knowledge it is impossible to know why she does this, and therefore it is impossible to know what we must do to fix the error.

In other words, we do not yet have a sufficient theoretical understanding of this kind of algorithm. This does not mean, of course, that we have to "demonize it"; All that is required is not to rush to expand its use in vital sectors.

But with all the media exaggerations and hype associated with the industry, this is unlikely to happen.