Uncategorized

Dinner with Peter Florjančič

It was today. He is a legend, 99 years old. He invented airbags so early, that his royalties would expire before they started to put them in cars. But earned a lot of money with his other inventions which are many. He knew everyone from Habsburgs to Yugoslavia’s royal family (as a child), Faruk of Egipt, almost every big European and some American businessmen, was a Chaplin’s neighbor in Switzerland ( as an adult)… A lot of photos of a lucky and brilliant man among the international jet-set.

He said he was lucky, I said that the luck is responsible for at most the half of it, the rest was his brilliance, which is genetic. Which is a kind of luck, too, I admit.

Many interesting anecdotes, but not as interesting as his cleverness in problems solving. Problems spotting at the first place, maybe. Where to substitute wood with plastics and make millions? How to spray perfumes? For hundreds of years, nobody bothered with this problem. But he did and solved it.

Then we spoke about AI. Which isn’t as clever as he was in his prime days, but it’s about to become. I didn’t formulate this so directly but he understood it anyway, I guess. I don’t think he believed that. Or even considered it remotely plausible for a second. I told him, that I am not promising him anything.  Only that chess has fallen recently and that when chemistry will fall too, there will be no more medical problems anymore. He expressed some deep general pessimism then, which I hate to see, for may it be a bad omen. To whom express it.

His life minus his daughter’s tragic death was as good as a man’s life can be.

Advertisements
Standard
algorithms, artificial intelligence

AI Skeptics

I am talking about those goalpost-moving crowd which says “You will be never able to make computers play chess!”. When computers do chess, they just move this goal to something else unachievable according to them, like Go. Pretending meanwhile, how trivial the chess was. It’s just an algorithm, they say.

It’s a well-known hypocrisy of this particular sect, nothing new here. But can we somehow use this pile of dishonest intellectual garbage for something interesting and informative?

Let me try! Until we have no algorithm, we have an open and “impossible” AI problem. Then, we have at least an lousy algorithm. Then we have a better algorithm. Then we have a superhuman level solving algorithm. Therefore, one day we will have the so-called AGI, when we will have one billion or more algorithms stacked so nifty, that they will trigger the most promising one among them, to solve any problem which may appear. A new algorithm will be devised when needed. In their free time, all those algorithms will be under optimization process and re-stacked often. Every aspect of this algorithm-hive will be the subject of a constant effort to improve.

And this we will call AGI. The above-mentioned skeptic club will call it “an increasingly large pile of self-improving algorithms for various tasks, nothing new”.

Fine.

 

 

Standard
algorithms, artificial intelligence

Doctors and nurses and such …

… are notoriously difficult to schedule.  The right amount of them at every moment of the day or night, each working at some acceptable pace, about the right number of hours per month, with various absences, holidays and much more is a par excellence hard to plan.

If is it more chess-like complicated or Go-like complicated, depends on a particular working place in question, but it’s almost always complicated!  Human schedulers are surprisingly good, as human chess players are surprisingly good. But only to a point when the machine with the algorithm arrives. A human is no match for the top engines.

There is a superhuman level scheduler called WoShi now. Partly responsible for the hiatus we had on this blog. We are field testing it right now.

doctors

 

Standard
probability

Coin Flipping Problem

Elo from LessWrong suggested this problem:

I write a sequence of n heads or tails on paper.

I then start flipping a sequence of n coins. If next coin does not match my sequence, I restart a new sequence of n coins.

What is the rule for the relationship between how many coins I need to flip to get the sequence I have pre-committed to and the length of the sequence of?

Discussion there:

http://lesswrong.com/r/discussion/lw/pg3/open_thread_september_25_october_1_2017/dxlt

Standard