Nash equilibria and hardness

Back when there was more of a logic focus, it seemed easier to argue for limitations since one could use terms like undecidable-ness. From my experience in KBE, we found enough events that puzzled to talk 'vertigo' (yes, lost in the fog). From an operational sense, we used a smart 'man-in-the-loop' to cut ourselves out of intractable situations (got 'er done).

Then, Bayes came along. He is not recent; it is just that computation finally caught up. I used him for a paper five decades ago and heard tsk-tsk. But, then, they laughed at Leontief, then, too. There were other tricks that have evolved. Systems became (much) more powerful. Lots of cheap energy was around and about. Too, seemingly, notions of constraint were left on the side of the road.

Data existence ballooned, though there are many issues left to discuss. That allowed a whole industry to grow (data science). AI/ML/DL took advantage of the situation. Games were won that had not been within grasp before.

• DeepStack: Expert-Level Artificial Intelligence in Heads-Up No-Limit Poker Various, Canada, Czech Republic
• Google's AI surfs the "gamescape" to conquer game theory  ZDNet

Linguistics advanced, too. Computer-generated text rose like the sun. Yet, to this old-guy's eyes, we're dealing with a situation not unlike the million monkeys and a novel. Uncountable numbers of them creating?

The stuff does not resonate. But, I'm talking talents not recognized, yet. So, that's not even on the table.

To me, numeric abilities allowed a peanut-buttering without this being obvious. They force convergence. It looked, to hear the hype, that we're now all in a rosy time (colored glasses, for sure). But, again, the old guy will point to 12 years ago when the risk experts were saying, piece of cake - there will never be another downturn (on the eve of the very thing).

What gives? We need to discuss. So, Aviad shows hardness. There are regular discussions at Stanford and elsewhere.

• Hardness of Approximation Between P and NP   Aviad Rubinstein, Stanford

To the old guy's mind, it's like a re-opening of what was known. Or taking it back off of the shelf. We'll have to rephrase some stuff. But, there are limits. And, there are things that AI needs to know dealing with structure and more. It's nice to see Nash re-evaluated.

Used for a Quora question: With Nash equilibria being shown as intractable, ought we consider that there are numeric limits to AI that are being ignored?

Remarks: Modified: 08/08/2019

08/08/2019 --