Excellent article. I wish democratic governments well in resisting the growth in power of big tech, and historically this has succeeded by following the money, so identifying how big tech makes its money, and where that is a public good like vast databases on individuals, taxing it, regulating it, nationalising it, splitting it. For a start, compelling publication of algorithms to a regulatory authority.

And?? So what should we do?

I think it is time for Constitutional Amendment Number 28!

One which says as this is a nation, Of, For, and By humans, and thus no digitally created laws may be binding.

The article above says the entire AI/SocialMedia/Tech are overlapping with government, and THUS we need laws to keep it where any law, in the broadest sense, MUST be created and enforced, by duly appointed, or elected, Human officials. By digital laws I possibly include algorithms which generate and manage any content, or metadata, which is private by law, or custom, and that controls that which influences human behavior, choices, manipulates citizens, or otherwise causes them to behave in ways they would not have without these artificial algorithms – should be limited.

Humans controlling Humans should be the law.

I wonder what George Washington would have thought of this. He didn’t agree with political parties, with good reason. This is what he said in his farewell address, and it is even more valid today:
“The alternate domination of one faction over another, sharpened by the spirit of revenge, natural to party dissension, which in different ages and countries has perpetrated the most horrid enormities, is itself a frightful despotism. But this leads at length to a more formal and permanent despotism. The disorders and miseries which result gradually incline the minds of men to seek security and repose in the absolute power of an individual; and sooner or later the chief of some prevailing faction, more able or more fortunate than his competitors, turns this disposition to the purposes of his own elevation, on the ruins of public liberty.”

“…The logic by which an AI reaches its conclusions is impossible to reconstruct even for those who built the underlying algorithms…”

An aside from the politics of the discussion:

Neural networks have a peculiar parallel to human expert decision making and knowledge, which is little discussed – certainly no non-tecchies I have discussed this with have ever fully understood this point – they will look at you blankly if you raise it. However, tecchies who think about the nature of knowledge and algorithmic vs human decision making, and the question if all human processing is ultimately algorithmic or not, will know instantly what I am talking about. Like neural nets, top-end human expertise equally cannot tell you exactly what it does to reach the conclusions it does. An illustration of this, is that there are a small number of traders who will make money in all circumstances and all kinds of markets. However, when their professed methods are turned into algorithms, they significantly underperform the trader. This happens even if the trader instigated the process of turning the methodologies into trading system algorithms. Investigation of the point at which the trader and the algorithm differ always yields an extra layer of “rules” – ‘oh yeah, if a is happening and b or c is looking like the case, then in that circumstance you should look at x and do y and z instead’ – and there is *always* another layer, it’s like pulling on a string that keeps coming.

This says several interesting things, one of them being, it seems not only can human experts not fully articulate the deeper levels of their expertise, they are in effect confabulating stories about what they do, even to themselves.

Neural nets appear to be some kind of multi-dimensional function approximators, bypassing mass scale number crunching, and instead arriving at sophisticated heuristic answers – in this at least they appear to be doing something similar to what human expertise does.