Hail The The Machine Overlord-In-Chief?
Greg Fish
91

Good analysis of the difficulty of taking current AI as the ultimate manager. The biggest current problem is 1) the algorithm being used and 2) inaccuracies in the training data sets. The first problem is an issue of secrecy allowed to protect the algorithm as property. That also often now applies to the data sets also. The second is problem may be caused by unintentional data distortions.

Fortunately growing awareness of these critical points of failure will produce solutions. The very first applies to human managers we now have and that is transparency. I’ll come back to this in a minute. You identified the primary solution and that is machine learning from primary sources. The logical way around this is more, not less. We need open algorithms with censorate (an imperial Chinese technique) machines that monitor constantly against human approved data models.

But even this is not a executive but a high level operations manager as day to day advisor. This is doable, critical and needed now. This is happening now anyway but with no oversight and no transparency. Thousands of people are the objects of administrative actions dictated by AI working from private (secret) data sets using private (secret) algorithms. You’re way too late. . .

We need to get control of this but it is already beyond our ability to understand how the AI based decisions are made. Humans simply can’t replicate the massive data manipulations. We’re going to need to establish complete transparency of algorithms and, at least aggregate data models. And, yes, the machines will need to see everything but effectively they already do.

The trade off is survival. Without a steady takeover by AI we will begin a rapid die off from climate disasters and related human driven chaos. We are already beyond our ability to politically manage current complexity from tech induced social change. In that sense Trump is a nasty precursor (kind of sentinel bleeding in a terminal cancer patient). It’s an ugly analogy but it’s an ugly situation.

So, yes, your critique is accurate. There is no way to have an AI executive at any level yet. But that is what we must prepare for because there is no other option. The decision has already been made and no one is going to stop it. Like everything else at this point we have to figure it out before things get any farther out of control.

The next stage requires heavy human augmentation. Then we can step back in charge. But we probably won’t want to. . .