I’m going to skip over all the practical impediments to this. If it has to get done, it will get done. We turned the entire economy over the the war effort in the early 40s and did things no one had even imagined could be done. Why? Because we felt we had no choice.
So I am thinking AI thoughts on one side of my mind and on the other, I am thinking carboni thoughts from The Ministry for the Future. [1] The carboni is a currency that was awarded to people who reclaimed carbon dioxide from the atmosphere or who prevented it from escaping into the atmosphere. There is no natural economic system that would reward that job, so they invented a currency that would and changed a lot of economic behavior that would otherwise not have occurred.
In the meantime, the AI side of my brain is trying to grapple with the fact that we keep defining jobs so that they can be done better by AI. There is no reason we could not define some jobs that can only be done by humans (and leaving others to whichever life form does it best) so that there is a balance between economic efficiency and human flourishing.
The way I have heard the problem posed took for granted that the most efficient (productive) way to provide various good and services is the best way. When that argument fails on the merits, the argument is that firms that refuse to work that way will be put out of business by firms that excel in it.
You could even define “human support jobs” in a way that would allow AI machines to outperform humans. What you cannot do is to say that jobs that are defined by human to human contact can be better done by AI. Some jobs could be made to come under the productivity metric; other jobs come under the human support of humans metric.
If that sounds like Affirmative Action to you, I am willing to let you be reminded. If the alternative were to turn out to be Affirmative Inaction or Affirmation Rather than Action, it wouldn’t sound as bad.
I, myself, think that we can be more discerning about what only another human can do for a person. I think we have not paid much attention to that form of the problem yet. But I also have a nightmare scenario that pushes me in the direction of the human care of human beings. It is that if we settle for the simulated care of human beings of which AI is already quite competent, we will begin to define what we value and what we need in terms of what it can produce for us.
Perhaps a silly example would help. Imagine that the speech programs with which they fit up AI systems are unable to pronounce the Sh- as in shibboleth. Like the Ephraimites, they could not pronounce the Sh-. Fine; use it as a test. But this was a battlefield situation. They can either pronounce it or they can’t. But if you were living with them, as many people so far have lived with robots, you would have the chance to decide that your really preferred a word that sounded like “Sibboleth” rather than the old way.
Some years ago, Sherry Turkle wrote a book with the title, Alone Together. It is a spectacularly good book, but along the lines of this essay, it is good to remember that the subtitle is: Why We Expect More From Technology and Less From Each Other. The short answer is that we develop emotional attachments to these AI devices and want to prefer what they offer us to what could be available from any other source: human or machine, natural or artificial.
The reason the system I am trying to imagine would require humans providing services to other humans is that if we do not, humans will be vulnerable to choosing the AI performance over the human one. And that is only the provision of services. There are also special skills required to receive those services in an affirming way. [2]
It seems sensible, I am sure, to lob the ball up as if it were a jump ball and allow the better performer to control the play thereafter. You could even say that we will simply let people choose which source of their “personal services” to use, the natural ones or the artificial ones.
But it is not sensible. AI systems can be tweaked to provide better service in the same way that viruses tweak their performance to stay ahead of anti-viral treatments. AI systems can perform in the way that is coolest or hottest, depending on the era; squarest or weirdest depending on the culture. The authentic response of one human being to another is not going to meet the preferences of people who allow their preferences to be messed with like that.
As nearly as I can see it, nothing short of defining personal services as services provided by one human and received by another human. That should take care of the competition between the two. It may very well, as in the carboni example, require a lot of rules and a new currency, but it seems to me that it is a step in the direction of thinking it through together.
[1] Kim Stanley Robinson’s wonderful sci-fi novel about the rescue of the planet.
[2] Giving and receiving are discrete sets of skills, neither of which can substitute for the other.

