Just in the Khosla Ventures portfolio alone, entrepreneurs already are trying to use machine learning technologies to replace human judgment in many areas including farm workers, warehouse workers, hamburger flippers, legal researchers, financial investment intermediaries, some areas of a cardiologist’s functions, ear-nose-throat (ENT) specialists, psychiatrists and many others.
I have omitted links to the companies to which he refers.
The rest of the essay argues, rather repetitively, that technology is becoming increasingly a substitute for, rather than a complement to, human labor. It strikes me as a nod toward Robin Hanson’s view of the future, although Khosla thinks that human brain emulation need not play a part.
What I would watch for over the next 15 years are developments that enable humans to evolve more rapidly, in order to compete with machines. Note that the cover story of the latest MIT technology review says (a bit prematurely) We Can Now Engineer the Human Race.
Human brain emulation looks technically too difficult. And since we can write software to do many–perhaps eventually *all*– human tasks *better* than humans do them, why bother with emulation?
Well, it’s not premature to say, “We can now *tinker with* the human race,” though we’re not doing much tinkering because of “ethical” concerns. (I think it is misleading to say, “to enable humans to *evolve* more rapidly,” if what you have in mind is tinkering/engineering rather than traditional evolution.)
I think Tyler Cowen is mostly on the right track with Average is Over. It’s not just about IQ anymore. There are lots of other characteristics I call, ‘choice CUTS’ – Complementary Uncorrelated Traits and Skills.
Things like conscientiousness, energy, drive, discipline, social savvy, charisma, ‘fashion-sense’, leadership qualities, and things that help people become what you call ‘marshmallow test passers’. Perhaps you might also add things like conformity or being particularly adept at toady-obsequious-sycophancy. It’s not always clear whether these things are part of ‘character’ or ‘personality’ or merely another kind of talent or gift.
Maybe also some rare and special affinity for producing a lot more market value than the normal person when working with the technology of your era. Most everyone can drive a car, but most people aren’t cut out to be racecar drivers.
In the past, I think being especially strong in one of these features could compensate for being average or weak in the others, and could still lead to an upper quartile or decile outcome.
Increasingly, it seems that people need the whole package to be really successful in almost any profession.
Evaluating future potential and selecting people is always a very difficult process. My hunch is that, in the past, people looked for various strengths along just a few key dimensions, which could be reasonably proxied by credentials or some simple metrics.
But now they still have a surfeit of people with those attributes and they need to filter for the additional ‘character’-type traits which are harder to observe or test for.
The solution to this problem seems to be an ‘easy in, easy out’ attrition-like strategy that the military often employs with regards to its personnel accessions pipeline. That is, you avoid acting as in loco parentis and stop spending a lot of time emphasizing character or helping students avoid predictable mistakes, and you give everybody plenty of opportunities to mess up (“give them enough rope to hang themselves”), and inevitably, most people will indeed mess up.
They’ll fail to notice the hypocrisy between what is said, “be yourself, have fun, don’t work too hard,” and what is actually true regarding the state of the competition, and when left to their own devices without a custodian or support network, they will fail to exercise the highest levels of discipline and self-control in the face of all kinds of temptations.
The people who are still able to run that gauntlet are the marshmallow test passers you are looking for. A related phenomenon is the probationary unpaid internship, the increasing prevalence of which shows us that the other credentials and scores still don’t provide selectors with the information they are really hoping to obtain.
Or also that they simply want to extend the unpaid portion of a career.
I wonder how much we are changing the work to accommodate the robots. Their brains don’t have to work like ours, they don’t have to play chess like we do, and they don’t have to fold clothes like we do. My Neato robot vacuum does a poor job but he makes it up on volume. I don’t mind. Our desires are evolving to accommodate their capabilities. There will be a lot of human work involved in figuring out these algorithms.