Stephen Hawking and Elon Musk made headlines this year by raising the specter of super advanced Artificially Intelligent machines taking over the world. The idea goes something like this:
1) Today humans dominate because we are creative and can make complex decisions.
2) Computers/Robots are “mindless” servant that do our bidding.
3) In the race to make things “more efficient” we will eventually make AI that can imagine and create.
4) When that AI does that it will realize it doesn’t need us anymore.
5) Bye bye, humans.
So my first reaction is… what’s the problem? I mean from a universal evolutionary standpoint, there’s not really any reason to believe that humans are special. If a meteor can wipe out the Dinosaurs (who dominated FAR longer than we have) in a few geologic seconds; why can’t we get wiped out by the results of our own making. Sure.. it is bad for HUMANS, but that doesn’t mean that it’s some kind of universal evil.
But let’s be more practical. Let’s assume that the Star Trek vision of the universe is “objectively good” or at least desirable for humanity. Does AI really threaten that and if so, how?
Between here and there
I think that the conversation has become far too black and white. On the one hand “AI will always be our loyal slaves” and on the other “eventually they will wipe us out.”
I’d like to talk a bit about the road in between starting with the very near future… say 50-150 years.
The last 150 years has been marked by “machines” and technology replacing base human labor. Robots build cars, stuff boxes, package things, cut down trees, break rocks, construct buildings and so on. As that storm has raged, humans have become more “efficient.” That is, it used to take many thousands of people many years to build a pyramid, but a few hundred people can build a skyscraper in a year or two. The side effect of this, of course, has been that humans have to compete with robots for labor “wages.”
So if you own an auto plant you have to decide if paying someone $x/hr over the course of 30 years is cheaper or more expensive than buying a robot for $x and maintaining it. That pushes car making (and everything else) to cheaper labor costs (robots or cheaper humans). The reason why this is different than before is that the labor wages are not moving between people but from people to machines… and thus, the owner of the machines. The flip side is that the cost of those things also drops dramatically so as long as there is other work, that displaced worker can live cheaper. And thus over the past 150 years the cost of things has generally reduced while the “productivity” has gone up.
Ok. That is a very primitive summary and you could fire a bunch of holes into it, but it’s not really what I want to talk about… it’s more setting the stage, so bear with me. My main point is that machines in the past 150 years have primarily replaced human physical activity as opposed to human mental activity.
The Next 150 Years Looks Quite Different
I think that is going to change dramatically.
The next stage in the eternal quest for efficiency and productivity is to reduce the cost of mental activity. We can already see this happening with things like self-driving cars. Driving is a suboptimal human activity. It’s boring, frustrating, dangerous and doesn’t really create value. If we can get robots to do it for us, GREAT!. Of course, that means we take out the people who make a living driving, but that’s just the natural cost of the ongoing technology evolution, right? Those people can learn to do something more useful, more fun, more helpful, and more interesting. Maybe they can start a blog or something…
Other areas also seem pretty obvious… how about ordering fast food? Or how about checking out at the grocery store? Or what about trying to find a special tool at Home Depot? Imagine if there’s Wall-e the helpful robot tool guy and you ask him, and he cheerfully shows you exactly what you need. No more listening to someone drone on about some uninteresting thing, no more waiting while he finds the person who might know the answer, no more “hmm… I’m not sure if we have that, let me go find out.”
Restaurants could also benefit. Who needs a moody waiter or an overstressed kid on their first day screwing up your order. Androids would always be polite, accurate, and efficient. You can even just pay with your iPhone and not even leave a tip!
Of course all those people lose their jobs too… but of course, they can find more productive things to do. Or so we hope.
But Wait A Minute…
I think we’re actually going somewhere else… maybe a step “further” or maybe a different step.
Actually I think robo-waiters are not going to be so easy… because when I go out to eat I VALUE the human interaction and while you can AI that pretty effectively, it’s very hard to AI the warmth and care. If you have a favorite restaurant you may know the waiter’s family (through pictures), you feel happy seeing him, and if he screws up your order a bit, that’s ok because he probably also remembers something or can notice when you’re down and cheer you up. These “semi-intimate” service jobs have a very high value component that goes beyond getting what you want quickly.
Same with the Home Depot person. She may ask you about what you’re trying to do and empathize with your frustration of not knowing how a toilet actually flushes. She may take time to explain it, crack a joke and make you feel better. That’s a pretty hard thing to AI away.
But… I think there’s another kind of service industry that WILL benefit from AI… and it’s much scarier when it comes to “retraining people.”
Let’s think about areas where errors lead to real consequences. Like, say, surgery? Or maybe military tactics? Or police work? Or perhaps judges? What about deciding whether a drug should be approved or not? Those are areas where AI androids might do BETTER than people and as a consumer I don’t care as much about the human side. When I’m knocked out on the OR table and someone is going to cut me open… who do I want? The human surgeon who just got out of a law suit and found out that his wife is cheating on him… maybe he had a drink after a 12 hour shift and was called back last second? Or do I want the feelingless, tireless and perfectly accurate robot that has millions of historic cases of surgery it can recall instantly to make the best possible decision and execute each cut perfectly?
Hmm… I think I want the robot.
Surgery and Driving are more similar than Waitering and Driving.
If I run a hospital, this is good news. I am pretty sure that when androids start doing medical procedures, malpractice rates will go DOWN. So all we need is one or two human professionals to make sure that the machines are working well and we’re good. Of course we still need the hardcore research but that’s a small fraction of human labor. So all you people working hard to become surgeons and make big bucks… sorry! You probably want to learn to be a very good waiter instead.
What about other parts of the medical field? General diagnosis, taking blood, urine samples, dispensing medications, etc. I think Androids are better at that too. Once the have all the information and cases available I think they will be more accurate at diagnosing than humans. And I think they will be BETTER on their own than with human/machine interaction – contrary to what expert humans will say of course :).
Why? Because these are classic areas where humans will be overconfident, bias in their anchoring, not have all the information, etc. What AI can’t replace is a person who shows empathy and compassion… but that’s a very different kind of job…normally more part of “nursing.” The doctor is the jerk who knows everything and the nurse is the nice person who cares. I think the jerk who knows everything is much easier to AI away. I need a nice person who makes me feel good, but I want the AI doing the diagnosis and deciding the treatment.
Of course, people will have to get used to this. The first generation will rebel and want the “good old days” but assuming the outcomes are superior, the new technology will win out.
I dug into medicine, but I think anywhere you see jobs that require calculated decision where bad outcomes are actually dangerous, you will see AI take over. And that’s a LOT of people. AND they can scale very quickly. A single AI can probably diagnose hundreds of people at the same time, given that we have enough input devices (some kind of tricorder, or maybe an army of humans who just hook you up to some machines). In essence the humans in these areas will do what the AI says, not the other way around.
Think about your own job. Does this describe what you do? Well… maybe not for long.
While I think we are very far from AI “taking over humanity and wiping it out” I do think we will face the very real challenge of high-risk outcome, judgment based jobs being replaced by AI. And those jobs ARE interesting and productive for people… it’s just… we’re not as good at them.
So we will face a different problem. Is it BETTER for people to be less efficient and have worse outcomes in exchange for being more professionally satisfied, or is the race to improved efficiency and better outcomes the more important goal?