In the search for Artificial General Intelligence (AGI), AI companies have managed to simulate nearly every mental illness real people are heir to. Deepmind thinks AGI will look like. Maria, the robot in Metropolis.)
I’ve seen software that’s autistic, that’s paranoid, that’s nihilistic, murderous and prone to every form of psychosis known to the DSM. AIs have been shown to be psychopaths and sociopaths. If AIs were people, they would be put in a mental hospital, the key tossed aside. While AI software may be a great tool when under proper human management, it can never be human.
How can you model the human mind when you don’t know how it works? We can see the mind’s structure and see where, in the brain, specific thought patterns originate. But we can’t follow the processes, nor come close to understanding them. We’re using up oceans, and tearing down Pleistocene forests, trying to make these things work. But your mind does everything it does on just 15 watts of power.
AI optimists are twisting themselves into pretzels, telling users how they can stop this (or at least slow it down). But that’s pretzel logic, and a demonstration of just why AI software should not be left alone, any more than the software in Desk Set, or the punch cards Doris Day fought in That Touch of Mink, the ones that replaced the army of accountants in The Apartment, should have been left alone.
Each one of our tools, from mainframes to the Cloud, required human thought to master. AI data engineers today are working almost twice as hard as before, thanks to integration complexity, tool sprawl and fragmentation of effort. I don’t know what all those laid-off Amazon managers were doing, but my guess is many will be quietly hired back.
Same As It Ever Was
AIs are look-up engines. They can only work with the data they have. They can’t go outside their programming or be truly creative. Any music an AI creates has been played before and any movie it makes has been made before. That’s the way the software works. (This is the first computer I ever tried to run a program on, installed in the basement of Rice University’s Herman Brown Hall in the late 1970s.)
AI is a tool. Just like database management is a tool, just like networking and every computer language is a tool. The only thing that has changed is the underlying hardware. Nvidia chips can process data at lightning speeds. It can Google faster than you can. That doesn’t make it creative.
The mental diseases AI suffers from are the same ones we suffer from. The difference is that AI diseases aren’t treated, they’re covered up. A human being’s mental diseases can be understood and treated, with therapy and with drugs. An AI’s mental diseases are only masked, and many AI programs are very good at ripping these masks off.
The Good News
In Young Frankenstein, Marty Feldman’s Igor gives Frankenstein the brain of “Abbe Normal,” after dropping the brain of Hans Delbruck, the “scientist and saint,” at the brain depository. What’s great about the picture, what distinguishes it from the original, is how the scientist then makes the monster into a productive citizen, able to sing, to dance, and (eventually) to give a speech that turns the villagers away.
That’s fiction, but that’s the task ahead. There’s an awful lot of human work needed to model the human mind, to turn biological models into mathematical ones, and to get the most help we can from what we already have. That’s human work, real jobs, not machine work.
The real monsters are the AI “Masters” who insist they’re going to replace us all with software. They’re not. Just as with previous computer breakthroughs, we’re going to see giant productivity leaps. There will be more work being done with less human effort, making more people available to do the big jobs that must be done to save humanity.
The coming AI Crash will clear out the morons. Our kids and grandkids will be left with new tools to meet the coming challenges, including the demographic fact that we, their elders, are going to die.
