The End of General AI Predictions?
Artificial General Intelligence is a classic sci fi topic. And it is appropriately so. What better to temp our imaginations than human–like “other” that is lacking or over abundant in certain human qualities such as empathy, purpose and intelligence. But AGI is being used in corporate predictions for economic conditions in the next 10 to 20 years. These predictions should be unplugged like Hal 9000.
Getting to AGI
AGI is the most attractive of thoughts about artificial intelligence because it is so easily anthropomorphized. The largest part of the human brain is devoted to understanding other humans. And so, when we seek to understand something that is not human, we want to imbue it with human characteristics. We see a face on the moon, we see that cats are happy or sad, rather than just conniving, and if a computer breaks, we say it’s because it hates us. AGI also has the greatest possibility: like nuclear fusion, if it could be created and harnessed, there could be almost unlimited riches for the creator. But the problem is best encapsulated in the old joke where a lost traveler asks for directions from a local who replies that although he knows the way, he wouldn’t start from here.
The problem with AGI is that we all seem to know what it looks like and what it does, but there isn’t a direct line to creating it. What we have at present is a number of Artificial Specific Intelligence (also called Narrow AI) which are exceedingly good at one or more things. ASI can beat humans at a game of chess, calculate flight paths of asteroids, determine which cat pictures you like best, provide you with relevant ads, and find information within large amounts of unstructured text (e.g., Ailira). Playing a chess game is far from a general ability to think and reason. And there isn’t a path from the creation of ASI to an AGI.
While it’s, of course, entirely possible that a company like DeepMind, which has some of the world’s top AI researchers trying to create AGI, might succeed, it might also be the case that, like nuclear fusion, AGI always ten years away. Other more promising possibilities that might have similar effects to AGI is incorporate some kind of “wetware” such as a hybrid organic computational device or even Whole Brain Emulation of the human mind to a computer. WBE involves creating a digital reproduction of every neuron in mind (without necessarily understanding what they do) such that you can emulate the hardware that the brain runs on. I am not saying that AGI will never happen. I hope that it does, but my criticism relates to the use of predictions on the timing of AGI.
Problems of Prediction
An event such as the creation of AGI is a stochastic break in a time series and means that previous prediction models do not apply. An example of the stochastic break is the COVID pandemic on previous economic models. Back on 2019 predictions for stock market, inflation, unemployment, international trade and so on up to present totally break down when there is the break that is a worldwide pandemic. The introduction of AGI would reduce the constraints on human capital (or human equivalent capital) to essentially nil. This means that the change in the economy is potentially unimaginable. The change between a world before and after AGI is the same change between hunter and gatherer and an agrarian society, then to an industrial society, then to an information age. A hunter gatherer could never imagine the stores of produce and time for developing skills that can come from having crops and agriculture. The increase in economic growth for hunter gatherers is essentially nil once populations have expanded across territories. Hunter gatherer societies can remain constant for tens of thousands of years. Agrarian societies can slowly increase economic output and exponentially more so than hunter gatherers. They reach a Malthusian equilibrium when the arable land has all been farmed. We can see with the problems of Thomas Malthus’ predictions that there is a limit on the amount of land that can be farmed and an exponential population and therefore ultimately the population must curb or humans will starve to death was broken through by the industrial revolution and the increase in productivity from technological advances. Again, the change between preindustrial and industrial society is an exponential one. It was impossible to predict the rapid economic growth that we’ve had over the last 300 years for a person who is living in an agricultural medieval society. So too is the change that will come from any introduction of AGI. That is just like the industrial revolution removed limits on agricultural output so too can AGI remove limits on human output, because the number of humans or human equivalent entities can be unlimited.
If you can create AGI that has a human like level of intelligence you can then copy and paste that program and have two. Or a number limited only by power and disk space. You could teach one of these AGI something like some kind of computer programming or scientific research or law or writing interesting blog posts. You could then have an infinite number of resources devoted towards research and development of new ideas and technology. It means that our current model of an economy as something that is limited by human capital is so wrong that it is totally irrelevant. A singularity beyond which we cannot see. Just like a hunter gatherer could not imagine a society like we have today.
Changing Our Predictions
We see predictions that some huge percentage of an industry is going to be replaced by AI or some task is going to be done entirely by AI we should remember that such prediction is based on AGI and not ASI. The impact of ASI and even more general automation over the last 100 years have been profound. But it should be noted that every job that was conducted 100 years ago is still conducted now, just perhaps in drastically changed proportions (for example agriculture). Today we still have farmers, and factory workers, and even blacksmiths. They may be fewer in number and with greater automation, but they have not been entirely replaced. They have instead been made much more efficient by specific technologies. I do understand that there is one exception to this and that is the job of elevator operator — when elevators were first created the pushing of buttons to different levels seemed to complex and risky a task to be left to the general public and required a specific operator. Elevator operators are now almost entirely redundant. Push back against predictions that involve anthropomorphised AI having radical changes to the economy. Instead theorising about the impact of AGI should remain in the realms of science fiction, where it may boldly go where no man has gone before.