Beyond AI

Plato the bot

Enthusiasts of artificial intelligence laud the benefits it’s about to bestow on humanity. It will find a cure for cancer. It will solve long-standing puzzles of science. It will take the drudgery out of work. It will make us wealthy beyond our dreams.

AI may indeed do some of this. But the boomers are bound to be disappointed. Artificial intelligence is overrated because intelligence itself is overrated. When we humans are unhappy, when we are divided against one another, when we struggle to make progress on crucial issues, it’s rarely from lack of intelligence. It’s because we can’t or won’t choose among competing values.

All the intelligence in the world, human or artificial, won’t get right-to-lifers and right-to-choosers to agree on abortion policy. They start from different premises and can’t help arriving at different conclusions.

One person thinks government programs serve vital purposes and is willing to accept the inefficiencies of bureaucracy. Another thinks the costs of the inefficiencies outweigh the benefits of the services provided. Neither is more intelligent than the other. Their tolerances simply differ.

You value the dynamism of capitalism and treat the consequent inequality as an incentive to personal betterment. I worry that inequality undermines democracy and good governance.

A measure of the irrelevance of intelligence to these problems is a simple test. Pick a position that you hold strongly. Can you imagine being talked out of that position by more information more intelligently conveyed? For a few people, an honest answer might be yes. But not for many.

When the matter of intelligence enters political discussions, it’s almost always to the effect that the other side lacks it. They’re too stupid to believe what I believe.

Consider this: If somehow you could decide for or against permitting the other side to get the latest model of AI, would you let them have it? If you think they’re deficient in intelligence, you should, because if they become smarter they’ll agree with you.

Again, some will say yes. But many would deem this equivalent to giving better arms to the bad guys.

In most problems that vex societies, the shortfall is not in intelligence but in wisdom. Artificial intelligence can be useful in many circumstances, but what we really need is artificial wisdom.

What would AW look like? How would it differ from AI?

Perhaps most strikingly, it would know what it didn’t know. AI doesn’t seem to hallucinate as much as it once did. But it still has trouble admitting ignorance. The first characteristic of wisdom, in humans or machines, is appreciation of its own limits.

Second, AW would seek multiple perspectives. When AI is set a task, it sets out single-mindedly to accomplish it. If one route fails, it tries another. But once a path is found, AI declares its mission accomplished. AW would realize this is just the start. It would examine the problem from a different angle. And it would keep all the viewpoints in mind as it prepares to proceed.

AI employs large language models (LLMs) that give different weights to the probabilities that different words follow the sequence already determined. AW would give different weights to the different viewpoints it discovers. Some views are broadly held. Others are deeply held. Others are mere conceits of a few. AW would take these differences into account in proposing action.

Imagine the budget committee of the House of Representatives preparing the budget. Suppose the committee has an AW bot—call it Plato. Plato will poll the myriad groups that will be affected by the budget. It will do so by searching the internet for articles, speeches and other sources of information on who wants what out of the budget. It could solicit additional input from interested parties. This input could be in written or spoken form. Plato would analyze the logic of the arguments adduced. It would tally numbers of responses. It would gauge the apparent conviction of the responders.

Plato would act, essentially, like a proxy of the budget committee itself. It would be indefatigable and, crucially, unswayed by donations or other political influences.

It would have its own AI agents that would draft and run simulations of possible budgets. From these possibilities it would choose the one that best met all the demands placed on the budget.

In other words, Plato would do what the budget committee is supposed to do.

But it would have the advantage of being a bot.

Ideally, the committee would agree ahead of time to accept or reject Plato’s recommendation in toto. No picking it apart or placing new ornaments on the Christmas tree.

The beauty of the scheme is that it would protect legislators from retribution by vested interests. The lawmakers could explain, credibly, that Plato’s proposal was the best for the country overall, even if some groups got more and some less.

Many of the problems facing us today are like the budget problem. Different groups have different interests. Congress is supposed to have the responsibility of assessing those interests and discerning what’s in the best interest of the country collectively. But often Congress gets paralyzed by the contradictory pulls of the competing interests. And nothing gets done.

Enlisting Plato could help things along. The real Plato recommended giving power to a philosopher-king, a leader at once wise and powerful. The bot Plato would be wise. It could help Congress be both wise and powerful.