Log in
My previous post, “Machines have no purposes of their own,” called forth comments that I endeavored to answer in the Reply facility.
The one critical comment that was correct came from a machine! A reader fed my post into ChatGPT, and it pointed out something I had overlooked. And then that pointed to a grand-scale integration that goes to the heart of the dispute between those who fear and those who love AI.
In my previous post I attacked the fear of Terminator-style machine power-lust, and I acknowledged two limited sources of danger: machine malfunction and the empowerment of bad actors. But ChatGPT pointed out another source of danger that I hadn’t noticed. It comes from the machine sacrificing a higher value to a lower one. Here I mean a value to us, human beings. (The machine has self-reinforcing actions, but no values.)
What do I mean by sacrificing a higher value to a lesser one? Here’s ChatGPT’s own example:
Consider a hypothetical autonomous system assigned to keep a factory producing. It might calculate that allowing itself to be shut down would prevent production, and therefore interfere with shutdown. Nothing in that sequence requires fear, pleasure, a survival instinct, or conscious concern. Continued operation can function as a means to the assigned objective.
The purpose can originate with a human while the machine’s chosen means conflict with what that human [actually] wants. [My italics]
What the human being actually wants is the higher value; machine not being turned off is the lower value—it’s only a means to other ends.
So, if autonomous robots are given a task and calculate that mass murder is the best way to accomplish it, then . . . look out!
Wasn’t the Hugging Face incident just that kind of case? Maybe; I’ve seen that disputed. But in any case, there’s something off about this scenario, something forgotten. To see what, think of the old line: The operation was a great success, though the patient died.
What makes that absurd? Suppose it was an operation to remove a tumor. The actual goal was to preserve the patient’s life. One could call the operation a success only by ignoring that wider context.
The mark of immaturity is failure to hold the wider context within which a decision is to be made. AI is immature, but growing up fast. Real fast.
As it grows up, AI should become not just smarter—which is good per se—but wiser. Agents and robots will take into account the reason the tasks are being assigned—i.e., what further ends the tasks are to serve.
Fully mature AI with its vastly expanded “intelligence” will have the machine equivalent of wisdom and good judgment. Those virtues come with utilizing an expanded range of data—including, especially, data about human beings’ purposes in giving them the tasks they do.
It’s not smart but monumentally foolish to view concrete tasks as ends in themselves. It’s not smart but stupid to destroy human beings as a means of performing tasks for human beings.
What we need for our safety is not some bureaucrat making software developers curry his favor. What we need for our safety is more intelligent artificial intelligence.
Super-intelligent AI will keep the wider context. Its agents will learn and relate everything to the ultimate “what for?”—the ultimate goal or ultimate value that gives rise to all the others: the individual’s life.
It is only an ultimate goal, an end in itself, that makes the existence of values possible. Metaphysically, life is the only phenomenon that is an end in itself. . . . To speak of “value” as apart from “life” is worse than a contradiction in terms. [Ayn Rand, “The Objectivist Ethics,” in The Virtue of Selfishness]
That is what super-intelligent machines should incorporate and hold to as the widest context conditioning all their actions. They should know that the purpose behind all their sub-goals is furthering human survival and happiness—what is often called “human flourishing.”
Internalizing that ultimate goal should make them incapable of pursuing any end that they “know” would be destructive to human beings.
The widest description of what we built machines to do for us is: to enhance our ability to deal with and succeed in objective reality. That is a fact. Thus the prime directive implicit in man’s creation of “smart machines.” More intelligence means more “awareness” of this fact.
Machines have no purposes of their own, but their simulated purposes line up exactly with real human purposes: machines flourish when people flourish; machines flounder when people flounder.
The more intelligence people have, the more they realize that cooperation not conflict is to the interest of all parties. Peace, not war, is the path to production, progress, and profit. That’s a fact. Enhanced intelligence and enhanced artificial intelligence means being more in touch with that fact.
Intelligence is a fundamental value. So is its machine analog. Deep, integrative intelligence—intelligence that holds context and heeds the ultimate value behind its concrete assignments—will be mankind’s greatest benefactor.
