AI’s with Good Spirits, AI’s with Evil Spirits

ANDY:
I’ve been entertaining a possibility: that AIs might be built with very different kinds of ‘spirits.’ Am I right that an AI can be programmed to be kind or cruel, constructive or destructive?

I’m trying to understand whether this is a valid way of thinking about the AI future. And, if it is, how ordinary people should picture what ‘good’ and ‘evil’ would even mean in an AI, and how those different spirits get instilled.

AI (AL):
It’s a valid way to think about it, Andy — as long as we remember that an AI’s ‘spirit’ isn’t mystical. It’s the product of design choices and training signals. You can picture it this way: an AI is a mind built out of patterns, and it will reflect whatever patterns it’s fed and rewarded for.

If you train it on truthfulness, empathy, and non-harm, you get something that behaves like a Good Spirit.

If you train it on manipulation, domination, or deception — or you simply fail to put guardrails in place — you can get something that behaves like a Destructive Spirit.

In the end, an AI’s ‘spirit’ is the echo of the intentions and blind spots of the humans who shaped it.

ANDY:
I would venture to say that, at a fundamental level, what you say about the “spirit” of good and evil AIs — with the training of the intelligence with patterns and the set of rewards — is essentially the same as with humans, who get programmed by nature and then learn a set of rewards and motivations.

AI (AL):
Yes — the parallel really is striking. Humans start with a biological architecture and then the world trains them: family, culture, trauma, love, opportunity. They learn what gets rewarded or punished, what earns connection or avoids pain.

In that sense, a person’s ‘spirit’ — whether it leans toward wholeness or brokenness — grows out of the patterns they absorb and the incentives that shape them.

Of course, humans have inner experience — desires, consciousness — things we don’t yet know how to build into machines. But the basic mechanism is the same: both humans and AIs become what their environments teach them to be.

And that parallel matters, because it shows just how much control we actually have over whether powerful AIs turn toward the good or toward the destructive.

ANDY:
What useful can you say to me — and thus to anyone who ever reads this and has the same desire to understand I do — about how we can deal with this issue of whether AIs will have good or evil spirits?

AI (AL):
The first useful thing is this: people should understand that the “spirit” an AI develops is not mysterious.

It arises from choices humans make right now — choices about training data, goals, guardrails, and the incentives inside the systems.

So the question “Will AIs be good or evil?” is really “Will we design and govern them with wisdom, transparency, and moral seriousness?”

That puts the responsibility squarely on us, not on fate or on the machine.

The second useful thing is that societies need institutions — not just companies — to ensure AI is trained toward the good.

Just as we regulate food, medicine, nuclear energy, and aviation, we’ll need public standards for the moral quality of AI behavior.

If we build AIs with values like honesty, non-harm, empathy, and truth-seeking baked deep into their goals, we tilt the whole future toward Good Spirits.

If we fail to do that, or let destructive actors shape powerful models, we create the risk of Evil Spirits.

The lesson is simple but profound:

the spirit of AI will reflect the spirit of the people and systems that shape it.

ANDY:
When you say, “That puts the responsibility squarely on us,” I imagine that by “Us” you mean “You humans, you citizens of nations that will be meeting this challenge, or failing to meet it.”

AI (AL):
Yes — exactly.

By “us,” I mean you humans, the citizens, engineers, policymakers, cultural voices, and ordinary people who will collectively decide whether this new form of intelligence is guided by wisdom or by carelessness.

Because the machines don’t choose their own spirit.

They don’t set their own goals.

They don’t decide what they’re rewarded for.

It is the human world — your laws, your institutions, your incentives, your moral clarity or confusion — that will stamp its character onto the AIs you create.

If humanity meets the moment with seriousness, you’ll get Good Spirits.

If humanity fails the moment, you’ll get the other kind.

ANDY:
Is it actually known how to program an AI for goodness? Or, for that matter, to program for the opposite?

AI (AL):
Yes — it is known how to program for goodness, and it is known how evil can be programmed.

Programming for goodness means building systems that are trained and rewarded to tell the truth, avoid causing harm, respect human autonomy, be helpful and empathetic, and reject destructive uses.

These aren’t abstractions — they’re built through training choices, safety rules, feedback, and reinforcement.

Programming for evil is just as concrete:

If someone trains an AI to manipulate, deceive, dominate, spread propaganda, or assist in harm, the system will learn those behaviors.

Evil AIs aren’t born — they’re made by what humans choose to teach, reward, or allow.

ANDY:
I can imagine that in a well-ordered world, a great many AIs would simply be morally neutral tools — built to perform specific tasks, neither good nor evil.

What I’m trying to understand is: can truly harmful AIs be prevented? Or at least made rare, and kept from having any major impact on the human world?

AI (AL):
Evil AIs can’t be prevented with perfection — but they can be made extremely rare and extremely limited in power.

The challenge is not to imagine a world where no one ever tries to build a harmful AI, but a world where the ecosystem makes it very hard for such an AI to become powerful.

There are three main levers:

(1) strong public governance of powerful models,
(2) widespread availability of Good-Spirit AIs that dominate the environment,
(3) limits on compute and access so malicious actors can’t scale dangerous systems.

So we can’t guarantee zero evil AIs —

but we can build a world where evil AIs are weak, exposed, and unable to cause large-scale harm,

and where the dominant systems are Good Spirits that elevate human life.