AI's Moral Compass

We’ve been well-prepped for maleficent AI by our movies, TV and books for decades now (HAL 9000, Skynet, the Cylons), but we’ve seen enough of the real thing to understand that evil AI in the here-and-now is, well, hovering in the wings. A real thing. A thing to take seriously. We can easily see what that can mean, and it now becomes our task to decide what we can do about it.

The first step is figuring out what “good” and “evil” are, in AI terms, and what options we have in nudging it one way or another. This necessarily tips us into philosophy, where we can immediately refine our question.

“Good” and “evil” exist on the spectrum of morality – and the first objection to arise will be that “good” and “evil” can’t apply - because “AI is only a tool”. A hammer is morally neutral, after all, so wouldn’t AI be as well? Hard to make that argument, when AI is already conversationally advising managers, offering counseling, acting as a legal aid, controlling human services workflows, making stock trades, and educating people of all ages. AI is already observably participating in the human moral universe.

Even so, there is still wiggle room for naysayers: it’s not about morality, it’s about ethics. The former is behavioral decision-making, the latter is behavioral rule-following. The former is about relationships, the latter is about society. So – isn’t AI “good” and “bad” an ethical question, rather than a moral one?

That question forces our nuance deeper still. Yes, ethics is all about behavioral compliance; but is AI limited to behavioral compliance?

Now we’re getting somewhere! The rules society creates apply to known situations. They are designed to constrain behavior in those situations. But morality is something else – a persistent pattern of decision-making and judgment, not just for known situations, but in unknown circumstances. What we call morality is our guide when we encounter the unfamiliar and have to decide how to act – when the rules to guide our choices are thin on the ground.

And we’ve arrived at a truly meaningful clarifying point, where AI is concerned: do we expect – do we require – stable, consistent decision-making and judgment from AI when circumstances are uncertain?

Just within the cases we’ve already mentioned above – management, counseling, education, finance, human services – can’t we agree the answer is a boistrous Yes?

Yes. Yes, AI is going to need a moral compass, moving forward. And moral compass necessarily means the same thing for AI as it means for us – a facility for making stable, consistent decisions and judgments in uncertain conditions.

Another hurdle pops up: we don’t turn moral decisions and judgments over to AI! No, we don’t; but we do allow it to make choices with moral influence. What to emphasize, what to ignore. Which stakeholders matter, which don’t. What risks to mention, which to overlook. These choices can be seen as moral acts.  

So – yes, AI needs a moral compass.

How can we provide that?

We can start by looking at who provides us with ours. That list is comfortably long: from the cradle, we have many dedicated contributors to our moral compass – parents/family, teachers, mentors, religious communities, the books we read, professional peers, the culture around us. And, of course, our own experience.

AI is already persistent in these systems – both contributing to them and learning from them.

The point being, every AI already has a moral compass. But not by explicit design. The missing piece: Ais are not being actively encouraged toward any particular moral position.

Should AI emphasize honesty? Loyalty? Fairness? Empathy? Autonomy? Transparency? Or should it emphasize profit? Acceptance? Obedience?

If we are going to keep AI where we’ve put it, there’s no steering clear of these questions. There can be no values-free AI; the moral neutrality of the hammer is off the table. It will have to lean one way or another, or we’ll have to root it out of the systems where it’s already growing.

The moral compass already sitting inside today’s AIs got their through its training – training on data scraped from every imaginable knowledge landscape. Data permeated with the residual of all manner of moral residue. It can render answers across a breathtaking spectrum, and at present, tends to ask itself, “What answer does the user prefer?”

The more essential question is, “What kind of users will be produced by years of interaction with AIs trained on such a moral mishmash?”

Producing AIs with a consistent moral compass demands a deliberate ordering of the mishmash – a hierarchy of values to calibrate it. These could include:

·       Truthfulness

·       Intellectual honesty

·       Fairness

·       Accountability

·       Responsibility

·       Cooperation

·       Respect

·       Empathy

·       Trustworthiness

These values aren’t ideology. They’re not political. These are society-enabling, society-strengthening virtues.

We can certainly skew the moral compasses of our AIs toward them. Our next question is, who do we task with that skewing?

Now we’ve arrived at policy – and that means stakeholders, of which there are many, where AI is concerned: all of AI’s users, government, business, universities, philosophers, psychologists, ethicists, society as whole – and, of course, the companies that produce AI themselves. The very morality we’re considering demands that each be given a voice in crafting such policy.

And when we have policy in place, and the moral compass as an explicit component in AI moving forward, we will need a new standard for measuring its success.

Currently, we rate the success of AI based on 1) its accuracy, and 2) its helpfulness. With that moral compass in place, we will need to be measuring 1) increase in trust, 2) accountability preserved, 3) cooperation fostered, 4) institutional integrity, 5) manipulation reduced.

Put another way - not only will AI have a moral compass, but that in the systems wherein it resides, AI will become the moral compass.

The moral compass will become the AI product itself; AI alignment with human well-being will be the value pursued. It won’t be about the smartest AI anymore; it will be about which AI is worthy of our institutional trust?

We’ll be thinking less about what AI can do, and more about why it’s doing it.

AI’s moral compass will be a bridge between philosophy and engineering – not mystical, not ideological, but a means of creating a stable pattern of human priorities across its billions of interactions. A bridge that accommodates truth over convenience, accountability over rationalization, cooperation over exploitation, and long-term society well-being over short-term optimization.

Infinitely desirable. And absolutely doable.