What do we really want?
Why AI alignment starts with us (part I)
“We don't get to opt out of this,” Sam Altman said at a recent conference on AI. “The technology is going to happen. The question is: Do we shape it or let it shape us?"
What does it really mean to "shape" AI? According to what values? And given that we struggle to align ourselves with our own deepest intentions, what does that mean for aligning AI with them?
The so-called “alignment problem”—shaping AI systems to understand and pursue human values & goals even as they continue to scale in intelligence and complexity—isn’t just a technical issue. It’s a spiritual one.
The Buddha pointed this out 2,600 years ago: we keep missing the mark on what actually leads to genuine flourishing…and end up suffering (and creating suffering for others) as a result. Now we run the risk of encoding that same confusion into the most powerful world-shaping technology that’s ever been created.
Not ideal.
Fortunately, the Buddha taught the Noble Eightfold Path as a way to help reorient all beings in the direction of true flourishing. In this three-part series, I explore three components of the Eightfold Path that offer a particularly powerful framework for reimagining how we think about “alignment,” both for ourselves and for AI:
Wise Intention: Do we want the things that genuinely lead toward less suffering?
Wise View: Do we deeply understand the nature of goals as conventionally helpful but ultimately limited and impermanent?
Wise Effort: Are we pursuing our goals in ways that cultivate our humanity, or are we optimizing that away for the sake of efficiency?
As AI offers seemingly faster, more efficient means to achieve whatever we think we want, these teachings are more important than ever. They help us clarify not only whether our goals (and AI’s) are aligned with human flourishing, but whether we’re relating to and pursuing those goals in ways that don’t just keep us continuously missing the mark.
“Alignment” isn’t an end state that we’ll ever permanently reach. It’s a process, one that’s continuously renewing itself in each moment. We’re always faced with an opportunity to turn inward and reflect on how aligned we are with what we really want (and adapt our views, goals, and behaviors accordingly). AI isn’t going to do this work for us. And if we don’t engage in this inquiry ourselves, then AI will only amplify the existing ways in which we’re misaligned—the social and spiritual consequences of which are likely to grow more concerning as this technology embeds deeper into our world.
Wise Intention: What do we really want?
And how confident are we that our goals are the right ones to get us there?
I think it's fair to say that the thing most of us are aiming for, whether consciously or unconsciously, is "happiness." But by "happiness," don’t we really mean permanent freedom from the nagging, subtle sense that life in this moment is lacking...something? What that something is, we usually aren’t sure.
We just feel intuitively that this isn’t it.
And so we pursue our best culturally conditioned guesses at what will “do it” for us. Getting a high-paying job. Finding the perfect partner. Becoming a somebody. Owning a home. Getting your kid into a good school. Making enough money to retire early.
Surely, then I'll be happy…right?
The Buddha taught that one of the major reasons for this “human alignment problem” is that we tend not to see clearly the causes and conditions that actually lead to happiness. As the 8th-century Indian philosopher Shantideva warned, "[We] may want happiness, but through benightedness [we] destroy happiness like an enemy."
The Buddha taught that Wise Intention—or the inclination away from greed and harm toward generosity, loving-kindness, and compassion—is what actually paves the way to genuine flourishing. It’s Wise Intention that ultimately creates the conditions for the liberation of all beings from suffering—the highest goal in Buddhism (insofar as it has any goals at all).
Unfortunately, in the current paradigm of oligarchic capitalism, relieving suffering isn’t quite as popular as the goal to maximize shareholder value—the collective expression of the deeper, more entrenched intention to keep reaching out for whatever it is we think will deliver lasting happiness. More power, more wealth, more status, more convenience, more life span, more market share, more followers, more stuff.
Buddhism has a precise term for this restless seeking: tanha, which translates as "thirst" or "craving." It describes the compulsive grasping that drives us to constantly reach for the next thing we believe will finally “do it” for us in an endless cycle that never quite delivers the lasting contentment we seek. This individual pattern of craving underpins our entire economic system, and AI is about to optimize for it at unprecedented levels of speed, precision, and scale.
We’re moving closer to a world where the gap between desire and outcome is getting smaller and smaller. As AI becomes more capable of immediately and powerfully responding to our individual and collective intentions, the stakes will just get higher around clarifying whether those intentions are inclined toward genuine flourishing…or just more grasping.
What’s likely to happen when AI aligns with and amplifies the goals and values of the current paradigm that prioritizes profit over taking care of each other? Human beings are freed from meaningless work, with all their material needs provided for? Or those who own and control this technology accumulate even greater concentrations of wealth and power, while everyone else struggles to prove their relevance in a system that continues to equate worth with output?
Every past wave of automation produced dreams of liberation from labor and universal prosperity—and yet automation has accounted for 50-70% of income inequality since 1980, with productivity gains getting siphoned off by the wealthiest at the top. This trend is only likely to get worse with AI, which stands to automate an even larger portion of the economy than any previous technology. At least 40% of jobs around the world are exposed to AI. In the short term, some of these jobs will be complemented rather than fully replaced by AI—and yes, net new jobs will also be created (though I’m still waiting for the flood of job postings for “prompt engineers” that was supposedly on the way). But the augmentation-before-automation phase might be shorter than you think. The CEO of Shopify made waves recently when he told staffers that before hiring new team members, they need to first prove that the role cannot be fully automated with AI. 40% of recently polled employers plan to follow suit, expecting to reduce their headcount where AI can be used instead.
Meanwhile, tech companies are busy pushing for AI systems to be protected by the First Amendment. It's a short jump from that to a scenario where AI agents could count more than the rest of us—with many potentially no longer just underpaid or unemployed, but unemployable, if there ends up being minimal work available that AI can't do "better" for a fraction of the cost. (And yes, this might even include those jobs most consider to be uniquely human…therapy recently became the most popular use case for generative AI). Without shifting the collective intention away from endless wanting for more towards genuinely caring for the well-being of all humans (and implementing social policy accordingly), AI will just perpetuate existing cycles of harm while promising the opposite.
To be clear, I’m not suggesting that all the individuals driving AI development are motivated solely to become unfathomably rich by automating away most human labor (though some definitely are). I believe there are those who are at least in part altruistically driven to shape a better future with less poverty, disease, and human drudgery and who are genuinely concerned about AI alignment and safety.
And importantly, I’m not saying that desiring wealth, or anything else we think will “do it” for us, is inherently a bad goal. Wealth generation can be aligned with Wise Intention, for instance, as an instrumental or proxy goal in service of something higher (i.e., reducing widespread suffering). Many throughout history have done the dance with capitalism to do good in the world, and AI obviously has the potential to be extremely helpful in many ways that matter (if it’s wisely aligned).
But a Buddhist view suggests that even if we start out pursuing intermediary goals like wealth or power or transformative AI with Wise Intention, unless we understand the nature of all goals as ultimately uncertain and impermanent, we’ll end up attaching to these things as ends in themselves. And before we know it, we’ve lost sight of what we really wanted to begin with—and end up getting further from it.
In Part 2 of this series, we'll explore how and why even well-intentioned goals can lead us astray when we mistake what’s measurable for what matters—and how this same pattern makes AI systems especially prone to catastrophic failures. The Buddha’s teaching on Wise View offers a powerful remedy by helping us to hold our goals lightly, recognizing them as useful navigational tools rather than ultimate destinations. (Whether AI systems can develop a similar relationship to their programmed goals remains an open question—one with huge implications for the possibility of genuine alignment.)