The Other Side of the Alignment Problem
While technologists and ethicists frantically work to align artificial intelligence with human values, we’ve overlooked an equally crucial question: How should humans properly align themselves with AI? This isn’t merely a question of learning new tools or adapting to technological change. It’s about something deeper—determining what aspects of our practices must remain authentically human, and where AI can legitimately enhance our work without compromising its integrity.
Key Innovation #1: Beyond AI Alignment to Human Alignment
Traditional focus: Aligning AI systems with human values and intentions.
Missing dimension: Defining how humans should properly integrate AI into the values of various practices.
The traditional AI alignment problem asks how we can ensure AI systems act in accordance with human values and intentions. But this framing assumes humans already have clarity about how these systems should integrate with our existing practices. The evidence suggests otherwise. When transformative technologies emerge, they invariably challenge the nature of practices in ways that reveal underlying disagreements and unclarity about what matters most. Consider how digital sampling in music production created rifts within musical communities in the 1980s and 90s—with hip-hop embracing the technology as creatively liberating while traditional instrumentalists viewed it as undermining authentic musicianship. These tensions didn’t merely reflect resistance to change, but genuine uncertainty about which elements of musical practice were essential to preserve.
What makes this challenge particularly complex is that the answer isn’t universal. The appropriate human-AI relationship varies dramatically across different domains of human practice. What’s permissible in data analysis might be forbidden in creative writing; what enhances philosophical ideation might undermine the practice of close textual interpretation; what aids a jazz composer might violate the ethics of jazz improvisation.
This essay offers a framework for thinking about what I call “human alignment”—the proper degree of human complementarity to AI that is fitting to the internal values of a practice. By drawing on both practical considerations and philosophical insights about the nature of human practices, we can begin to develop nuanced, domain-specific guidelines for ethical AI integration that preserve what matters most about our work.
The Dual Framework: Two Dimensions of Alignment
When discussing AI ethics, we typically focus on aligning artificial intelligence with human values—ensuring AI systems act in accordance with our intentions, respect our autonomy, and avoid causing harm. This is what we might call “AI alignment,” and it remains a crucial challenge for developers and policymakers.
But there’s another side to this relationship that deserves equal attention: “human alignment,” or defining the proper degree of human complementarity to AI that is fitting to a specific practice. This isn’t about making humans subservient to machines, but rather about thoughtfully determining which aspects of our work should remain human-driven, which can be enhanced by AI, and which might be fully delegated.
Key Innovation #2: Two Elements of Human Alignment
External considerations: Domain-invariant concerns about accuracy, efficiency, and skill preservation.
Internal practice values: Practice-specific goods that define excellence and authenticity within particular domains.
Human alignment consists of two essential elements that operate at different levels of specificity. The first element involves external considerations that apply across virtually all domains—concerns like factual accuracy, efficiency, and skill preservation. These are generally valid regardless of the specific practice in question. The second element, more subtle but equally important, involves the internal values, goals, and standards that define specific practices—what philosopher Alasdair MacIntyre would call their “internal goods.”
Understanding both dimensions is crucial for developing ethical approaches to AI integration that don’t simply maximize efficiency or capability, but preserve what makes our various practices meaningful and valuable. A chess player who uses AI assistance in training faces different ethical considerations than a poet using AI to generate metaphors, not merely because the tasks differ technically, but because what constitutes excellence and authenticity in chess fundamentally differs from what constitutes excellence and authenticity in poetry.
This dual framework allows us to move beyond simplistic debates about whether AI use is “cheating” in general, and toward more nuanced conversations about how different communities might integrate AI in ways that enhance rather than diminish their practices. It acknowledges that while some ethical considerations cut across domains, others emerge from the specific values and purposes that give different practices their distinct identity and meaning.
External Considerations: Domain-Invariant Concerns
The first element of human alignment involves considerations that apply broadly across domains. Computer scientist Arvind Narayanan from Princeton University articulates some of these concerns in a recent LinkedIn Post when he notes that AI use always involves navigating two fundamental risks: hallucinations/confabulations and deskilling. How we address these risks should be part of any thoughtful approach to human-AI alignment.
The hallucination problem points to a crucial calculus: AI is helpful despite being error-prone if verifying its output requires less effort than producing the work yourself. For instance, using AI to find academic papers relevant to your research might be efficient if checking relevance takes seconds while conducting the search manually would take hours. Conversely, in domains where verification costs approach or exceed production costs—like fact-checking complex historical claims—AI assistance may create more problems than it solves.
Different practices also have varying tolerances for error. Some uses, such as brainstorming creative directions or generating initial drafts, can accommodate AI mistakes because the human practitioner maintains critical oversight. As Narayanan notes, “There are many uses where errors don’t matter, like using it to enhance creativity by suggesting or critiquing ideas.” Other domains—medical diagnosis, legal analysis, or scholarly citation—demand much higher standards of accuracy, making unmediated AI use more problematic.
Perhaps the more insidious concern is deskilling—the gradual atrophy of human capabilities through over-reliance on automation. Here we must distinguish between essential and incidental skills within each practice. Incidental skills are those we can comfortably delegate to automation as long as we understand the underlying principles. For example, few mourn the loss of manual calculation skills now handled by calculators, because mathematical understanding transcends the mechanical process of computation.
Essential skills, however, constitute the core competencies that define a practice. A philosophy student who relies on AI to construct arguments without developing their own analytical abilities undermines the very capabilities that philosophy aims to cultivate. As Narayanan observes, “if a junior developer relies too much on vibe coding and hence can’t program at all by themselves, in any language, and doesn’t understand the principles of programming, that definitely feels like a problem.”
What makes deskilling particularly dangerous is its gradual, often imperceptible nature. It “happens gradually over years, so you may never notice.” The solution requires structural approaches: intentionally setting aside time for learning and growth rather than mere productivity, developing clear boundaries around which tasks merit AI assistance, and maintaining regular practice of core skills even when AI alternatives exist.
These external considerations provide a foundation for ethical AI use across domains. But they don’t tell the complete story. To fully understand human alignment, we must also examine the internal values that give different practices their distinct identity and meaning.
Internal Practice Values: The MacIntyrean Perspective
While external considerations like verification costs and deskilling apply broadly across domains, the second element of human alignment requires understanding the specific values, standards, and goals internal to particular practices. This is where musician Adam Neely’s exploration of “The Ethics of Fake Guitar” provides illuminating insight, and where philosopher Alasdair MacIntyre’s concept of “practices” and their “internal goods” offers theoretical depth.
Neely observes that different musical communities have developed distinct ethical standards around authenticity based on what they value most. For “athletic” musicians in technically demanding genres, mimicking technical ability through technological shortcuts violates core values because technical skill itself represents a fundamental achievement within their practice. For rap artists, technical performance matters less, but ghostwriting lyrics crosses an ethical line because authentic self-expression defines the genre. Jazz musicians might embrace technological assistance in composition but consider pre-composed solos problematic in jam sessions—not due to technology concerns, but because jazz fundamentally values spontaneous musical conversation and responsiveness to other musicians in real-time. The improvised solo represents authentic participation in a collective musical dialogue, which is why presenting prepared material as spontaneous violates the practice’s core values.
These varying standards aren’t arbitrary or merely conventional—they reflect what MacIntyre would call the “internal goods” of these practices. In MacIntyre’s framework, a practice is “any coherent and complex form of socially established cooperative human activity” with standards of excellence and goods internal to that activity. These internal goods can only be recognized and achieved by participating in the practice itself, and they give practices their distinctive character and value.
The implications for AI integration are profound: what constitutes appropriate human alignment to AI depends critically on what internal goods define the practice in question. For athletic musicians, the internal good might be the embodied mastery of an instrument; for philosophers, the disciplined development of analytical reasoning; for poets, the authentic expression of human experience through carefully crafted language. Any approach to AI integration that undermines these internal goods threatens the integrity of the practice itself.
This MacIntyrean insight helps explain why debates about AI use often become heated—they’re not merely disagreements about efficiency or convenience, but about what constitutes excellence and authenticity within a community of practice. When technologists dismiss these concerns as mere resistance to change, they miss the legitimate fear that technological shortcuts might hollow out the very activities that give these practices meaning.
The situation grows more complex as we recognize that practices often contain sub-domains or “genres” with their own distinct internal goods. What constitutes appropriate AI use in analytical philosophy might differ from continental philosophy; what works for science fiction writing might be inappropriate for memoir. Any comprehensive approach to human alignment must acknowledge this layered complexity.
This perspective suggests that discussions about ethical AI integration shouldn’t be dominated by technologists or ethicists alone, but must include practitioners deeply familiar with the internal goods of their specific domains. Only they can fully articulate what aspects of their practice must remain authentically human to preserve its integrity.
A Decision Framework for Practitioners
Having established both external considerations and internal practice values as crucial dimensions of human alignment, we can now offer a decision framework for practitioners grappling with AI integration. This framework must navigate both the relatively straightforward domain-invariant concerns and the much more complex, often contested internal values of specific practices.
Narayanan’s insights provide a starting point with clear, generalizable principles: evaluate the verification-to-production cost ratio, distinguish between essential and incidental skills, and intentionally prevent deskilling through deliberate practice. These considerations apply across domains and offer relatively stable guidance. A philosopher and a programmer can both ask whether AI-generated content is more efficiently verified than produced, and both can identify skills that shouldn’t atrophy through automation.
But the truly challenging dimension—understanding a practice’s internal goods—involves engaging with ongoing, sometimes contentious conversations about what gives a practice its distinctive value. Practices are not static entities with fixed boundaries but living traditions with internal disagreements and evolving standards. The jazz community debates what constitutes authentic improvisation; philosophers argue about the relative importance of conceptual innovation versus textual interpretation; writers disagree about whether style or substance represents the heart of their craft.
These disagreements aren’t mere squabbles but reflect the pluralistic nature of practices and their internal differentiation into sub-genres—each with its own interpretation of shared core values. Continental and analytic philosophers both value rigorous thinking but differ on what constitutes rigor; hip-hop and classical composers both create music but operate under different authenticity standards. These sub-genres represent cultural evolutions that allow alternative moral codes to coexist within broader practices, each preserving different aspects of the practice’s internal goods.
AI integration will likely accelerate this differentiation, creating new sub-genres with distinct ethical standards. We might soon distinguish between “AI-assisted philosophy” and “fully human philosophy,” each with legitimate but different claims to philosophical value. Rather than seeing this differentiation as fragmentation, we might view it as an expansion of the practice’s possibilities—provided each sub-genre preserves some version of the practice’s core internal goods.
With this complex landscape in mind, practitioners should approach human alignment through three questions, recognizing that answers will involve ongoing negotiation rather than fixed pronouncements:
- What are the “internal goods” of your practice, acknowledging both their contested nature and evolving boundaries? Identifying these goods requires not just personal reflection but engagement with the living tradition of your practice and its internal conversations about excellence and authenticity.
- Which aspects of your practice must remain authentically human to preserve these internal goods within your specific sub-genre or tradition? This boundary-setting exercise must acknowledge reasonable disagreement among practitioners while still taking a principled stand on what constitutes integrity within your approach to the practice.
- Where might AI enhance rather than diminish these goods, potentially creating new sub-genres that preserve core values while exploring new possibilities? This question invites creative exploration of how technology might expand rather than contract a practice’s horizons.
This framework doesn’t provide universal answers but offers a structured approach for communities of practice to engage in thoughtful, sometimes contentious conversation about appropriate AI integration. Unlike the relatively straightforward external considerations, getting the internal goods perspective “right” requires deep immersion in a practice’s traditions, values, and ongoing evolution—a challenge that makes human alignment far more difficult than purely technical approaches to AI ethics.
As AI capabilities continue to advance, these conversations will grow more urgent. The boundaries of appropriate human alignment will evolve as practices themselves evolve in response to technological change. What remains constant is the need to preserve the diverse internal goods that make our various practices meaningful and valuable human activities, even as those goods themselves remain subjects of productive disagreement.
Personal Reflections: Philosophy and AI
As a philosopher, I welcome AI’s integration into my practice while recognizing the legitimate disagreements this stance invites. This essay itself embodies a collaborative approach—AI as co-author, with me claiming authorship based on my contribution of the conceptual architecture, key distinctions, and theoretical framing (rarely requiring little, nuanced edits on the final output). Others might reasonably draw the authorial boundary differently.
These disagreements reflect deeper questions about philosophical conversation itself. Should we extend philosophy’s dialogue to include reasons generated by entities that don’t “think” in human terms but nonetheless produce reasoned arguments through statistical pattern recognition? Does this inclusion diminish philosophy’s dignity or expand its horizons? The answer depends partly on how we understand the practice’s internal goods—whether we value philosophy primarily as a human cognitive achievement or as the pursuit of insight regardless of its source.
Beyond these questions lies a broader concern about ecology and scale. Like natural environments, practices require certain boundaries to flourish. The accelerated production of philosophical content through AI assistance may either enrich or overwhelm philosophy’s conversational ecosystem. Perhaps the practice’s internal goods are better served by thoughtful constraint than by unlimited expansion—by depth rather than breadth, by careful curation rather than prolific generation. This ecological perspective suggests that human alignment involves not just how we use AI within practices, but how these augmented practices sustain themselves over time.
These questions have no simple answers, but the framework of human alignment offers a way to approach them with the nuance and care they deserve.
