In 2016, South Korean player Lee Sedol — considered one of the greatest Go players in history — lost four games of five to AlphaGo. After his single victory in Game 4, he described it as the most meaningful win of his career. The machine had already rendered his competitive dominance obsolete. Yet his joy was unmistakably human: it came from struggle, not outcome.
That asymmetry — a human finding meaning in the very difficulty an AI eliminates — cuts to the heart of what philosophers call the human flourishing question.
Aristotle's concept of eudaimonia — translated variously as happiness, flourishing, or living well — was never simply about comfort or pleasure. It referred to the full actualization of distinctively human capacities: reason, social life, virtue, and purposeful activity. On this view, a life of effortless ease provided by an omnipotent servant would not be a flourishing life — it would be an impoverished one.
Contemporary psychologist Martin Seligman, whose PERMA framework (Positive emotion, Engagement, Relationships, Meaning, Accomplishment) emerged from decades of clinical research, similarly emphasizes that accomplishment matters independently of reward. The doing is constitutive of wellbeing, not merely instrumental to it.
Both traditions raise a pointed question: if AI systems perform the tasks through which humans previously exercised skill, built relationships, and found meaning — does something essential get lost?
MD Anderson Cancer Center spent $62 million developing an IBM Watson-powered oncology advisor. After three years, the project was suspended. Watson had been trained on hypothetical cases rather than real clinical records and recommended treatments oncologists considered unsafe. The episode illustrated that cognitive offloading in high-stakes domains carries costs well beyond technical accuracy — it also erodes the cultivated judgment that physicians develop through years of difficult cases.
The Substitution Worry holds that AI replacing human cognitive activity degrades the capacities and satisfactions tied to exercising those abilities. Chess players who rely heavily on engine analysis report that their intrinsic motivation to calculate diminishes. Radiologists trained in AI-augmented environments show reduced unaided diagnostic accuracy over time — a phenomenon documented in studies from Stanford Medicine between 2019 and 2022.
The Augmentation Optimism position argues that AI can free humans from drudgery and expand the range of genuinely human activities. Surgeons using the da Vinci robotic system perform procedures impossible with the unaided hand. Scientists using AlphaFold2 (released 2021) resolved protein structures that would have taken careers — freeing researchers to ask higher-order biological questions. This view holds that what matters is not cognitive effort per se, but purposeful engagement with genuinely human goals.
The Dependency Risk perspective focuses not on individuals but on civilizational resilience. GPS navigation has measurably reduced the hippocampal activity associated with spatial memory. As documented in a 2020 UCL study, habitual GPS users show less engagement of the hippocampus and caudate nucleus during navigation. The concern is systemic: a civilization that cannot function without its tools is fragile in ways a flourishing civilization should not be.
The question is not whether AI is used, but how the relationship is structured. A musician using software to master recordings retains the compositional and performative challenges that generate meaning. A musician who uses AI to generate melodies, harmonies, and arrangements has a fundamentally different relationship to the creative act — and to the satisfaction it yields.
Lee Sedol retired from professional Go in 2019, citing AlphaGo as one reason — he described the experience of knowing no human could ever match the machine as removing something essential from the game. That same year, a new Go variant called Fog of War Go emerged, deliberately designed to resist AI domination. Humans were engineering a space where their particular kind of intelligence could still matter. Whether that is adaptation or retreat is precisely the question this module pursues.
Think about a skill or practice that matters to you — something you spend real time developing. It could be writing, cooking, a sport, music, coding, clinical judgment, or anything else. Your AI conversation partner will help you examine where AI assistance would enhance your flourishing versus where it might undermine the very activities that generate meaning for you.
Complete at least 3 exchanges. Push back on the AI's responses to deepen the analysis.
In 2023, the Hollywood writers' strike — the longest in decades — centered not on wages but on a specific demand: AI could not be used to generate or rewrite scripts. The Writers Guild of America understood that allowing AI into the creative pipeline would not merely reduce income. It would sever the connection between the writer's intelligence and the work that bore their name. They were striking, in part, for the right to be the author of their own labor.
Psychological research consistently finds that work satisfaction correlates more strongly with autonomy, mastery, and purpose than with compensation. Self-Determination Theory, developed by Edward Deci and Richard Ryan at the University of Rochester through decades of empirical work, identifies three fundamental psychological needs: autonomy (acting from genuine choice), competence (exercising skill effectively), and relatedness (meaningful connection to others). Work, when structured well, satisfies all three.
When AI performs the cognitive core of a job — the diagnostic reasoning, the creative generation, the analytical judgment — it potentially satisfies the output requirement while hollowing out the psychological experience. The surgeon who executes a procedure planned entirely by an AI system still earns their salary. But whether they are exercising their competence or executing instructions is a genuinely different question.
A series of studies from Stanford Medicine examined radiologists trained in AI-augmented environments. Residents who relied heavily on AI flagging for preliminary reads showed measurably reduced accuracy when the AI was removed — a phenomenon the researchers described as a "skill formation gap." The AI had not replaced radiology; it had interrupted the accumulation of diagnostic pattern recognition that constitutes radiological expertise. Junior radiologists were learning to confirm AI outputs rather than build independent perceptual skill.
Work is not only a source of income or activity; it is a primary domain through which modern people construct identity. Sociologist Richard Sennett, in The Craftsman (2008), argued that skilled work is one of the last sites where humans encounter genuine resistance — material or intellectual — that forces self-development. The craftsman's identity is inseparable from the difficulty of the craft.
This creates a specific problem for AI-mediated work: the work may persist in name while the difficulty that gave it identity-forming power disappears. A writer who uses AI to generate first drafts is still a writer in some sense. But the specific experience of confronting a blank page, of wrestling language into form, of discovering what you think by writing it down — that experience, which many writers identify as the core of their practice — is gone.
The 2023 WGA strike resulted in a negotiated agreement limiting AI-generated material: AI could not be used to write or rewrite literary material, and any AI-assisted material had to be disclosed. The agreement acknowledged that the identity of a writer was bound to the act of writing — not the text that resulted.
The optimistic reading notes that every wave of automation has historically created new categories of human work. Agricultural mechanization freed labor that built industrial economies. Word processors did not end professional writing — they made it more accessible and prolific. The question is whether AI-era automation differs in kind: previous automation replaced physical and routine cognitive work, while AI increasingly targets non-routine cognitive work — the domain previously considered uniquely human.
High-meaning, high-skill work (surgery, composing, scientific research) — AI augmentation can expand what is possible without necessarily removing the human challenge. But the balance is fragile and depends heavily on how AI is integrated.
Medium-skill, high-volume work (legal document review, routine journalism, customer service) — AI substitution is already substantial and accelerating. The meaningful work in these domains — the judgment calls, the craft — is surrounded by volume that AI eliminates, but the margin of human contribution narrows.
Low-skill, high-frequency work (data entry, basic image classification, form processing) — Full AI substitution is largely complete. The human flourishing question here is distributive: what do people whose labor value was in these categories do instead?
The WGA agreement's disclosure requirement reflects a broader principle: when humans are credited with work, something about the relationship between person and product matters independently of the output's quality. A beautiful building designed entirely by an AI system attributed to a human architect involves a kind of deception — not because the building is less beautiful, but because the relationship between architect and work that gives the credit meaning has been severed.
Choose a professional domain — your own field or one you know well. Work with your AI conversation partner to map exactly where in that domain AI assistance augments human meaning versus where it begins to displace the activities that make the work identity-forming. Apply Self-Determination Theory (autonomy, competence, relatedness) as your analytical lens.
Complete at least 3 exchanges. Push the analysis toward specific activities, not generalities.
In 2021, Frances Haugen — a former Facebook data scientist — testified before the U.S. Senate and released tens of thousands of internal documents. Among them: research showing Facebook's own teams had identified that Instagram use was associated with body image issues in teenage girls, and that the platform's recommendation algorithm was documented to push users toward progressively more extreme content. The company had chosen engagement over wellbeing. The documents showed this was not ignorance — it was a deliberate design choice.
The philosopher Simone Weil wrote that attention is the rarest and purest form of generosity. Cognitive scientists have extended this: sustained, directed attention is not merely an instrument of productivity — it is the mechanism through which humans form values, exercise judgment, and cultivate relationships. You cannot read deeply, deliberate carefully, or be genuinely present with another person without it.
AI systems designed to maximize engagement are, by definition, designed to capture and hold attention — to interrupt, redirect, and occupy cognitive resources. The attention economy researcher Tim Wu documented in The Attention Merchants (2016) how advertising-supported media has always monetized human attention. What AI adds is the capacity to optimize this capture with unprecedented precision, personalizing the hook for each individual nervous system.
In 2019, Guillaume Chaslot — a former YouTube engineer who had worked on the recommendation algorithm — published analysis showing the algorithm systematically recommended progressively more extreme content because extreme content generated longer watch times. This was not a bug; it was the algorithm performing exactly as designed. The optimization target (engagement) produced systematic radicalization as a side effect. YouTube subsequently modified its algorithm in 2019 and 2022, reducing recommendations of borderline content — but the core economic incentive remained.
Personalization feels like a form of freedom: the platform shows you what you want. But this conflates preference-satisfaction with autonomy. Genuine autonomy, in the philosophical tradition running from Kant through contemporary philosophers like Christine Korsgaard, involves acting from reasons you endorse on reflection — not simply having your existing impulses fulfilled.
A recommendation algorithm that learns your patterns and feeds them back to you at higher intensity is not serving your autonomous preferences — it is amplifying your impulsive patterns while bypassing your reflective capacity. This is precisely the distinction between wanting and valuing: you can want something in the moment that you do not, on reflection, value. Systems optimized for the former systematically undermine the latter.
The philosopher Harry Frankfurt's distinction between first-order desires (what you want right now) and second-order desires (what you want to want) is illuminating here. Flourishing, on most philosophical accounts, requires that your first-order desires align with your second-order values. Engagement-maximizing AI consistently targets first-order desires, often at the expense of second-order alignment.
Internal Facebook research released through the Haugen documents found that 32% of teenage girls said that when they felt bad about their bodies, Instagram made them feel worse. The research also found that "teens blame Instagram for increases in the rate of anxiety and depression." Critically, the research team noted the platform was aware and had not acted on earlier findings. This represents a documented case of AI-mediated design choices systematically undermining human psychological wellbeing at scale.
The question is not whether AI can be used in attention-capture systems — it can and will be. The question is whether those systems can be re-oriented toward what researchers call time well spent rather than time maximized. The Center for Humane Technology, founded by former Google design ethicist Tristan Harris, has documented specific design alternatives: recommending content that users rate as meaningful after the fact (not just engaging in the moment), building in friction that creates deliberate pauses, and designing defaults that favor depth over volume.
In 2018, Apple introduced Screen Time and in 2019 added usage controls to iOS specifically in response to advocacy from investor groups arguing that iPhone usage patterns were harming children. Google followed with Digital Wellbeing tools. These were partial responses — the economic incentives remained intact — but they acknowledged a principle: platform design has consequences for human flourishing that the platforms themselves bear some responsibility for.
The philosopher Michael Sandel has argued that some goods are corrupted by being optimized. Friendship is not served by an algorithm that predicts who you will enjoy talking to most — the serendipity and effort and risk of genuine friendship are constitutive of its value. Attention may be similar: the capacity to choose what you attend to, and to sustain that attention against resistance, may be inseparable from the kind of selfhood that makes flourishing possible.
You are a product designer at a major social platform. Your challenge: propose a specific feature — or a modification to an existing feature — that reorients the platform's optimization target from engagement-maximization toward something users would endorse on reflection. Your AI partner will stress-test your proposal, push for specificity, and raise the economic and technical objections you'll need to answer.
Complete at least 3 exchanges. Be concrete — name the specific behavioral target, the design mechanism, and how you'd measure success.
In 2017, the city of Amsterdam launched an initiative to use AI to assist residents navigating social services — not to replace caseworkers, but to ensure no one fell through the gaps. The system identified residents at risk of housing instability, debt spirals, or care gaps and prompted human caseworkers to reach out. By design, the AI never made decisions — it surfaced information to humans who did. Amsterdam's Responsible AI team described this explicitly as a flourishing-oriented design choice: AI amplifies human attention and care; humans exercise judgment and relationship.
The previous three lessons have established the problem space: AI can erode skill, hollow out work identity, and capture attention in ways that undermine the autonomy and engagement that flourishing requires. This lesson asks the constructive question: given what we know about flourishing, what design principles should govern AI systems?
Philosopher and AI ethicist Shannon Vallor at the University of Edinburgh has proposed that AI systems should be evaluated not only on efficiency metrics but on techno-moral virtues — whether they cultivate or erode the capacities for self-reflection, care, collaboration, and practical wisdom that human flourishing requires. This is not merely a regulatory aspiration; it is a design criterion.
When Khan Academy launched its AI tutoring assistant Khanmigo in 2023, a deliberate design choice was embedded: Khanmigo does not give students answers. Instead, it asks questions, surfaces hints, and guides students through their own reasoning process. Sal Khan explicitly described this as a flourishing-oriented choice — the goal of education is not knowledge transfer but capability development, and an AI that answers questions directly undermines that goal. Early research from Khan Academy showed students using Khanmigo showed greater engagement with reasoning processes than those using answer-providing tools.
1. Competence-Preserving Design. AI should be designed to develop human skill, not substitute for it. The Khanmigo model — questioning rather than answering — is the archetype. In medical AI, this translates to tools that show their reasoning and require clinicians to evaluate it, rather than tools that output conclusions for clinicians to sign off on. The goal is for human expertise to grow through engagement with AI, not atrophy through delegation to it.
2. Transparency and Legibility. AI systems should make their reasoning visible to the humans who use them. When a physician understands why a diagnostic AI flagged an anomaly, they can exercise judgment about whether the flag is relevant. When they only receive an output, they cannot. The EU AI Act (2024) mandates explainability requirements for high-risk AI systems specifically for this reason — legibility preserves human agency.
3. Autonomy-Preserving Defaults. Defaults should be oriented toward user reflection rather than impulsive engagement. This means friction by design: confirmation steps before consequential decisions, summaries before scroll-continuation, prompts that surface the user's stated goals and ask whether current behavior serves them. Apple's Screen Time and iOS notification controls represent partial implementations of this principle.
4. Relationship-Centered Rather Than Relationship-Replacing. AI in caregiving, education, and therapy should augment human relationships rather than replace them. The Amsterdam social services model is illustrative: AI identifies; humans connect. Research on AI-based mental health chatbots (Woebot, Wysa) consistently shows they work best as bridges to human therapy, not substitutes for it. The relationship — with its full human complexity — cannot be substituted without loss.
5. Collective and Distributive Considerations. Individual flourishing is embedded in social flourishing. AI systems that concentrate benefits among skilled users while eliminating income pathways for others do not serve human flourishing at the societal level, even if they benefit their direct users. The distributive question — who gains capacity, who loses it — is part of the flourishing calculus.
The EU AI Act, which entered into force in August 2024, contains several provisions that reflect flourishing considerations without using that language. High-risk AI systems must be designed with human oversight requirements. AI systems in education must not exploit the characteristics of users. General-purpose AI systems that interact with users must identify themselves as AI. While framed in rights and safety language, these requirements operationalize the insight that human agency and self-determination are conditions of flourishing that AI systems can systematically undermine — and that regulation can partially protect.
In 2023, a coalition of philosophers, AI researchers, and public health experts released the Helsinki Declaration on AI and Human Dignity, arguing that AI should be evaluated against human flourishing criteria: does the system expand or contract human agency? Does it cultivate or erode human capability? Does it strengthen or weaken the social relationships through which humans find meaning? The declaration called for flourishing impact assessments alongside the safety and rights assessments that regulatory frameworks typically require.
The declaration's central claim: an AI system that poses no safety risk and violates no rights may still be a bad AI system if it systematically degrades the conditions of human flourishing. This represents a conceptual expansion of the criteria by which AI should be evaluated — from "does it cause harm?" to "does it make human life genuinely better?"
The question that opened this module — can a species hand off its cognitive labor and still call the result a good life? — does not have a single answer. It depends on what is handed off, to what end, and under what design constraints. The Amsterdam caseworker model, the Khanmigo tutoring model, and the AlphaFold research model all suggest the answer can be yes — if the AI is designed around what humans need to flourish, not merely around what is technically possible or commercially optimal. That design choice, made millions of times in millions of systems, is one of the defining questions of the coming decades.
Choose an AI system you actually use — a writing assistant, a navigation app, a recommendation system, a tutoring tool, a health tracker. Work through a structured flourishing audit: which of the five design principles does it currently satisfy? Which does it violate? What single concrete change would most improve its flourishing profile? Your AI partner will push you to be specific and will challenge weak reasoning.
Complete at least 3 exchanges. Move from description to diagnosis to specific prescription.