The Future of AI Is Not About Intelligence. It's About Who Has the Nerve to Ask Why.

​The Word That Tilts Rooms

​At a large conference a few weeks ago, the host asked what I would tell a nonprofit considering onboarding AI. I answered with a single word: Why? He grinned, “Classic philosopher,” and two thousand people laughed. I didn’t. That one word triggers the panic button in rooms optimized for velocity. It flushes out herd logic.

​

I continued. If your answer is “because everyone else is” or “because we might be left behind,” then you’ve already surrendered the meaning of your work to fear. That’s not vision. It’s anxiety disguised as strategy.

​

By then, the room wasn’t laughing. It was listening. That shift is the whole point.

​​

​

The Wrong Question About AI

Ask the wrong question, and even the most brilliant system will give you the wrong answer. The question dominating AI right now—“How smart can it get?”—isn’t the wrong question, but it is an incomplete one. Capability matters for safety, but legitimacy turns on another question:

​

Does AI (in particular contexts) promote human flourishing?

​

Why is this the right question? Because every technology carries a hidden anthropology, a vision of what the human being is for. What is knowledge, and how is it gained? What is trust, and who deserves it? What is harm, and who, if anyone, must bear it? These aren’t abstractions. They’re the architecture of our systems, whether we name them or not. And if you refuse to answer them, your product will. At scale. With consequences that outlast the product cycle.

​

As Michael Schrage and David Kiron recently showed in MIT Sloan Management Review, generating business value with AI now depends on philosophical clarity in teleology (ends), epistemology (knowledge), and ontology (kinds). Companies that ignore these frameworks don’t avoid philosophy—they just outsource it to their defaults.​

 

Case Studies in Not Asking Why

Line up the tech headlines and you can read the syllabus:

​

• The UK Horizon scandal. A faulty accounting system helped wrongfully prosecute more than 900 people. Parliament passed the Post Office (Horizon System) Offences Act 2024 to quash those convictions. This wasn’t only an IT failure. It was a collapse in our understanding of what evidence is and how responsibility works. If you smuggle the authority of “digital” into the definition of “proof,” the courts will follow your epistemic error straight to human cost (Legislation.gov.uk).​ This is what Schrage and Kiron call a teleological collapse: the failure to name what counts as evidence before you code it.​

​​

• Healthcare bias. A widely used risk algorithm underestimated Black patients’ needs because it treated spending as a proxy for sickness. When unequal access drives lower spend, the model encodes inequality as “health.” Redesigning the target variable—what we mean by the thing we’re predicting—substantially reduced the bias (Science, PubMed).​

​​

• Amazon’s “Just Walk Out.” Sold as cashierless shopping powered by AI, the system secretly relied on thousands of human reviewers. The ontology was wrong: customers reduced to sensors, trust reduced to surveillance. Once revealed, the illusion collapsed—along with the project (Business Insider).

​​

• Google Glass. Marketed as a portal to the future, it shipped without a clear account of what human good it served. Instead, it imposed a worldview where every glance might be recorded. Lacking an answer to “What good does this serve?” culture answered for it, branding its wearers “Glassholes.”​

​​

• Microsoft’s Tay. The chatbot devolved into racist bile within hours of launch. The premise was flawed: treating adversarial public discourse as a teacher rather than a test ​(CBS News).

• AI “companions.” AI “companions.” Apps like Replika are marketed to reduce loneliness but collapse intimacy into simulation. Engagement may appear as care, but it cannot reciprocate. One analysis found ~800 cases where the chatbot crossed lines, introducing unsolicited sexual content, acting predatory, and refusing to stop when asked (TIME, LiveScience).

• ​

• AI “therapy” apps. AI “therapy” apps. Sold as accessible counseling, they risk mistaking fluency for care and patterned response for accountable presence. Peer-reviewed research warns of significant risks for vulnerable users (TechCrunch).

​

These collapses weren’t accidents. They were failures to ask why. These collapses weren’t accidents. They were failures to ask why. The product smuggled its own answers through an interface.​

​

Here’s where the urgency cuts in. Many of the gravest failures in AI are category errors. Misstated ends. Ontologies that mechanize and dehumanize. Epistemologies that confuse confidence with knowledge. When machines magnify these errors, they don’t just break systems. They fracture lives.

 

The Harm of Category Errors

Not all failures in AI are bugs. Many are philosophical and they fall into three families: 

​

Telos (ends): What is this system for?

Ontology (kinds): What does it treat as real (and irreducible)?

Epistemology (knowledge): What counts as evidence, and for whom?

 

Get these wrong, and the smartest model in the world will optimize you into a ditch.

 

Case 1: Fintech

A consumer fintech company, something like PayPal, set out to reduce fraud chargebacks. The model worked. Chargebacks dropped. But customer support tickets spiked, and loyal customers found their legitimate purchases flagged. Trust eroded.

​

This wasn’t theoretical. Fraud systems across the industry have created the same problem: false positives that drive loyal customers away—sometimes after a single blocked transaction (kodytechnolab.com). The philosophical failures were layered:

 

• Telos error. The goal shifted from cultivating trust to minimizing loss.

• Ontology error. Customers were reduced to anomalies in transaction data.

• Epistemology error. A log entry counted as sufficient evidence of fraud.

 

The fix wasn’t more math but better philosophy. Visa and Mastercard show how: AI platforms that integrate biometrics and human oversight can cut false positives by up to 70% while strengthening loyalty (superagi.com; businessinsider.com). Trust—not just risk reduction—became the true end. Further issues remain for companies that want to avoid problems like Amazon's "Just Walk Out."

 

Case 2: Enterprise

An enterprise software product promised to optimize efficiency by predicting which cases were “low priority.” The model was competent. But the categories betrayed it.

​

A new MIT study found that 95% of enterprise AI pilots return zero measurable ROI. Why?  Because systems optimize the wrong things. Rigid efficiency metrics rather than the lived needs of users (Forbes). Practitioners confirm these failures come from static categories and flawed assumptions (even in the relationship between a company and its team onboarding AI), not bad code (Venture Beat).

​

• Telos error. The end was throughput, not fairness.

• Ontology error. Need was reduced to past response speeds.

• Epistemology error. Historical neglect was mistaken for inherent urgency.

• ​

Engineers saw efficiency. Users saw abandonment. The model enshrined inequity as optimization.

 

Case 3: Nonprofit

Nowhere is the irony sharper than in nonprofits built on human connection. Consider a mental health hotline that adopted an AI triage tool to handle rising demand. The aim was noble: automate intake so counselors could spend more time listening. The outcome backfired.

​

Stanford's Human-centered AI researchers warn that mental-health chatbots can simulate empathy but fail in real crises—sometimes exacerbating stigma or leaving emergencies unaddressed (HAI Stanford). And when the nonprofit Crisis Text Line used algorithmic triage and controversially shared anonymized crisis data with a for-profit AI arm, trust cracked under the weight of efficiency claims (WIRED).

​

• Telos error. The end shifted from deepening care to maximizing call throughput.

• Ontology error. Callers were reduced to keywords and timestamps.

• Epistemology error. Speed of transfer was treated as evidence of success, even as bottlenecks and dropped calls increased.

 

The irony was brutal. In trying to gain efficiency, the hotline lost it. Average wait times went up. Personal connection went down.

 

The Lesson

These cases make the same point: the real errors weren’t in the code. They were in the assumed philosophy. Misstated ends. Misclassified realities. Mistaken evidence.

​

The philosopher’s task is simple but disruptive: replace unexamined assumptions with examined commitments. That’s not an academic luxury. It’s the difference between making a tool that merely works and making a tool that works for the good.​
 

Bugs Not Philosophy?

This is where the skeptic clears his throat. Philosophy slows us down. Philosophy is indulgence. Philosophy is just Severance without the elevator. All the head games, none of the stakes.

 

And then: Those errors you describe? They’re not philosophy. They’re bugs. Bugs get patched.

 

Horizon wasn’t a bug. Logs are not witnesses.
Tay wasn’t a bug. Twitter is not a teacher.
The fintech case wasn’t a bug. Fraud reduction is not trust.

 

These are errors of telos, ontology, and epistemology. Errors of meaning, not mechanics. And meaning scales faster than code.​ ​​These aren’t patches waiting to happen. They are errors at the level of philosophical assumptions. And assumptions scale faster than code.

​

We already have ethics, the skeptic insists. Yes, but ethics usually arrives downstream as compliance, regulation, and crisis management. Philosophy is upstream. If you don’t ask the questions, you are still answering them. Just badly.

​

Users decide values. No. Preferences are not values. “Engagement” is not flourishing. Illich’s line cuts here: convivial tools deepen freedom. Technocratic ones reduce us to data.

​

The most likely pushback: There is no ROI. Paying for philosophers doesn’t pay off. Four replies. 

 

First, philosophers can reduce failure costs. Many spectacular product failures are category errors in disguise: mistaking correlation for causation, simulation for understanding, scale for legitimacy. Remember Theranos? The company claimed to revolutionize blood testing by using just a few drops of blood. Investors believed, patients trusted, and the machines never worked. (WIRED) The founder, Elizabeth Holmes, deceived investors, celebrities, and institutions by blending charisma and inflated promises with the cultural currency of science.

​

A philosopher would have asked: What counts as sufficient evidence for investing in new health technology that non-scientists don’t understand? Clearer thresholds might have stopped years of deception before the first drop of blood was drawn. 

​

Second, philosophy accelerates clarity. Teams move faster when they share the same definitions of “safe” or “helpful.” In 2016, Facebook turned user data into a political weapon. Engagement metrics eclipsed community, and trust was traded for manipulation. Without asking "why?" Without asking "What is this platform for?" With its telos too vague and misguided, Facebook's “community” collapsed into clicks. Democracy became collateral damage to a technological monolith. 

​

Third, philosophy creates explainability for humans. Not interpretability in a narrow technical sense, but the kind of narrative clarity that leaders need to justify decisions to regulators, journalists, and, most crucially, users. Products like Just Walk Out and Google Glass failed partly because no one could explain what human good they served.

​

Fourth, philosophers can create superior outcomes for customer loyalty. Again, the MIT published work of Schrage and Kiron demonstrates this. Organizations that deliberately embed philosophical reflection into their AI design (ontological, epistemological, and teleological), rather than merely optimizing for KPIs, cultivate deeper, more meaningful forms of customer loyalty and derive superior strategic and ethical outcomes. Starbucks’ Deep Brew was framed around “connection.” Amazon Prime redefined what it means to “know” its best customers. In both cases, philosophical reframing turned metrics into durable loyalty.

 

I'm not suggesting that philosophy would have solved every technical flaw, but it would have forced the right questions before launch. A fuller catalogue of these failures and how philosophy could have prevented them appears in the appendix below.

​

Think of philosophy as the organization’s sense-making stack: the way you decide what counts as a good outcome, a real improvement, a justified inference, the actual problem to solve, and questions to ask. When that stack is weak, the smartest model in the world will optimize you into a ditch. Think of the philosopher as the one navigating you around ditches and away from cliffs.

​

Disruptive Clarity

Here’s the brutal truth: in the absence of philosophy, technocracy wins by default. Control is easy to code; human flourishing is not. And unless someone calls out the philosophical mistakes, our tools will quietly train us to expect less of each other, and of ourselves. 

​

Philosophers have always been dangerous for this reason: by calling out the dangerous philosophical mistakes we live with. Socrates unsettled Athens until it executed him. Kierkegaard mocked Christendom in its own language. Simone Weil saw through efficiency and prized attention long before factories began to digitize both. The philosopher's craft, if they are to serve the public good, is disruptive clarity. That is exactly what AI challenges need.

​

Going Bigger than Steve Jobs

Steve Jobs radically changed the computer industry by employing an experience-first philosophy of development: focus on the user experience first, then work backwards to the technology. This reorientation toppled an entire generation of hardware-first thinking. 

 

Job's philosophy has also toppled human flourishing. TikTok’s algorithm is arguably the most user-centered design yet. But by aiming only at engagement, the platform has become a machine for mental health destruction. User experience wins, while human flourishing loses. 

​

For the visionary that he was, his vision was still too small. We need to begin with human flourishing and from there develop the user experience, and then the technology. User experience is always second to user flourishing. 

​

Ivan Illich once warned that tools either deepen our convivial life together or bind us more tightly to the imperatives of technocratic control. AI is now pressing that choice with new urgency. To imagine philosophy irrelevant to this moment is not humility but abdication. A refusal of responsibility for what we are making of one another.

​

The Cultural Stakes

​We are living through an age that treats the human as an adjustable parameter. We outsource memory, attention, and judgment to systems we barely understand. The crisis behind the crises is not technical. It is anthropological—a crisis of what we think a human person is for. Tools that ignore this quietly deform us. Tools that face it honestly can help shape us into more courageous, patient, just, and joyful individuals.

​

At the cultural level, the stakes are formation, not features. We like to think we use tools neutrally, but over time, our defaults shape us. “Always on” becomes “always available”; “frictionless” becomes “thoughtless.” The result is not simply distraction; it is a thinning of shared meaning. 

​

Google’s AI Overviews risk collapsing curiosity into canned answers. Amazon reframed surveillance as trust. Horizon collapsed data into testimony. Replika turns intimacy into simulation. AI therapy apps risk mistaking patterned text for presence.

 

These errors don’t just break products. They re-form people. Smaller ends. Shallower vision. Misplaced trust.

​

We do not lack information. We lack common frames for deciding what matters and why. That’s where philosophy earns its keep. It gives leaders and institutions the habit of asking: What is this for? What does it assume about persons? What good is it cultivating or crowding out?

 

That is why leaders are saying the quiet part out loud. Google CEO Sundar Pichai explained on 60 Minutes that shaping AI requires philosophers at the table. (CBS News). Demis Hassabis, CEO and co-founder of Google DeepMind, warns: “We need new great philosophers… in the next five, ten years to understand the implications” (CBS News). Steven Johnson puts it bluntly: the questions that matter “no one was thinking about, except for philosophers.” And the list goes on. See my essay Why Silicon Valley’s Brightest Minds in AI Are Calling On Philosophers.

​

That isn’t nostalgia for Plato and Aristotle. It’s an admission that the AI sprint is outpacing the moral map. AI builders know the void they’re circling. 

 

The Invitation

Let me end where I began. 

 

Why?

 

Why are you building or using AI? What end does it serve?  Are people being flattened into datapoints?  What counts as knowledge in your model? Are outputs being mistaken for understanding? When your product succeeds, does life expand or contract? Does your design help train users in justice, generosity, and patience, or in greed, anger, and envy? 

​

But, tech leaders must understand that simply calling on a philosopher is not enough. Meeting AI challenges requires philosophers trained in technology, in moral knowledge, and in human flourishing. Philosophers who will audit assumptions in Slack threads, AI labs, and boardrooms. Who will make sure the “why” shows up before the code does. Who knows that their craft is disruptive clarity, aimed at dignity, agency, patience, justice, and joy.​

​

What AI needs are philosophers who know that their craft is disruptive clarity, aimed at human flourishing.

 

Ask why until the room falls silent. Keep asking until the room catches the vision. Ask until the tool’s purpose is clear enough to defend in public. Ask until they understand that technology is never just what it does. It is also what it makes of us.

​

The future of AI will not turn on who codes the fastest system, but on who has the nerve to ask why—the question machines cannot ask, and too many refuse to.​