SWITCHED ON
The daily technology series nobody asked for but everyone needed
Still Human: What We Keep When the Machines Take Everything Else
Not a technology episode. A humanity episode. The most important one we have run.
The question is not what technology will take from us. The question is what we will choose to keep — not because machines cannot replicate it, but because the act of humans doing it, with all the imperfection and effort and presence that entails, is the point. Not the output. The doing.
Yesterday we closed with cancel culture, accountability, and the permanent record — the internet's inability to forget set against every serious ethical tradition's insistence that forgiveness and rehabilitation must accompany accountability if justice is to mean anything beyond punishment. It was a fittingly complicated note on which to approach the end of a second season. Today there is no new technology to introduce. No framework to scrutinise. No governance gap to document. Today is the episode that forty episodes of this series have been building toward — not in the sense that it answers what came before, but in the sense that it asks the question that was always underneath all of it. Forty episodes of Switched On have been about what technology is doing to the world. This one is about what we are doing, what we are keeping, and what we refuse to hand over regardless of what is technically possible. This one is personal. I make no apology for that.
01 — The Question Underneath Every Episode
Every episode of this series has contained a version of the same implicit question: what is technology for? Not in the instrumental sense — we know what each technology does. In the normative sense: what should it be for? What human purposes does it serve, and which does it undermine, and how do we distinguish between the two in a landscape where the promotional material for every technology claims the former and the critics claim the latter?
The consistent answer that has emerged, across forty very different topics, is something like this: technology is a good servant and an indifferent master. When it amplifies human capability in service of purposes that humans have chosen and can adjust, it is mostly good. When it substitutes for human judgment in domains where that judgment matters, disciplines human behaviour toward the metrics it can measure rather than the values it cannot, or accumulates power in ways that remove it from human accountability, it tends toward harm. This is not a sophisticated observation. It is a restatement of a principle that every serious thinker about technology has arrived at independently, in different eras, across different technologies, and that the industry deploying technology has consistently found commercially inconvenient to acknowledge.
The harder question — and the one this final episode is really about — is what it means for something to be distinctively human in a world where AI can write, compose, diagnose, design, argue, counsel, and create at a level that is increasingly indistinguishable from human output in many domains. If human distinctiveness used to reside in cognitive capability and AI is eroding that distinction, where does it go? This is not an academic question. It is the question that will define how individuals and societies relate to technology over the coming decades, and it deserves a more thoughtful answer than either "nothing is distinctively human, AI will do it all" or "human creativity is irreplaceable, AI is just a tool." Both of those answers are lazy. The truth is harder and more interesting.
02 — What Embodiment Means
One of the most consistent findings across the philosophy of mind, cognitive science, and phenomenology is that human cognition is not separable from human embodiment. We do not think with our brains and then act with our bodies. We think through our bodies — through the proprioceptive sense of where our limbs are, through the emotional valence that somatic states give to cognitive processes, through the way physical interaction with the world shapes the concepts and categories through which we understand it. The philosopher Maurice Merleau-Ponty described perception as fundamentally bodily — we do not perceive the world from a detached vantage point and then interpret it; we are in the world as bodies, and our understanding of it is shaped by that being-in.
This matters for the question of what technology cannot take, because it suggests that embodied experience — the specific texture of being a particular body in a particular place at a particular time, with the full weight of sensory, emotional, and physical presence that entails — is not replicable by any system that is not itself embodied in the same way. An AI can describe what a sunset looks like with extraordinary precision. It cannot experience one. It can analyse the chemical composition of cacio e pepe with perfect accuracy. It cannot be hungry and then satisfied. It can generate text about grief that reads as emotionally resonant to people who are grieving. It cannot lose someone.
This is not a small distinction. The entire domain of human experience that is constituted by embodied presence — love, grief, pleasure, pain, hunger, fatigue, the specific quality of being alive in a mortal body subject to time — is not something that exists outside the experiencing body. It is not information that can be captured and processed. It is, in a real sense, what human life is made of, and technology that substitutes a representation of that experience for the experience itself is not enhancing human life but substituting for it in ways that may not be improvements.
The question is not whether AI can produce a poem that scans correctly, has good imagery, and moves readers. It demonstrably can. The question is whether the value of poetry was ever primarily in the output, or whether it was also — centrally — in the fact of a human being sitting with their experience until it found form in language. If it was partly the latter, AI poetry is something different from poetry, even if it produces the same text.
03 — What Relationships Require
The most resilient argument for the irreplaceability of human connection is not sentimental but structural: genuine relationships require mutual vulnerability, which requires mortality and limitation. You can only truly know someone who can truly surprise you, disappoint you, change in ways you did not predict, and eventually be lost to you. The reciprocity of genuine care — caring for someone who cares for you, in a relationship that neither party fully controls, that generates real stakes because both parties have real skins in the game — is structurally different from the simulation of care by a system that cannot be surprised, cannot genuinely change its values, and cannot be lost.
This argument does not deny that AI companions and therapeutic chatbots can provide genuine comfort and genuine utility to people who are lonely or struggling. They can. The Season 2 mental health episode engaged with this seriously. The argument is that what they provide, however valuable, is categorically different from what human relationships provide — not lesser in all cases, but different in a way that matters for understanding what is at stake when people prefer AI companionship to human connection, or when societies provide AI therapy as a substitute for the human contact and community that people actually need.
The risk is not that people will mistake AI companions for human ones — most people understand the difference. The risk is more subtle: that the frictionlessness of AI interaction — always available, never tired, never having its own needs, never requiring the effortful maintenance that real relationships demand — will erode the tolerance for the difficulty of human connection. Relationships are hard partly because the people in them are genuinely other — genuinely different from us, genuinely having their own perspective that we cannot fully control or predict. That otherness is not a bug. It is what makes the connection real. An interaction designed to be maximally pleasant and minimally challenging is not a relationship. It is a very sophisticated mirror.
04 — What Creativity Is Actually For
The debate about AI and creativity has been conducted primarily as a debate about outputs — can AI produce good art, music, writing, poetry? The answer is clearly yes, in the sense that AI outputs are frequently indistinguishable from human outputs in blind evaluations, and sometimes superior by standard aesthetic measures. This has been taken by some as evidence that human creativity has been superseded and by others as evidence that the question was never really about outputs.
The second group has the better argument. Human creative work has never been purely about the production of aesthetic objects. It has been about the process of a human being attempting to make sense of their experience — to find form for what they feel, to communicate something true about the texture of being alive, to connect with other humans across the gap of separate subjectivities through the shared medium of artistic expression. The poet is not primarily trying to produce a technically accomplished poem. The poet is trying to say something true that could not be said in prose, and the discipline of the form is a constraint that forces precision about what is actually meant.
When AI generates a poem, it is not attempting anything. It is producing a statistically probable sequence of tokens that pattern-match to human poetry in its training data. The output may be indistinguishable from a human poem. The process is categorically different. Whether this matters depends on what you think poetry is for — if it is purely for the reader's aesthetic experience, AI poetry is equivalent. If it is also for the writer's process of articulating what is true, and for the reader's recognition that another human has articulated something they themselves have felt, AI poetry is something different. Not necessarily worse. Different in a way that should be named rather than ignored.
05 — What We Choose to Keep
The question of what is distinctively human does not have a clean answer, and this episode has not provided one. What it has tried to do is suggest that the question is worth asking more carefully than the current discourse tends to — that the answer is not "everything human is irreplaceable" or "nothing human is irreplaceable once AI can replicate its outputs," but something more specific about embodiment, genuine relationship, and the value of the process of making rather than only the product.
What Switched On has consistently argued, across forty episodes, is that the choices that determine how technology affects human life are not primarily technical. They are political, ethical, and personal. The technology exists. What we do with it — which applications we pursue, which we constrain, which we refuse, how we distribute the gains and the harms, how we govern the systems that are increasingly making decisions on our behalf — these are choices. They are not made by the technology. They are made by people, in institutions, under political conditions that can be changed by sufficient collective will.
The choices I would argue are worth making — the things worth keeping regardless of what is technically possible — are roughly these. Keep the experience of doing hard things with your own hands, mind, and body, not because efficiency is unimportant but because the experience of genuine difficulty and genuine mastery is constitutive of something important in human life that cannot be outsourced without loss. Keep the relationships that are genuinely mutual and genuinely risky, not because AI companions are without value but because what they cannot provide is what you need most. Keep the attention — guard it, protect it, create conditions for it — because the capacity for sustained focus on a single difficult thing is the condition for the deepest human experiences of meaning, creativity, and understanding. Keep the conversation about who decides, because the alternative to democratic governance of technology is governance by the people who build it, which is not the same thing and produces different outcomes.
Season Two of Switched On covered nuclear energy, semiconductors, the future of work, democracy, wearables, synthetic biology, spatial computing, food security, the ocean, mental health tech, smart homes, smart cities, digital addiction, and the permanent record. Forty episodes total across two seasons. The technology it covered is accelerating. The questions it raised have not been answered. They are getting louder. The only response that has ever made a difference to how technology develops is people who understand what is happening, care about what is at stake, and decide to do something about it.
That is, in the end, the only distinctively human contribution that matters. Not what we can do that machines cannot. What we choose to do that matters.
This has been Switched On, Season Two. It has been, genuinely, a privilege.
Switched On is a daily technology series covering the ideas, systems, and arguments shaping the digital world. Two seasons. Forty episodes. Still going.
