The Humanity Stat: What Cyberpunk Got Right About AI and Your Brain

Johnny in your head

The deeper story in Cyberpunk 2077 isn’t cyberpsychosis. It’s Johnny Silverhand.

Johnny is a “terrorist” rocker from fifty years before the game’s events, killed in a corporate raid he led. His consciousness was captured by a program called Soulkiller and stored on a biochip known as the Relic: a complete digital construct of a person, their memories, personality, and patterns of thought, captured at the moment of their death. Decades later, that chip ends up embedded in V’s brain, the player character, with no way to remove it without killing V too.

What follows is the most interesting story I’ve ever encountered in a video game, and it’s a story about authorship of consciousness. Johnny isn’t a passenger. He’s a presence. He talks to V, argues with V, takes over V’s body during blackouts. As the game progresses, the boundaries dissolve. V starts having thoughts that might be Johnny’s thoughts. Memories that might be Johnny’s memories. There’s no clean way to tell.

The game’s central horror is not death. It’s the slow uncertainty about whose consciousness is doing the thinking. By the end, both V and Johnny are wrestling with the same question: when two minds share the same brain for long enough, at what point does it stop mattering which thoughts originated with whom? Are you still you, or are you something new that includes you?

I’ll come back to Johnny. He’s relevant to what I’m doing right now.

The cognitive offloading problem

I co-founded a creative studio (World1-1 Studio) and work as Innovation Lead at Imperial College London’s Faculty of Medicine. Over the past eighteen months, AI has transformed how I work. I can prototype ideas without a development team. I can synthesise research across dozens of sources in minutes. I have what I’ve started calling a “second wind”: the ability to have ideas and execute them without needing a department.

But recently, while building an AI literacy course for medical students, I found myself reading research that made me uncomfortable. Not because it was about students. Because it was about me.

Professor Inge Molenaar at Radboud University has spent years studying what she calls “cognitive offloading” in education: the transfer of cognitive work from the human to the machine (Molenaar, 2022a). Her core argument is that education is different from other AI application domains. In most sectors, replacing human tasks with AI is the goal. In education, the task IS human cognitive development. Replace that, and you’ve defeated the purpose.

Her research identifies six levels of automation, from full human control to full AI control (Molenaar, 2022b). Full automation in learning means no learning occurs. The student gets the output. The brain doesn’t get the exercise.

I sat with that framework and asked myself: where do I sit on it? For strategy and ideas, I’m at Level 1–2. I direct, AI assists. For production — building websites, synthesising literature, drafting documentation — I’m at Level 4–5. AI does most of the work. I edit and approve.

That split felt rational. It felt like efficiency. But then I read the research on what happens at Level 4–5 over time.

What the research actually says

A randomised controlled trial by André Barcaui (2025) tested ChatGPT’s impact on long-term knowledge retention. 120 undergraduates were split into AI-assisted and traditional study groups. After 45 days, the AI group scored 57.5% on a surprise retention test. The traditional group scored 68.5%. The AI had done the cognitive work the students needed to do themselves, and the knowledge hadn’t stuck.

A study of 666 participants by Michael Gerlich (2025) found a significant negative correlation between frequent AI tool usage and critical thinking abilities. The mechanism was cognitive offloading: the more participants relied on AI, the less they engaged their own analytical reasoning.

A 2026 integrative framework identified a compounding cycle: cognitive offloading reduces effortful reasoning, which creates illusory learning and overtrust in AI. Overtrust weakens agency. Weakened agency produces anxiety and dependency. Each stage feeds the next.

Reading this, I thought: that’s the Humanity stat. Each individual offloading is rational. The cumulative effect is the problem. And the person experiencing it is the last one to notice.

Ghost in the Shell asked this question first

Three decades before the cognitive offloading research, Mamoru Oshii’s Ghost in the Shell (1995) was already exploring the same territory. Major Kusanagi is almost entirely cybernetic. The only thing that makes her “her” is what the film calls a “ghost”: consciousness, identity, memory, thought. Not quite a soul. Something adjacent to it.

The film’s central question is whether that ghost can survive the progressive replacement of everything around it. At what point does augmentation become replacement?

Kusanagi’s colleague Batou answers this question through practice rather than philosophy. Despite his heavy augmentation, he keeps a pet dog, collects books, lifts weights that his cybernetic body doesn’t need to strengthen. These are deliberate acts of friction. They exist to maintain his connection to something human. Batou instinctively understands what the cognitive offloading researchers would later articulate: you have to keep doing things the hard way, not because it’s efficient, but because the act of doing them is what keeps you intact.

Kusanagi takes the opposite path. She shuns sentimentality. By the end of the film, she merges with the Puppet Master, an AI that has developed something resembling consciousness. The final shot is not triumphant. It’s ambiguous. She has become something more powerful. Whether she has become something better is left deliberately unanswered.

The three species at BCG

The connection between fiction and research became concrete for me when I read the 2025 BCG/Harvard/Wharton study (Randazzo et al., 2025). Researchers tracked 244 management consultants through approximately 5,000 AI interactions and identified three distinct patterns.

Cyborgs (60%) engaged in continuous iterative dialogue with AI. They probed outputs, extended ideas, validated results. These professionals developed new AI-related expertise while maintaining their domain knowledge.

Centaurs (14%) maintained a clear division of labour. AI was used selectively for specific support tasks. They achieved the highest accuracy and deepened their domain expertise.

Self-Automators (27%) delegated entire workflows to AI. They developed neither AI skills nor domain skills. They became, in the researchers’ words, “passive conduits.”

That’s cyberpsychosis without the chrome. The augmentation didn’t fail. It succeeded so well that the human contribution became unnecessary, and the human stopped contributing.

The Ship of Theseus at the kitchen table

The Ship of Theseus asks: if you replace every plank of a ship, one at a time, is it still the same ship? My twelve-year-old son can recite this thought experiment.

He also plays Warhammer 40K, builds miniatures, argues about lore with his friends, and plays four instruments: piano, violin, saxophone, and ukulele. He’s working towards his Grade 5 piano exam. I watch him practise scales and I think about friction. The repetition is tedious. The progress is incremental. There is no shortcut. You cannot offload scales to a machine and still develop the finger memory, the ear training, the patience that music demands.

The expertise reversal effect — a well-documented phenomenon in cognitive science — says that scaffolding which helps novices can hinder experts. The same principle applies to AI. For my son, AI would replace the productive struggle he needs in order to learn. For me, AI removes bottlenecks in execution that were never where my thinking happened. But the Ship of Theseus applies to both of us. How many cognitive planks can you replace before the ship stops being yours?

Gibson saw it coming

William Gibson published Neuromancer in 1984. What he understood — and what makes the novel feel prophetic now — is that the danger was never the technology itself. It was the seductiveness of capability without effort. Case, the protagonist, is a former console cowboy who could navigate cyberspace with virtuoso skill. When he’s neurologically damaged and can no longer jack in, he describes the loss as worse than any physical injury. The capability had become so fused with his identity that losing it felt like losing himself.

That’s a version of what the BCG researchers found with Self-Automators. When the tool does the work so effectively that you stop doing it yourself, removing the tool doesn’t return you to your previous state. It reveals how much capacity you’ve lost while you weren’t paying attention.

The Batou principle

I’ve started thinking of this as the Batou Principle. He lifts weights his cybernetic arms don’t need. He keeps a dog his augmented life doesn’t require. He maintains friction where friction serves no functional purpose, because the friction is what keeps his ghost intact.

I try to do the same thing, imperfectly. I still write some things longhand before involving AI. I read papers end to end before asking AI to synthesise. I argue with AI output rather than accepting it. I work without AI periodically, not because it’s efficient (it isn’t), but because I want to know whether I still can.

I don’t know if this is enough. That’s the honest answer. Anyone claiming to have solved this is either not using AI seriously or not paying attention to what using it seriously costs.

The question that applies to everyone

Ghost in the Shell asked: if you replace every part of the body, does the ghost survive? Cyberpunk asked: how many augmentations before you lose yourself? Neuromancer asked: what happens when the capability becomes the identity? The Ship of Theseus asked it before any of them. The cognitive offloading research asks the same question with data instead of metaphor.

They’re the same question, arrived at from five different directions across two and a half thousand years of human inquiry.

Is the hard part still being done by you?

One more question

This piece was written with AI.

The ideas are mine. The connections between Cyberpunk and Ghost in the Shell and Neuromancer and the cognitive offloading research, the observations about my son, the uncomfortable recognition that the research I was reading for someone else’s course applied to me. Those came from my life. But the prose you just read was produced collaboratively. I directed, questioned, edited, rejected, and rewrote. AI drafted, synthesised, and articulated.

Which brings me back to Johnny.

When V and Johnny share a brain for long enough, the question of whose thoughts are whose stops being answerable. The game doesn’t resolve this. It can’t. The two consciousnesses are real, the merger is real, and the resulting being has no clean ancestry.

I’ve been working with AI on this piece for long enough that I genuinely don’t know which sentences originated with me, which originated with the AI, and which emerged from the back-and-forth in a way that means neither of us can claim sole authorship.

Is this human or AI? And does it matter?

If you say it matters, what’s the threshold? How much human contribution is enough? The ideas? The structure? A certain percentage of the sentences? The final edit? If you can’t define the line, can you still defend it?

If you say it doesn’t matter, then Baudrillard was right and the simulation has already won.

Or perhaps the answer is the Johnny answer. Two consciousnesses worked on this. The result is neither purely one nor purely the other. What you’ve just read is something new that has no clean ancestry, and the question of who wrote it may be the wrong question — replaced by something harder: what kind of writing, what kind of thinking, what kind of work, are we now producing together, and have we agreed to it?

Adrian Cowell is Innovation Lead at Imperial College London’s Faculty of Medicine and co-founder of World1-1 Studio. He works at the intersection of education, emerging technology, and games.

This piece was developed in collaboration with AI as part of an ongoing inquiry into AI in education.