May 19, 2026 | 14 min read
Listen to the full episode: 🎧 Spotify 🎧 Apple Podcasts ▶️ YouTube
Free resource: Outro music by CLAWS. Used with permission. Listen on Spotify
Direct answer: Spencer Kelly, Cambridge computer scientist and former BBC Click presenter, argues that the most important distinction in the entire AI conversation is one the industry consistently fails to make: intelligence and consciousness are not the same thing. AI systems can demonstrate extraordinary intelligence, producing outputs that exceed human experts across a growing range of tasks, while operating in complete darkness, with no awareness, no understanding, and no inner experience of any kind. Drawing on two decades of technology journalism, academic grounding in AI from before the current era, and the specific failure of his own predictions about Spotify and the iPad, Spencer makes the case that knowing what AI actually is rather than what it appears to be is the most commercially significant thing any leader deploying it can understand. His argument is not against AI adoption. It is against adoption without the cognitive clarity about what is being deployed. The lights are off. The system does not understand anything it is doing. And the governance, liability, and strategic frameworks being built around AI in healthcare are almost universally failing to account for that.
Intelligence Is Not Consciousness. And That Distinction Changes Everything.
The single most important sentence in this episode is also the one that will be most uncomfortable for anyone building, deploying, or investing in AI in healthcare.
Spencer Kelly says it plainly, with no qualification and no softening. "There is nothing behind those words. The lights are off. There is no experience. There is no judgement. There is no critical thinking. There is just nothing."
This is not the language of an AI sceptic. Spencer has studied artificial intelligence since his Cambridge dissertation in the 1990s. He has reported on it for BBC Click across two decades of material advances. He builds with it daily, makes music with it, writes with it, explores it with genuine enthusiasm. His argument is not that AI is useless or dangerous or overhyped. His argument is that it is precisely understood by almost nobody deploying it at scale.
The distinction he draws is between intelligence and consciousness. Intelligence, in the most operationally useful definition he offers, is what an entity deploys when it does not know what to do. A chess-playing programme that was hard-coded to respond to specific positions has intelligence in a narrow sense. It can handle the situations it was built for. What it cannot do is encounter a situation outside its instructions and reason about what to try. Large language models and modern AI systems have moved far beyond that narrow form. They can handle novel inputs, synthesise across domains, and generate outputs in situations they have never explicitly encountered. That is intelligence at an extraordinary level.
Consciousness is different. Consciousness is the awareness that you are doing something. It is the inner experience of a process rather than just the execution of it. And Spencer's position is that AI systems do not have this, will not have it on the current architectural trajectory, and that the entire AI industry is allowing a conflation of the two concepts to serve their fundraising and public positioning interests.
"There is nothing behind those words. The lights are off. There is no experience. There is no judgement. There is no critical thinking. There is just nothing. That is the way you have got to use these things."
For dental leaders and healthtech founders, this has a specific commercial implication. Every AI diagnostic tool, every clinical decision support system, every treatment planning algorithm being deployed into dental practices right now is operating on the same basis. The outputs can be extraordinarily good. Better than a human clinician at detecting early caries on a bitewing radiograph. Better than a practice manager at identifying scheduling inefficiencies. Better than a billing team at catching coding errors before submission. But the system producing those outputs has no understanding of what a tooth is, what a patient is, or what harm means.
The governance and liability frameworks being constructed around dental AI in the UK are almost universally silent on this. The GDC's fitness to practise standards predate AI deployment at clinical scale. The MHRA's medical device classifications were written for a world where algorithms followed explicit rules. The indemnity structures dental practitioners operate under were designed for human error. None of them adequately account for the specific failure modes of a system with extraordinary intelligence and no consciousness whatsoever.
What Being Wrong About Spotify Taught Spencer About Disruption — and What It Teaches Dentistry About AI
In the mid-2000s Spencer sat across from Daniel Ek, the founder of Spotify, and told him the service would never be commercially viable. His analysis was technically precise: the cost of licensing music at the scale Ek was describing, combined with the free-tier model he was proposing, made profitability structurally impossible. The economics did not work.
He was correct about the economics. He was completely wrong about everything that mattered.
Spotify made losses of hundreds of millions of dollars for more than a decade. It did not matter. What mattered was that Spotify built infrastructure, established habits, and created library lock-in at sufficient scale to make the previous model of music consumption structurally irrelevant before the incumbents could respond. Once people's libraries were on Spotify, once the recommendation engine understood their tastes, once the social and playlist infrastructure was built, the pricing could be turned on. The economics caught up with the disruption rather than the other way around.
"I hadn't understood the power of disruption. As long as you have enough money to disrupt and end the previous model, you can do what you want. And eventually, Spotify and other music streaming services have completely changed the way we consume music."
Spencer draws the same structure around AI adoption in healthcare. The dental practices evaluating AI deployment primarily through an immediate ROI framework are asking the wrong question. The question is not whether the AI tool pays for itself in year one. The question is whether the practices that adopt early will be building infrastructure, capabilities, and data assets that make the position of the non-adopting practice structurally weaker over a five to ten year horizon, regardless of the current economics.
The Blockbuster equivalent in UK dentistry is the practice that evaluates AI through cost avoidance rather than structural positioning. Blockbuster had the same access to technology, the same customer relationships, and more established infrastructure than Netflix for years. What it did not have was the willingness to accept short-term economic pain in exchange for the infrastructure that would define the next generation of the market. The dental practices and groups making their AI adoption decisions on the basis of immediate financial justification are making the same calculation that looked rational from inside Blockbuster in 2003.
Is AI Actually Different This Time? Spencer's Most Honest Answer.
Spencer has been asked a version of this question about every significant technology for twenty years. The printing press automated the copying of manuscripts and scribes predicted the end of their profession. The agricultural revolution eliminated most farm labour over a century. The fridge ended the business of delivering ice by horse and cart. Car-building robots were going to eliminate manufacturing employment. Each wave of automation produced predictions of mass technological unemployment. Each wave, measured over any meaningful time horizon, created more work than it destroyed.
The standard argument from AI optimists is that this cycle will be identical. AI is just automation at a higher level of cognitive sophistication. The same economic logic that created new jobs from every previous automation wave will operate here. New categories of work will emerge that we cannot yet name, just as social media manager was not a job category before Facebook existed.
Spencer's answer is the most honest I have heard from someone with his combination of technical background and journalism experience. He is genuinely torn.
"Half of me thinks surely this is going to be the same. But then the other half thinks, this time it does feel different. Because we have always said the machines can do the dangerous, physical, heavy lifting. But we will still be making the decisions. And AI is creeping into that now."
The specific distinction he draws is the one that matters for healthcare. Every previous automation wave targeted physical or procedural work. The justification for human employment always retreated to cognitive and creative tasks. Machines could build cars but they could not design them. Machines could sort mail but they could not write it. Machines could process data but they could not interpret it. That cognitive distinction was the stable ground on which human employment rested through every previous disruption.
AI is now targeting that cognitive ground. Not all of it, and not in a way that suggests imminent wholesale replacement of human judgement in complex domains. But enough of it, at sufficient speed, that the previous stable boundary between human and machine capability is moving for the first time in the history of automation.
Spencer does not resolve this uncertainty. He holds it. And that intellectual honesty, from someone who has been close enough to the technology for long enough to have earned the right to make confident predictions, is itself the most useful signal. The people making confident predictions about AI's impact on employment in either direction are working from the same evidence Spencer has and reaching more certain conclusions. That asymmetry should be noted.
What Dentistry Gets Wrong About This Conversation
The dental AI conversation in the UK in 2026 is almost entirely a conversation about products. Which diagnostic tool. Which imaging software. Which PMS has the best AI feature set. Which DSO is furthest along the deployment curve. These are reasonable questions. They are not the most important ones.
The most important question, the one Spencer's framing forces to the surface, is what dentistry is actually asking AI to do and whether the governance structures around those tasks are adequate given that AI has no understanding of what it is doing.
AI diagnostic tools can identify caries on a radiograph with greater consistency and at higher sensitivity than many human clinicians. That is not a claim about AI understanding. It is a claim about pattern matching at scale. The system has processed millions of radiographs and learned to identify the pixel patterns associated with carious lesions with extraordinary precision. It has no understanding of what a tooth is, what pain means, or what the clinical consequences of a missed diagnosis are. The accuracy of the output does not change the nature of the process.
The clinical governance implication is that deploying AI diagnostics in dentistry is not the same as deploying a more experienced clinician. It is deploying a very precise pattern matcher in a context where the mismatches, the cases that fall outside the training distribution, are exactly the ones where clinical judgement matters most. The practices and groups that understand this will build oversight structures proportionate to it. The ones that treat AI as a smarter colleague will be the ones generating the liability cases that define the regulatory environment for everyone else.
For a deeper analysis of how the legal framework is failing to keep pace with clinical AI deployment in UK healthcare, see When AI Gets It Wrong: Liability, Governance and the Legal Frontier of Dentistry at techdental.com/insights.
Key Takeaways
1. Intelligence and consciousness are not the same thing. AI systems can demonstrate extraordinary intelligence while operating with no awareness, no understanding, and no inner experience. Building governance frameworks around AI in healthcare without accounting for this distinction is the most common and most consequential error in AI deployment today.
2. The Spotify lesson is about infrastructure, not economics. Spencer was correct that Spotify's economics were impossible in the short term. He was wrong about what mattered. The dental practices evaluating AI adoption primarily through immediate ROI are asking the Spotify question rather than the Netflix question. The infrastructure shift is the point, not the payback period.
3. The stable cognitive boundary between human and machine work is moving for the first time. Every previous automation wave displaced physical or procedural work. AI is displacing cognitive and creative work. Whether this produces the same net positive outcome as previous waves is genuinely uncertain. The people who tell you they know which way this goes are more confident than the evidence warrants.
4. Being wrong in public is a more useful qualification than being confidently right. Spencer's credibility on AI comes not from his predictions but from his systematic examination of why he got things wrong. The intellectual posture of someone who has had their predictions fail twice in front of a global BBC audience, and who has continued to engage seriously with the subject, is exactly the posture that AI commentary in healthcare currently lacks.
5. The safest job in the world right now might be a plumber. This is not a joke. AI's extraordinary capability in language, image, and knowledge tasks is matched by its complete inability to navigate the physical, contextual, and unpredictable world of manual trades. The practices and groups thinking seriously about which clinical roles are most and least exposed to AI displacement should start with Spencer's physical world argument rather than the vendor's feature list.
About TechDental
TechDental is a strategic intelligence platform for founders, executives, operators and investors shaping the future of dentistry. Through high-level analysis and systems-focused conversations, we explore how AI, governance frameworks and operating model design influence performance, scalability and enterprise value in dental organisations.
LinkedIn: https://www.linkedin.com/in/drrandeep/
The future belongs to those who deploy technology with discipline.
© 2026 RIG Enterprises Limited. All Rights Reserved. TechDental is a trading name of RIG Enterprises Limited (Company No. 11223423), incorporated in England and Wales on 23 February 2018, registered at 1a City Gate, 185 Dyke Road, Hove, England, BN3 1TL. Opinions attributed to named guests reflect the views of those individuals as expressed during recorded interviews. No part of this article may be reproduced without the prior written consent of RIG Enterprises Limited.