Talks

The future of AI: law, regulation, creativity, health — Ryan Abbott with Dinis Guarda

53 min read

Ryan Abbott · 2021 · 1:28:09 · ▶ Watch

Summary

Central question. The host asks, over and over, what happens to law, ethics and human life when AI does things that only people used to do. Abbott answers with one lens: law should follow behaviour, so the same rule should cover a person and a machine doing the same thing.

Main thesis (Abbott).

  1. AI should be regulated now, by a mix of binding law, ethics, norms and industry standards. Existing rules often pre-date AI and were not written for it.
  2. AI should be regulated “more like a legal entity” because it behaves like one. That is not the same as giving it rights, which he rejects for today’s AI.
  3. On social manipulation, platforms have a duty to restrict misinformation, governments have a role, and AI is both the problem and the tool for the clean-up.
  4. Countries should talk to each other and harmonise where they can, especially on IP and data. The EU leads on ethics. WIPO and the UN are right to convene.
  5. In creative and inventive fields AI is near a tipping point. The legal uncertainty about who owns AI output is itself a cost to business.
  6. In healthcare there is “almost a moral imperative” to bring AI in, and eventually to phase out human services where AI is much better.
  7. AI discrimination is a real risk but not new, and AI can be audited more easily than a human hiring manager.
  8. Small firms need an IP strategy advisor, not a pile of patents. Traditional and alternative medicine deserve proper integration. If AI does most work, society needs fair distribution (he names universal basic income) and fair access to longevity treatments.

How it closes. On longevity, the host raises 200 to 500 year lives. Abbott says the ethical worry is that “peter thiel will get to live to be 500 and everyone else is going to die in 80” and that access must be wide.

Stance. Optimistic, pragmatic, lawyerly. He answers dark scenarios from the host (data-driven selection of people, Cambridge Analytica, Nazi-era IBM machines) by saying the law already prohibits the underlying wrong and AI mostly makes it cheaper. He pushes back on the host’s “near-singularity” framing and keeps “strong AI” and “cyborg rights” in the future.

Note on the format. This is a host-led interview. The host often talks for minutes before asking. Abbott’s answers are the 🎙 material; the host’s claims are labelled as the host’s.

Map — talking points

#Point
1Host’s long introduction of Abbott
2Early life: family of lawyers, medicine first, then law, then life-science IP
3How AI came in: contract research firms say AI does drug discovery, so who is the inventor?
4The Reasonable Robot: tax, tort, IP, criminal law; same standard for person and machine
5Host: humanity, evolution, dystopia vs Star Trek, AI built under different national laws
6Regulating AI: it is a mix of law, ethics, norms, standards; old rules not built for AI; rights vs “legal entity”; strong AI and cyborgs in the future
7Host: political manipulation, Cambridge Analytica, Brexit, Orson Welles
8Social manipulation: platforms’ duty, role of government, AI as both problem and solution
9Host: governments lack understanding; three “governances”; China’s digital currency
10Global regulation: consultations (USPTO, UK IPO), harmonise data and IP, EU ethics lead, UN role
11Case studies: Siemens at WIPO; the Artificial Inventor Project (Thaler)
12Host: three interactions (human-human, human-machine, machine-machine); his own AI film work
13Creative industries: “creative singularity”, UK vs US copyright, AI music, AI actors
14Criminal liability for AI (Abbott and Sarch paper)
15Host’s health-data scenario: genetic and credit-score selection of people
16Healthcare and bias: anti-discrimination law exists; AI bias vs human bias; AI diagnosis; moral imperative
17Post-work future, universal basic income
18Legal tech: why law firms resist; Specifio, Ping, Iprova
19Startups and IP: get a strategic advisor; learn basics; WIPO small-business guides
20Alternative medicine: 2011 paper on medical students; evidence problem; botanicals; integration
21Longevity and ethics: access, “haves and have-nots”

Putting it together

Mental model. Abbott treats AI governance as a problem of consistency, not of new categories. The law already has rules for harmful behaviour, for discrimination, for liability, and for ownership. These rules were written for people (or companies). When a machine does the same thing, the gap between the two rule sets produces bad outcomes: machines get blocked from being owners or are over-penalised, while humans are under-penalised. Fix the gaps by aligning standards, and do not rush to give machines rights. Alongside this, he holds an optimistic view of the effects: AI can limit bias, help doctors, and make wealth plentiful, as long as society handles distribution and access. His method is to take each dark scenario from the host and show the existing legal wrong beneath it.

Open tensions.

  • Behaviour standard vs. development risks. A rule on conduct does not reach how systems are built.
  • “Same rule” vs. real difference. Machines scale, copy and fail differently from people; a single standard may not fit.
  • Platform duty vs. free speech. He assigns an obligation to platforms and a role to governments without saying who decides what is true.
  • Incentive logic. IP for machine output rewards investors, but cheap output may need no reward.
  • Moral imperative vs. validation. “Phase out human services” depends on proof across populations.
  • Abundance vs. concentration. UBI is mentioned as if it will arrive; the history in idea 9 says it needs a fight.
  • Evidence ladder. He asks for less evidence for low-risk traditional care than for drugs.

Reading path.

  • Start: abbott-reasonable-robot-ai-and-the-law for the framework; then abbott-patent-law-ai-inventors for inventorship.
  • Then: Abbott’s The Reasonable Robot (Cambridge UP, 2020); Lessig, Code; Benkler, Faris and Roberts, Network Propaganda; Bradford, The Brussels Effect.
  • Deeper: Calo (2015) on robotics and cyberlaw; Bryson, Diamantis and Grant (2017) on AI personhood; Obermeyer et al. (2019); Kleinberg et al. (2019); Acemoglu and Johnson, Power and Progress; Daniels, Just Health.

See also: abbott-reasonable-robot-ai-and-the-law · abbott-patent-law-ai-inventors

Part 1 — Before you read

Key terms

TermPlain meaningWatch out for
Behaviour-based standardLaw judges what was done, not what was thoughtCriminal law cares about mental state; Abbott says tort, tax and IP are mostly outcome-focused
Reasonable person standardThe test of what a typical careful person would have doneIn tort it asks about negligence. A product is judged by strict liability, which is a different test
Strict liabilityYou are liable for the harm whether or not you were carefulUsed for products in tort, and for a few crimes
Legal entity / legal personSomething the law can give duties and rights, like a companyA “legal person” is not a “human”; Abbott separates regulating AI like an entity from giving it rights because it deserves them
Strong AI / AGIAI that is aware it exists and has its own wantsAbbott says today’s AI is not this and calls strong AI “a whole different ballgame”
Creative singularityHis term for the point where AI becomes a main source of creative and inventive outputNot Kurzweil’s technological singularity; the host links the two
Freedom to operate (FTO)A search to check you are not infringing someone else’s patentDifferent from “can I get a patent?”
Computer-generated workUK copyright term for a work with no human authorThe US has no equivalent; see idea 4
Algorithmic biasA model treats groups differently because of its data or designHumans do this too; the question is which is easier to detect and fix
CAM / integrative medicineComplementary and alternative medicine; “integrative” means used alongside standard careWords differ between traditions, and “alternative” often means instead of standard care, which Abbott says he is not arguing for

People and things named (reading list)

  • Orson Welles / H.G. Wells (host): the 1938 radio broadcast of The War of the Worlds and the myth of mass panic.
  • Cambridge Analytica (host): a political data firm; the host says he worked with it for a while.
  • Dr Stephen Thaler and the DABUS system; the Artificial Inventor Project (artificialinventor.com). See abbott-patent-law-ai-inventors.
  • Alexander Sarch: co-author of “Punishing Artificial Intelligence: Legal Fiction or Science Fiction”. See abbott-reasonable-robot-ai-and-the-law.
  • WIPO (World Intellectual Property Organization): ran “conversations on AI and IP”; Abbott was in the first two.
  • Siemens: the company’s IP chief described a machine-designed part at WIPO.
  • OpenAI Jukebox: a 2020 model that made songs “in the sound of” famous singers.
  • Specifio, Ping, Iprova: legal-tech companies Abbott likes.
  • Yiviva (probably): a Yale spin-out testing a traditional Chinese formula.
  • Peter Thiel: the longevity example at the close.
  • Harari: Sapiens, 21 Lessons (host).
  • Ray Kurzweil: singularity forecast (host).

Abbott’s own frame

Abbott starts from one observation. AI is “stepping into the shoes of people” and “there are two different legal regimes that apply to behavior by a person and behavior by a machine”. Sometimes that has bad outcomes. His book tests this in four areas: tax, tort, IP and criminal law. The tort example: a self-driving car is held strictly liable as a product, a human driver only for negligence. “we don’t really want to have different liability standards for this because that discourages people from using machines even if they’re safer”. And he calls the aim “a principle of ai regulation that i think will result in all of us getting more benefit from ai and helping to minimize its risks”.

That is all this page needs of the framework. For depth, read abbott-reasonable-robot-ai-and-the-law. For the inventorship case, read abbott-patent-law-ai-inventors.

The current debate (as of 30 Oct 2026)

Added context, not Abbott’s view. The talk is five years old. These are the points where later events changed the picture. Searches were run on 30 Oct 2026; where sources conflict, I say so.

  1. Inventorship: Abbott’s test cases lost in the main courts. The US Federal Circuit held in Thaler v. Vidal (2022) that only natural persons can be inventors, and the US Supreme Court declined review in April 2023 (The Register). The UK Supreme Court unanimously held on 20 Dec 2023 that an inventor must be a natural person (IPWatchdog). Abbott’s remark that “as the law currently stands there’s just no protection for that sort of thing” is still the position in both countries. Some other jurisdictions went the other way (South Africa and, for a time, Australia), per the sources I found; details are in the sibling page.
  2. Copyright: the US “human authorship” rule stood. The D.C. Circuit affirmed in Thaler v. Perlmutter (2025) that copyright needs human authorship, and the Supreme Court denied review on 2 March 2026 (Mayer Brown; Baker Donelson). That confirms Abbott’s description of the US. The open question is how much human input counts when AI is a tool.
  3. UK copyright for machine-made work may go. The 1988 provision Abbott praises (the producer of a computer-generated work is its author) was put to consultation in Dec 2024 with removal as the preferred option. A March 2026 government report recommends removing it while keeping protection for AI-assisted works. It is a proposal; no bill has followed (HSF Kramer).
  4. The EU moved from guidelines to law, then delayed parts of it. Abbott calls the EU “one of the leaders” on ethical AI. The AI Act came after the talk. A “Digital Omnibus” then moved the high-risk deadlines. The sources I found agree on the new dates (standalone high-risk systems to 2 Dec 2027; those embedded in regulated products such as medical devices to 2 Aug 2028) but differ on the adoption steps and on the transparency rules (Sidley; CMS). Treat exact dates as to be checked against the Official Journal.
  5. Platforms: regulation arrived in the EU, retreated in the US. The EU fined X EUR 120 million in Dec 2025, the first non-compliance decision under the Digital Services Act (NBC News). In the US, Meta ended its third-party fact-checking programme in January 2025 and moved to Community Notes (Dawn). Abbott’s view that platforms “have an obligation” is the position of the EU path, not the US path.
  6. AI music: licensing, not a ruling. Abbott predicted that music would be “hugely disruptive” in five to ten years. The labels sued Suno and Udio in 2024. In Oct 2025 Universal settled with Udio and agreed to a licensed platform (Music Business Worldwide). I did not find the current state of the other cases.
  7. Medical AI: many devices, no replacement. The FDA list of AI-enabled devices passed 1,000 entries, about three quarters in radiology (FDA list; Applied Radiology). Radiologist pay and demand rose, which cuts against the old claim that radiologists would be obsolete (Fortune, May 2026; Fortune is a secondary source for the Medscape figure). This fits Abbott’s own hedge, “we are a very long way off from seeing an ai instead of a doctor”, but not yet his “five or ten years” of outperformance in non-trivial tasks.
  8. Yiviva’s drug. The formula (YIV-906) reported randomised phase 2b results in liver cancer in 2025 (trial record; company results were presented at ASCO and ESMO GI 2025 per Santelog). I did not check the later regulatory status.
  9. Not found. I found no source on the current status of Specifio, Ping or Iprova (acquisitions, closures). Treat Abbott’s endorsements as 2021 snapshots.

The ideas

*These are the traditions behind what Abbott and the host say. Abbott names almost none of them. Dates and titles are†

How the traditions meet

                      Abbott's lens:
          same behaviour -> same legal standard
                         |
   2.1 how to regulate   |   2.2 who counts as a "person"
   what you don't        |   (company, AI, human)
   understand            |
        \                |                 /
         \               v                /
  2.3 platforms,  --> REGULATION  <-- 2.4 treaties,
  propaganda,          OF AI              harmonisation,
  attention                               EU "Brussels effect"
         \                |                 /
          \               |                /
   2.5 IP incentives      |     2.6 medicine, AI, bias
   and machine authors    |     (standard of care, audit)
                          |
              2.7 tradition vs. evidence
                          |
              2.8 post-work, distribution,
                  who gets long life

Where each school of thought is covered:

  • Regulating a technology you do not yet understand: idea 1
  • Legal personhood: persons, things and companies: idea 1
  • Propaganda, attention and platform governance: idea 2
  • Global governance and IP harmonisation: idea 3
  • Why we give IP at all, and what machine authorship does to it: idea 4
  • Medicine, AI and fairness: idea 5
  • Evidence, tradition and integrative medicine: idea 8
  • Post-work, distribution and longevity: idea 9

1. Regulating AI development

In short. Abbott says AI needs regulating because people can do very bad things with it, as with every other technology. By regulation he means more than law: moral codes, cultural norms and industry standards count too. Rules written before AI already apply, and AI may sometimes need to be treated more like a legal entity, not because it deserves rights but because it has to be regulated.

🎙 In the talk.

  • The host asks whether AI will bring a dystopia or a “Star Trek” future, and notes AI built in the US, UK, Russia or China sits under different laws.
  • Abbott agrees AI needs regulating. The reason: “there is a potential for people to do very bad things with ai” and law restricts harmful behaviour in every other area. He also wants the benefits “spread somewhat fairly”.
  • He widens “regulation”: it is not only law but “people’s own moral and ethical codes its cultural norms its industry standards”. He gives a binding example (no machine gun on a drone).
  • Existing rules matter: “a body of regulations that wasn’t designed with ai in mind”. He compares AI to the internet, software and cars.
  • Culture varies. “you see a very different attitude towards things like data ownership or personal privacy in china versus the united kingdom”. Some of this should be harmonised, perhaps through WIPO or the European Commission.
  • On rights: “it isn’t that ai deserves rights because it is morally deserving but it is that sometimes ai needs to be regulated more like a legal entity”.
  • On the future: strong AI “is a whole different ballgame”. Cyborg rights are for “the future”, but “not too early to be talking about them”.

📚 Deeper theory.

  • Abbott’s list of constraints (law, ethics, norms, standards) is the Lessig modality idea (idea 1). It means regulators are not choosing between “law” and “nothing”.
  • The “old rules” point is the pacing problem. A common response is gap-filling by analogy (Abbott’s route) versus new statute (the EU AI Act route; see “current debate” item 4).
  • “Regulate like a legal entity but give no rights” is corporate personhood (Roots). It matters because it separates duties from moral status. A company has duties and some rights but no feelings.
  • The host’s three-way split (US, EU/UK, China) matches the standard comparison: market-led, rights-led, state-led. This is a broad simplification.
  • The “dystopia vs Star Trek” framing is a narrative device, not a legal one. Abbott does not take it up; he talks about specific risks.

📚 Roots: Regulating a technology you do not yet understand

The problem. Rules must be made before the effects are known, but once the technology is entrenched the rules are hard to change.

Chain.

  • David Collingridge, The Social Control of Technology (1980). The “control dilemma”: early on you cannot predict the harm, later you cannot easily change course.
  • Lawrence Lessig, Code (1999). Four modalities that constrain behaviour: law, social norms, markets, and architecture (code). Abbott lists “moral and ethical codes… cultural norms… industry standards”; this is Lessig’s family of constraints, though he does not cite it.
  • Ryan Calo, “Robotics and the Lessons of Cyberlaw” (2015). Asks which features of robots will force the law to change, not just stretch.
  • Roger Brownsword, Law, Technology and Society (2019). Argues regulators must watch “technological management”, where rules are built into the tech itself.
  • Gary Marchant et al., on the “pacing problem” (2011). Technology moves faster than law; “soft law” (guidelines, standards) fills the gap.
  • Adam Thierer, “permissionless innovation” vs the precautionary principle (Europe). The two poles of the argument.

Feeds the talk. Abbott’s claim that the existing body of rules “wasn’t designed with ai in mind” is the pacing problem. His choice to apply the old standard of reasonableness to machines is the stretch option, not the new law option.

📚 Roots: Legal personhood: persons, things and companies

The problem. Law has two boxes: persons (who hold rights and duties) and things (which are owned). Where does an autonomous system go?

Chain.

  • Roman law and the medieval universitas. Groups can be treated as one person.
  • Santa Clara County v. Southern Pacific (1886) and Salomon v Salomon (1897). Companies as persons, separate from owners. (The US headnote on Santa Clara is often misread; from memory.)
  • John Dewey, “The Historic Background of Corporate Legal Personality” (1926). Personhood is a legal tool, not a claim about souls.
  • Lawrence Solum, “Legal Personhood for Artificial Intelligences” (1992). Early test of whether AI could hold rights and duties.
  • European Parliament resolution on robotics (2017) floated “electronic personality”; an open letter of experts in 2018 opposed it. Later EU law did not take it up.
  • Joanna Bryson, Mihailis Diamantis and Thomas Grant, “Of, for, and by the people” (2017). Warn that AI personhood could be a liability shield for the people behind it.
  • Christopher Stone, “Should Trees Have Standing?” (1972). Rights for non-humans, as a legal device.

Feeds the talk. Abbott says AI should be regulated “more like a legal entity” but not given rights because it is “morally deserving”. That is the corporate-personhood route (function, not soul), and Bryson’s warning is the strongest argument against going further.

Where others push back.

  • Bryson and others warn that treating AI as an entity can let the humans behind it hide (Roots).
  • Some argue that “regulate behaviour, not the technology” misses risks from how models are built (training data, scale), which behaviour-based rules do not reach. This is the case for rules on AI development, which is the chapter’s title. Abbott answers mainly about use.
  • Industry voices argue against early rules (the “permissionless innovation” side in Roots).

What they leave out.

  • Who enforces, and with what budget. Neither speaker names a regulator.
  • Military AI beyond the one drone example.
  • Any concrete rule for development (testing, audits, licensing) despite the chapter label. The answer stays at the level of principle.

Check yourself.

  1. Why does Abbott say AI should be regulated “like a legal entity” without having rights?
  2. What are the four kinds of constraint in his list, and which one is not binding law?
  3. Why is “old rules not made for AI” a reason for both gap-filling and new law?

2. Social manipulation and platforms

In short. Abbott agrees that AI’s impact on opinion is strong today. Misinformation is old, but now there is more data on each person, and it moves faster and further. He says platforms have a duty to enforce limits on it and governments must play a role. AI is also part of the answer, since it can help moderate content.

🎙 In the talk.

  • The host’s setup is long. He says a large share of Americans believe a “parallel reality” about the 2020 election, links Brexit and Facebook “to manipulate opinion”, says he worked for a time with Cambridge Analytica, and says even simple retargeting is already a strong tool. He compares radio and Orson Welles and says Facebook, Google, Amazon and Alibaba hold more data than countries.
  • Abbott agrees on the urgency: “the impact of ai is very strong today”. He accepts the host’s point that misinformation is old. What is new: more data on each person, and speed and reach.
  • Duty: “these platforms have an obligation to be enforcing you know restrictions on people disseminating misinformation” and governments must “play a role”. His main example is COVID-vaccine misinformation, which “will cost lives”.
  • He says regulators are “coming around” to the view that you cannot leave online speech wholly unregulated, and he says Facebook has not wanted to act because “the way it operates right now is very profitable”.
  • AI cuts both ways: “ai you know is not only part of the problem in having you know even simple ai promoting misinformation but also part of the solution”. Human moderators suffer from child abuse and violent material, so there is “a bit of an ai arms race here to be had”.

📚 Deeper theory.

  • The history the host uses (Welles) is itself contested (Roots). The better-supported claim is the structural one: platforms optimise for engagement, and targeting is cheap.
  • The causal claim “Facebook changed Brexit/US votes” is weaker than the host implies. Benkler’s group (Roots) finds a media-ecosystem effect that predates the platforms. Abbott never asserts the strong causal claim; the host does.
  • A live legal split: the US First Amendment and Section 230 make state-ordered takedowns hard; the EU DSA uses systemic duties (risk assessments) instead of case-by-case speech rules. Abbott’s “governments have a role” is easier to implement in the second model.
  • “Platforms have an obligation” has two readings: (a) a private duty to moderate; (b) a legal duty set by the state. He leans on (b) with industry help.
  • AI as moderator raises its own problems: error rates, language gaps, and moderators in the Global South†.

Added context (2026). The EU fined X EUR 120 million under the DSA in Dec 2025 (NBC News). Meta dropped US third-party fact-checking in Jan 2025 (Dawn). So the direction Abbott predicted held in Europe and reversed in the US platform’s own policy.

📚 Roots: Propaganda, attention and platform governance

The problem. Who is responsible for what spreads in a public sphere owned by private companies?

Chain.

  • Walter Lippmann, Public Opinion (1922) and Edward Bernays, Propaganda (1928). Opinion can be manufactured.
  • Hadley Cantril, The Invasion from Mars (1940). The study behind the Welles story the host cites. Later historians (such as W. Joseph Campbell, Getting It Wrong) argue the panic was exaggerated (NPR).
  • Jacques Ellul, Propaganda (1962). Modern propaganda works through technique and saturation, not lies alone.
  • Section 230 of the US Communications Decency Act (1996). Platforms are not publishers of user content. This is the legal background for the US debate.
  • Eli Pariser, The Filter Bubble (2011) and Shoshana Zuboff, The Age of Surveillance Capitalism (2019). Targeting and behaviour prediction as a business model.
  • Cambridge Analytica (2018). Harvested Facebook data used for political targeting. The effect on votes is contested; the UK regulator’s later report found less than the headlines suggested†.
  • Yochai Benkler, Robert Faris, Hal Roberts, Network Propaganda (2018). Finds the US problem was driven more by an asymmetric media ecosystem than by platform algorithms alone. This is the main academic pushback on the host’s “Facebook did it” story.
  • EU Digital Services Act (2022). Risk-based duties on very large platforms.

Feeds the talk. The host’s long claim is that data and targeting changed politics. Abbott accepts platform responsibility but never says how much of the effect is proven.

Where others push back.

  • Free-speech scholars: who decides what counts as “misinformation”? In the COVID case the official line changed (for example on early mask guidance), which makes state-set truth risky.
  • Empirical researchers: exposure to false news is concentrated in a small group of heavy users, so broad rules may be a poor tool.
  • Platform defenders: community-based correction (Community Notes) scales better than expert fact-checking.

What they leave out.

  • Generative AI. In 2021 the problem is simple targeting and cheap text; the cost of fake images, voices and video had not collapsed yet. The talk cannot cover that.
  • The ad business model as the cause. Abbott mentions profit but not what rule would change it.
  • Any statement on who is liable when a platform’s own recommender spreads the false item.

Check yourself.

  1. Which parts of the Brexit/Cambridge Analytica story are the host’s claims and which does Abbott endorse?
  2. What two things does Abbott say are new compared to older fake news?
  3. In what way does he say AI is “part of the solution”?

3. Global regulation, policy education and harmonisation

In short. Abbott says policymakers who are not technical need educating, and he calls public consultation by the US and UK patent offices the right place to start. Data and AI are global, so countries should at least talk to each other and agree technical standards. A system that protects some activities in some countries but not others is often a problem.

🎙 In the talk.

  • The host says governments show an “alarming lack of understanding”, that control of tech sits mainly in patents, and that China moved first on a central bank digital currency.
  • Abbott agrees that educating non-technical policymakers matters. He says patent offices in the US and UK have “recently announced calls for consultation” with “dozens of questions” on AI and IP.
  • “that sort of public engagement process is the right place to start” and “no one working in silos understands the whole picture”.
  • Because data and AI are “global”, countries should talk: “at a minimum country should be speaking with each other”. He wants technical standards for interoperability, and warns of a system with “protection for some activities in some countries but not others which may often be problematic”.
  • Ethics: “i think the eu is one of the leaders in really ensuring ethical uses of ai”, including non-discrimination and health privacy. He says the UN system is “right to be taking a lead”.

📚 Deeper theory.

  • Consultations are a standard soft-law step. They gather evidence before statute and let firms, doctors and inventors speak. The weakness is that well-funded groups answer most.
  • IP is the most harmonised area of law because treaties date from the 1880s (Roots). That is why Abbott’s team could file the same inventions in a dozen offices and see different outcomes: harmonised form, divergent substance on who counts as an inventor.
  • The Brussels effect (Roots) explains why “EU leads” matters beyond the EU. It also explains why other powers resist: a rule written in Brussels becomes de facto global.
  • The host’s point that “control is on patents” is partly right. Patents matter in pharma and hardware; for AI models, trade secrets, compute and data access matter more†.
  • Interoperability standards (ISO, IEEE) are a different tool from treaties. They are often the faster route.

Added context (2026). The AI Act’s high-risk deadlines were pushed back by the Digital Omnibus, with different sources giving different adoption steps (Sidley). That is a sign that “EU leads” and “EU slows” can both be true.

📚 Roots: Global governance and IP harmonisation

The problem. Ideas and data cross borders; laws do not.

Chain.

  • Paris Convention (1883) and Berne Convention (1886). The base of international patent and copyright law. Treaties share rules by “national treatment”.
  • WIPO (created 1967; UN agency since 1974) and TRIPS (1994, WTO). Minimum standards for IP in trade. Abbott works with both.
  • Patent Cooperation Treaty (1970). One filing, many countries. Abbott’s team filed in about a dozen places.
  • Anu Bradford, The Brussels Effect (2020). EU rules spread because firms standardise on the strictest market. This is the mechanism behind “the EU is one of the leaders”.
  • OECD AI Principles (2019) and UNESCO Recommendation on the Ethics of AI (2021). Soft-law attempts at shared norms.
  • WIPO Conversation on IP and AI (from 2019). The forum Abbott describes.

Feeds the talk. Abbott’s line that countries “should be speaking with each other” is the harmonisation tradition. His warning of “protection for some activities in some countries but not others” is the very problem treaties like the PCT try to solve.

Where others push back.

  • Critics say harmonisation tends to favour the strongest IP holders (the TRIPS critique).
  • Others say regulatory competition is healthy and one global rule is not needed.
  • On the host’s claim that China is “10 years” ahead: the e-CNY was still limited in 2021;†

What they leave out.

  • Which countries are not at the table (most of the Global South), and what they would want.
  • Enforcement across borders.
  • Data protection law by name (GDPR), though he points at it.

Check yourself.

  1. What does Abbott say a consultation is good for?
  2. Why does he say a split in protection between countries is “problematic”?
  3. What is the difference between a treaty and an interoperability standard?

4. AI in creative industries and ownership

In short. Abbott describes a “creative singularity” in which AI makes art, music and films. Who owns the output is tangled: the AI maker, the person who gave the input and the owners of the training data may all have a claim. The UK and US treat machine-made work differently, and that uncertainty is bad for business. He expects AI music to be a hugely disruptive commercial force within five or ten years.

🎙 In the talk.

  • The host describes his own films made with AI and image banks, says some films are “created completely by ai” in “already very processed” industries, and worries small start-ups can be “killed” by big patent holders. He also cites a figure of about one million actors.
  • Abbott calls the shift “the creative singularity”, from one of his articles. Patents are harder than copyright, which has a very low bar.
  • For a business, ownership is tangled: who made the AI, who gave the input, whose data was used, and “many people can own that data” so “using data like that can be infringement”.
  • UK vs US: in the UK “if you have a machine make something without a traditional human author the producer of the work owns the copyright and is legally deemed the author”; in the US “if you don’t have a traditional human author no one owns the thing”. He adds a hedge on the date.
  • AI music: songs “in the sound of frank sinatra or taylor swift” are “still a little wonky” but not bad. He says “this is really going to be a hugely disruptive commercial thing in five or ten years”.
  • Open questions: where the data came from, who “pushes the button”, whether the programmer knew the output. “uncertainty like that is bad for commercial enterprises”.
  • Actors: “these million actors are pretty soon going to be competing with ai actors”, maybe of dead stars. Many firms haven’t really given enough thought to this.

📚 Deeper theory.

  • The UK s 9(3) rule (Roots) treats the arranger as the author. The logic is practical: protect the investor so the market works. The US takes the opposite path: copyright is for human expression.
  • The US position rests on the 19th-century case law (Burrow-Giles) and Copyright Office practice. The Naruto monkey-selfie cases are the animal analogy.
  • Training-data questions are separate from ownership-of-output questions. Abbott mixes them in one list because the business faces both. Courts treat them separately: one is infringement by the input, the other authorship of the output.
  • Copyright’s low originality bar vs patents’ high bar is why art and music will reach the courts first.
  • Actors and “digital replicas” are mostly a publicity rights and contract question, not copyright. The legal tools differ by US state and country†.
  • Economic effect: if output is cheap, prices fall and the share going to human creators may fall too. The incentive logic in Roots may need rethinking for humans, not just machines.

Added context (2026).

  • US: the human-authorship rule was confirmed by the D.C. Circuit and the Supreme Court let it stand in March 2026 (Mayer Brown).
  • UK: the 1988 rule Abbott cites is now proposed for removal (HSF Kramer). The contrast he draws may disappear.
  • Music: Universal settled with Udio in Oct 2025 and moved to a licensed model (MBW). That is an example of the “who gets paid for the data” question being solved by contract, not by a court ruling.
  • Actors: I did not find a source on AI actors in my searches; the 2023 US actors’ strike and later digital-replica rules are†

📚 Roots: Why we give IP at all, and what machine authorship does to it

The problem. If IP is a bribe to make people create, what happens when the creator needs no bribe?

Chain.

  • Statute of Anne (1710) and the US Constitution’s IP clause (1787). Incentive to promote “progress”.
  • Landes and Posner, The Economic Structure of Intellectual Property Law (2003). The incentive/access trade-off.
  • Burrow-Giles v. Sarony (1884) and Feist v. Rural (1991). US copyright needs a human author and a modicum of originality. Abbott says copyright’s bar “is incredibly low” compared with patents.
  • The US Copyright Office line that works need human authorship (the Naruto v. Slater “monkey selfie” decisions, 2018, are the animal parallel).
  • UK CDPA 1988, s 9(3). The “person by whom the arrangements necessary for the creation of the work are undertaken” is the author. Abbott cites this.
  • James Boyle, The Public Domain (2008) and critics of over-propertisation. The worry that more IP blocks later creators.
  • Mark Lemley, “Faith-Based Intellectual Property” (2015) and Lemley and Casey, “Fair Learning” (2021). Argue that training on works should often be fair use. A live dispute behind the host’s images from data banks point.

Feeds the talk. Abbott’s whole case for machine inventorship is incentive theory: protect the investor in the machine so they build machines that invent. The critique in idea 4 is that the incentive may be unneeded if machine output is cheap.

Where others push back.

  • Some argue an “arranger as author” rule gives monopolies over cheap output and should be dropped (the direction the UK report leans).
  • Artists’ groups argue that training without consent is the first harm, and that the ownership debate hides it.
  • Others say AI music will not displace live performers, who sell presence.

What they leave out.

  • Moral rights, attribution and consent of the people whose work trained the model.
  • Who benefits when the “producer” is a platform rather than an artist.
  • Whether low-cost AI output will raise the value of human-made work.

Check yourself.

  1. Why does Abbott say patents are harder to get than copyright for AI output?
  2. State the UK and US positions on a work with no human author as he gives them.
  3. What are the three or four separate legal questions packed into “who owns the Taylor Swift-style song”?

5. Healthcare, bias and the moral imperative to automate

In short. Abbott calls the host’s health-data scenario chilling, but says the core harm is discrimination, which AI only makes easier and which law already bars in part. Human hiring has always been biased, and AI could make bias easier to see and to stop. He thinks AI will outperform doctors at some tasks in five or ten years, and that if a robot is better 99 percent of the time we almost have a moral imperative to use it.

🎙 In the talk.

  • The host raises a scenario: take 10 or 20 million people from a social network, add their health data, segment by risk, and then decide who to “take out of society”. He cites Nazi-era IBM machines.
  • Abbott calls it “chilling” but says the core is discrimination on health information, which “ai is simply enabling it to be much easier to do”. Law already bars much of it: “you can’t not hire someone because of their cancer risk”. Credit scores are a less clear line. Regulation could limit what social networks collect.
  • Bias: algorithmic hiring tools do behave badly, but “human resource managers have been making bias decisions on human cvs” for as long as there have been CVs.
  • The turn: “ai also has the potential to limit human discrimination and make it more difficult”. With an AI “it’s easier to see if they’re doing it and to just make rules to keep it from doing that”.
  • Clinical AI: some companies claim AI beats a doctor at narrow tasks (skin lesions, pathology slides, lung imaging). “we are a very long way off from seeing an ai instead of a doctor”, but in five or ten years he expects AI “outperforming human doctors” in non-trivial things.
  • Surgery: the da Vinci robot “is really just a sophisticated scalpel” today. In 20 years it may remove a cancer better than a surgeon. If the surgeon says the robot is better “99 percent” of the time, “we almost have a moral imperative in some sense to integrate ai into things like patient care and in fact at some point probably to be transitioning out human services”.
  • He sees this as positive: if AI does everything, “we would have as a society really an unlimited amount of wealth and people wouldn’t have to work”, with “we’d have to have a universal basic income”. The worry is “making sure that we get there efficiently in a way that isn’t leaving people behind”.

📚 Deeper theory.

  • The standard of care in malpractice is judged by custom, but T.J. Hooper (Roots) says custom can itself be negligent. Once AI is much safer, the law may require its use. That is the legal route to Abbott’s “moral imperative”.
  • Abbott’s symmetry (same rule for human and machine) implies a ratchet: as machines get safer, the human standard must be compared to the machine standard. This is the book’s tort argument (abbott-reasonable-robot-ai-and-the-law).
  • “AI is easier to audit than a person” is real but conditional. It needs access to the model, to training data and to outcome data. Many hiring tools are sold as black boxes. Obermeyer’s 2019 study (Roots) shows audits are possible but only with data.
  • Different fairness definitions can conflict (Roots). “Make rules to keep it from doing that” requires choosing which fairness rule, and that is a political choice.
  • Narrow-task performance (reading a scan) is not the same as the whole job (talking to the patient, deciding with uncertainty). Radiologists and surgeons do more than read images.
  • Transition: “transitioning out human services” raises training, skill decay and accountability questions. If humans stop doing a task, who checks the machine? (From memory; the “automation complacency” literature.)

Added context (2026).

  • The FDA’s AI-device list is above 1,000 entries, about three quarters radiology (FDA; Applied Radiology).
  • Radiology demand and pay rose through 2025, so AI has not replaced radiologists (Fortune). Abbott’s own caution holds; his timeline for “outperforming doctors in non-trivial things” has partly arrived on narrow tasks and not on whole jobs.
  • EU AI Act high-risk duties for AI in medical devices were moved to 2 Aug 2028 (Sidley; sources differ).

📚 Roots: Medicine, AI and fairness

The problem. When a model beats a clinician, who is responsible, and does it treat people fairly?

Chain.

  • Paul Meehl, Clinical versus Statistical Prediction (1954). Simple rules often beat expert judgement. The old root of “AI beats the doctor”.
  • The T.J. Hooper case (1932, Learned Hand). Industry custom does not decide the standard of care; a whole industry can be negligent in not adopting a cheap safety tool. This is the legal ancestor of Abbott’s “moral imperative”.
  • Esteva et al., skin cancer classification (Nature, 2017) and Topol, Deep Medicine (2019). The evidence that models can match specialists at narrow tasks; Topol argues for a “doctor plus AI” future.
  • Geoffrey Hinton’s 2016 remark that radiologists should stop training. Often cited as a failed forecast (see the 2026 context above).
  • Barocas and Selbst, “Big Data’s Disparate Impact” (2016) and ProPublica on COMPAS (2016). How bias enters models; and Kleinberg et al. (2016), showing that fairness definitions can conflict mathematically.
  • Obermeyer et al., Science (2019). A US health algorithm under-served Black patients because it used cost as a proxy for need.
  • Kleinberg, Ludwig, Mullainathan, Sunstein, “Discrimination in the Age of Algorithms” (2019). Argue algorithms are easier to audit than people, which is Abbott’s point.
  • GINA (US, 2008). Bars genetic discrimination in health insurance and employment. This is the sort of law Abbott points to.

Feeds the talk. The host’s scenario of selecting people by DNA and credit is an old discrimination problem with a new tool. Abbott’s reply is: law exists, enforcement is the issue, and algorithms leave a record.

Where others push back.

  • Clinicians and ethicists: “moral imperative” can become coercion to adopt tools before they are validated across populations.
  • Equity researchers: models trained on one population can fail on another.
  • Economists: the UBI/abundance picture assumes distribution happens; history shows gains accrue to owners without policy (idea 9).
  • Patient groups: some want a human in the loop for dignity, not only for outcome.

What they leave out.

  • Liability when an AI-assisted decision fails (a topic of his book; here only implied).
  • Data consent and health-data ownership, though the host raises it.
  • Who pays for the AI and who is excluded.
  • Medical-device regulation (how a tool is approved), which is the real gate.

Check yourself.

  1. Why does Abbott say the host’s health-data scenario is an old problem?
  2. What two conditions make “AI is easier to audit than a hiring manager” true?
  3. How does the T.J. Hooper idea support his “moral imperative”?

In short. Abbott says law, like medicine, is not naturally eager to adopt technology. The hourly billing model gives firms little reason to disrupt themselves. He points to tools that already work: one that drafts a patent application, one that tracks time automatically, and one that helps inventors think across fields.

🎙 In the talk.

  • The host asks about legal tech, RegTech, smart contracts replacing lawyers’ work, and platforms people can use now.
  • Abbott says law, like medicine, is “not naturally ambitious about bringing on technology”. Some firms are forward-looking. The resistance is tied to business models, “the way that firms bill hourly”. They may lack incentives to disrupt their own model “until that gets forced upon some of them”.
  • Specifio: give it a patent claim and it drafts an application; it needs attorney review, but “it’s something that could take writing a patent application from 15 hours down to five”.
  • Ping: automated timekeeping. Lawyers dislike recording hours, clients get accuracy, and “lawyers tend to under bill their hours” so billings go up.
  • Iprova: uses AI to help inventors “invent in ways they wouldn’t have thought of before” by bringing in other fields and prior art.
  • The host says the point is to talk about good things to get out of “paranoia”.

📚 Deeper theory.

  • The billable hour creates a conflict: efficiency cuts revenue. Richard and Daniel Susskind, The Future of the Professions (2015), make this argument for all professions†.
  • The three examples are three different uses: drafting (generates a document), admin (records time), invention support (widens the search). Only the third is close to Abbott’s research topic of machine invention.
  • Specifio-style drafting links to a patent-law point: a patent application must “enable” the invention, and a model that writes the text may add detail the inventor never had. Review matters†.
  • Time-recording tools affect pricing: if time is captured more fully, hours billed rise, but clients may push to move to fixed fees.
  • The host’s “lawyers replaced by code” idea (smart contracts) is a separate topic. Abbott answers with firm-level tools instead of contracts-as-code.

Added context (2026). I found no source on these three companies’ status in 2026, so treat the names as 2021 examples;† I did not research the later rise of general-purpose AI drafting tools, so I make no claim about them.

Where others push back.

  • Clients now ask firms to price by outcome rather than by hour, which makes the hourly-billing explanation only part of the story.
  • Professional-conduct rules (competence, confidentiality, unauthorised practice) also slow adoption.

What they leave out.

  • Courts’ own use of AI and the risk of false citations.
  • Access to justice: the best case for legal tech is serving people who cannot afford a lawyer.
  • Data privacy when client files go to vendors.

Check yourself.

  1. Why does Abbott think hourly billing slows legal-tech adoption?
  2. What is the difference between Specifio and Iprova as uses of AI?
  3. Why is attorney review still needed for an AI-drafted patent application?

7. Startups and IP strategy

In short. Abbott says IP is expensive and big firms have in-house teams, while small firms need an advisor who knows what is critical to the business. His advice is to learn a basic amount of IP, use the WIPO guides for small businesses, then hire a lawyer for specific advice. He sums it up as an ounce of prevention being worth a pound of cure.

🎙 In the talk.

  • The host says there are 430 million micro-SMEs and millions of startups, and fears a big firm’s patent can kill a small one.
  • Abbott: IP is “expensive”, and big companies have in-house teams. Small firms need “an advisor who understands strategically what is really critical to the business”. The questions: “does your business need a patent why does it need a patent”, do you need a trademark, what are the risks?
  • Small firms cannot file widely, and “it is also very difficult for small businesses to do freedom to operate analyses”.
  • His advice: learn “a basic amount” of IP, use WIPO’s small-business division guides (they differ by industry), then hire a lawyer or strategic advisor for specific advice. “an ounce of prevention is worth a pound of cure”.

📚 Deeper theory.

  • Startups use patents for three different reasons: to block copying, to raise money (signalling to investors), and to trade in a dispute. Many do not need the first. Abbott’s “does your business need a patent” starts from this.
  • Freedom-to-operate work is expensive because patent claims are hard to read and search. Firms choose between a search, a licence, designing around, or accepting the risk†.
  • Trade secrets and speed to market can be better than patents for software and AI. They need no filing but need internal controls.
  • Copyright protects code and expression; patents protect functions. Software patents are limited in the US and Europe†.
  • Patent trolls and cross-licensing: large firms stockpile patents mainly for defence and bargaining. A small firm cannot join that game. Defensive options include open licensing networks†.
  • Costs vary by field. Medical devices and biotech depend on patents; software and art much less. That is why WIPO’s guides are sector-specific.

Where others push back.

  • Some founders argue the real risk is not patents but running out of money; IP spend should be small and late.
  • Studies find mixed evidence that early patents help start-ups raise money†.

What they leave out.

  • AI-specific startup issues: ownership of model weights, training-data licences, and open-source licences.
  • Employee and contractor IP assignments, a common cause of disputes.
  • Cost figures: he gives none.

Check yourself.

  1. Which three questions does Abbott say a startup should ask about a patent?
  2. What is a freedom-to-operate analysis and why is it hard for a small firm?
  3. Why do the useful IP facts differ between medical devices and software?

8. Alternative medicine

In short. Abbott calls alternative medicine low-tech and high-touch. Many practices have been used safely for a very long time, but the costly trials that prove effect only happen when a company can profit through IP. So for many therapies the evidence bar will never be met, yet he thinks the safe and cheap ones belong in conventional care, as an addition to it and not a replacement.

🎙 In the talk.

  • The host raises Abbott’s 2011 paper on medical students’ attitudes to complementary, alternative and integrative medicine and says massage helps his own circulation.
  • Abbott calls it “the opposite of you know the high-tech area it’s low-tech high-touch”. Many practices have been used safely by many people “since time immemorial”.
  • The evidence problem: a COVID-scale trial costs billions, and that only happens when a company will profit through IP and high prices. So for many therapies “we’re never going to have that threshold of evidence”.
  • Still, they are safe, cheap and used for a long time, and patient-reported outcomes are good, so “there is really room for many of these therapies to be brought into conventional medicine”.
  • Limits: “if you got hit by a bus i wouldn’t go to an acupuncturist”, and for cancer, “i would be getting acupuncture but i wouldn’t be doing it instead of chemotherapy”, only for energy and side effects.
  • Botanicals: many drugs come from herbs. Two US prescription botanicals: a green-tea extract for genital warts and a tree sap for diarrhoea in HIV/AIDS patients. A “yale spinout” (caption: “eviva”) has a traditional Chinese formula in phase 2 as a chemotherapy adjuvant; it is “their own proprietary version”.
  • The host adds a quote from a “professor larry”: traditional medicine is “our cultural historical operating system” (speaker unresolved).

📚 Deeper theory.

  • The IP point is the sharpest idea here. A treatment that cannot be patented gets little trial funding. Old remedies cannot be patented as such, so they stay untested. A company can patent a standardised extract or a new use, which is why Yiviva’s version is “a different drug”.
  • This ties to the Tu Youyou case (Roots), where state funding, not private IP, paid for the research.
  • The botanical-drug pathway (Roots) lets a mixture be approved without isolating one compound, but the sponsor must show consistent batches and a clinical effect.
  • “Evidence” has layers. Pragmatic trials (compare real-world outcomes) and patient-reported measures are weaker than double-blind RCTs but cheaper. Abbott accepts the weaker layer when risk is low.
  • The critique (Ernst, Singh) is that low-risk is not the same as zero-risk: harm can come from delay, interactions, or false hope. Abbott’s bus and chemotherapy lines are his own guardrail.
  • Placebo and ritual may have real effects on pain and nausea; whether that is a reason to use a treatment is an ethics question, not a data one.
  • Acupuncture evidence is mixed: effects over no treatment are clear, over sham small (Roots)†.

Added context (2026). Yiviva’s YIV-906 reported phase 2b liver-cancer results in 2025 (trial record; Santelog). The 2021 talk described it as a chemotherapy adjuvant in phase 2. I read these as the same compound; the match is not confirmed.

📚 Roots: Evidence, tradition and integrative medicine

The problem. How should medicine treat practices that are old, cheap, popular and hard to trial?

Chain.

  • Evidence-based medicine (Sackett, Guyatt, early 1990s). Randomised trials sit high in the hierarchy.
  • NCCAM (US, 1998), renamed NCCIH in 2014. A US government centre for complementary and integrative health.
  • Tu Youyou and artemisinin (Nobel, 2015). The best known case of a traditional Chinese remedy becoming a modern drug. (Abbott does not cite it.)
  • FDA Botanical Drug Guidance (2004; final 2016). A route for plant mixtures as prescription drugs. Abbott’s two examples came through it.
  • Edzard Ernst and Simon Singh, Trick or Treatment? (2008) and the “science-based medicine” movement. The main critique: integration gives unproven care a halo.
  • Vickers et al., acupuncture meta-analysis (2012). Found acupuncture beats no treatment, and that the gap to sham is small, which is the dispute in one finding. (Details†)

Feeds the talk. Abbott’s position is a middle one: use proven care first, add cheap safe traditional care for side effects, and accept that an RCT may never be paid for if no one can patent the result.

Where others push back.

  • Science-based-medicine writers say integration lends legitimacy to practices that fail good trials (Roots).
  • Traditional-medicine advocates say RCTs fit single drugs, not whole systems, and misjudge them.
  • Regulators worry about quality control in herbal products.

What they leave out.

  • India’s AYUSH systems, which are state-backed and might be a major example (the talk is US/UK-framed; from memory).
  • How an “integrated” service would be paid for and licensed.
  • Any specific condition-by-condition list of what is supported.

Check yourself.

  1. Why, in Abbott’s account, do cheap traditional therapies lack RCTs?
  2. What are his two examples of safe use alongside standard care?
  3. Why is the Yiviva formula “a different drug” from the traditional one?

9. Longevity, distribution and who gets the future-end

In short. Abbott does not think living two to five hundred years sounds too long. His main worry is fairness: one billionaire living to 500 while everyone else dies at 80. He says that when these technologies exist there is an ethical and moral obligation to make them widely available and not create further disparities.

🎙 In the talk.

  • The host cites life-expectancy growth, a large investment by Thiel in life extension, and the idea that our children’s children may live 200 to 500 years.
  • Abbott: “two to five hundred years doesn’t seem too long for me to live”. He doubts resources are the limit, since longer-lived, productive people could add to resources, and it is “a good reason for us to kind of increase our stewardship of our natural environment”.
  • His main ethical worry: “peter thiel will get to live to be 500 and everyone else is going to die in 80”. His answer: “there is an ethical and a moral obligation when these technologies exist to make them widely available and not to create further disparities”.
  • This echoes his earlier line on AI wealth: a fair system so it is not “jeff bezos owning you know a trillion trillion dollars”.

📚 Deeper theory.

  • The distribution argument is Rawlsian (Roots): a technology is justified if the worst off also gain. Daniels’s “just health” frames it as fair equality of opportunity.
  • “Resources are not the limit” is a contested empirical claim. Population, energy and ecological limits are debated†.
  • Social effects of long lives: pensions, retirement age, inheritance, political turnover, and entrenched wealth. Many economists argue the effect is on the age structure of institutions, not only on fairness of access.
  • Cost curves: some medical technologies start expensive then fall in price; others (specialty drugs) stay priced for profit. The IP system (his field) is one cause. He links IP to access at the start of the talk with the vaccine-in-Africa example.
  • The host’s life-span numbers are loose (see slips); the larger forecast is speculation.

📚 Roots: Post-work, distribution and longevity

The problem. If machines do most work, who gets the output? If life gets longer, who gets the treatment?

Chain.

  • John Maynard Keynes, “Economic Possibilities for our Grandchildren” (1930). A 15-hour week by 2030. Wrong on hours, right on wealth.
  • John Rawls, A Theory of Justice (1971). Fairness as what you would pick behind a “veil of ignorance”. The base of most arguments that new health technology should reach the worst off.
  • Philippe Van Parijs, Real Freedom for All (1995) and Guy Standing. The modern case for a basic income.
  • Frey and Osborne (2013) on job automation; Daron Acemoglu and Simon Johnson, Power and Progress (2023) argue the gains go to owners unless policy and technology choices change.
  • Norman Daniels, Just Health (2008) and Nick Bostrom’s work on enhancement and longevity. Debate over whether life extension is a medical good or a status good.

Feeds the talk. Abbott’s “unlimited amount of wealth” with a “universal basic income” is the Keynes/Van Parijs line. His point about Thiel is a Rawlsian access worry.

Where others push back.

  • Some argue life-extension is a low-priority use of research money compared with existing disease burdens.
  • Others argue a “moral obligation” to share is easier said than enforced; patents and pricing exist in the same system Abbott works within.
  • Some ethicists (such as Leon Kass) argue limits on lifespan are part of what gives life meaning†.

What they leave out.

  • Any mechanism for “making them widely available” (compulsory licences, public funding, procurement).
  • The overlap with his earlier claim that AI will concentrate wealth; the two ideas are linked but he does not tie them together.
  • A date or evidence for the host’s forecasts.

Check yourself.

  1. What is Abbott’s “ethical concern that seems most pressing” about life extension?
  2. How does it connect to his remarks on universal basic income?
  3. What does a patent system have to do with access, as he sketches it at the start of the talk?

10. Two shorter threads (see siblings)

In short. Two shorter threads. On inventorship, Abbott says engineers who saw an AI-designed part would not call themselves its inventor, and current law protects it only if a human is named. On crime, he opposes holding AI criminally liable: it would be a huge change to the legal system and is not needed.

Inventorship and the Siemens story. Abbott describes a Siemens engineer group who looked at an optimal AI-designed part and “said well i didn’t invent it this just came out of the machine”. Offices in the UK and Europe rejected the Thaler applications because they had no human inventor. He says the law has “no protection for that sort of thing unless you lie about it and put a human being”. Later rulings are summarised above in “current debate” items 1 and 2; the full treatment is in abbott-patent-law-ai-inventors.

Criminal liability for AI. With Sarch he asks whether AI could ever be liable for a crime. A tool used by a person is the person’s crime. Abbott says “we’re not in favor of holding ai criminally liable because it would be a huge change to the legal system and there’s really not a need for it”, but notes companies are already criminally liable and some crimes need no bad mindset. The full treatment is in abbott-reasonable-robot-ai-and-the-law.