Global AI Debates · Fall 2026 · Resources
Choose your topic and age division. Understand the motion, explore both sides, and use the sources to develop your own argument.
Primary School Speech · Ages 7–10 · Fall 2026
Explore whether developing self-driving cars could make travel safer and fairer. Read arguments on both sides, examine real evidence, and plan your own five-minute speech.
“We support the development of self-driving cars.”
Read with a parent or teacher where helpful. Choose your own position and explain your reasons.
01 / Research and preparation
Imagine a car taking someone to the library without a person steering it. Could that help people? What would we need to check before trusting it? Read the reasons on both sides, then choose the reasons you think are strongest. Use them to plan your own speech. You do not have to agree with this guide.
A self-driving car uses sensors and a computer to notice what is around it and control its movement. Some systems can do the driving only in certain places and conditions. A car that helps a human steer or brake can still need that person to watch the road. These are different kinds of technology. [C1]
Development means designing, building, and testing the technology. You can support research while asking for strict safety rules. You do not have to support every road test. If you oppose development, explain why the risks or costs are too great, or why another use of the money matters more.
Read C1 with an adult to understand the technology. Use C2 to explore safety evidence and C3 to think about who could benefit. The source guide at the end explains each link. Sources were checked on 24 October 2026.
02 / Research and preparation
These are possible arguments to investigate. The conclusion is a judgement you must defend; a source does not make the decision for you.
A carefully designed system might avoid some mistakes made by tired or distracted drivers. A 2026 study by the Insurance Institute for Highway Safety compared driverless Waymo cars with human drivers in four US cities. It counted crashes per mile that people would usually report to the police. In that comparison, the Waymo cars had 68% fewer crashes. This is encouraging evidence for one service in the conditions studied. It does not show that all self-driving cars are safer everywhere. [C2]
An opponent might ask whether those results apply near your school or in different weather. A strong reply would support further testing and require evidence for each setting, rather than promise that crashes will disappear.
A person who cannot drive might gain another way to reach work, friends, or a doctor. The US Department of Transportation has supported designs to help disabled passengers find, enter, and use automated vehicles. [C3] The argument is that development could expand freedom if vehicles are accessible and affordable.
The challenge is that removing the driver can also remove someone who helps a passenger. Supporters should explain how that help would still be provided.
We cannot learn everything about a new technology without investigating it. Supporters can argue for staged development: controlled tests, published results, checks by people outside the company, and clear stopping rules. This is a proposed approach, not proof that every test on public roads is acceptable.
The hard question is who bears the risk while others learn. Explain how pedestrians and local communities would be protected and heard.
03 / Research and preparation
A pedestrian shares the road with an experimental vehicle without booking a ride. An official investigation into a fatal 2018 Uber test-car crash identified failures involving the operator, system design, and company safety practices. [C4] This shows why safety depends on how development is managed. It does not establish the performance of every current system.
A supporter can reply that better safeguards and new designs can address past failures. A critic then needs to explain whether the remaining risks justify pausing development or changing where and how it happens.
If one fleet travels much farther than another, its total number of crashes can be higher even when its crash rate is lower. The US safety agency warns that its reports have limits, including differences in reporting and missing information about driving exposure. [C6] Ask about crashes per mile, the roads used, and the seriousness of injuries.
An ongoing investigation concerns a Waymo vehicle that struck a child near a school in January 2026. Its preliminary report records minor injuries; the cause remains under investigation. [C5] Discuss this carefully. One incident cannot settle the whole safety debate, and an unfinished investigation cannot establish blame.
Money, road space, and public attention are limited. A critic can ask whether safer crossings, accessible buses, or better pavements should come first. This is a comparison students should investigate locally; this guide does not establish that any one option always costs less or saves more lives.
Automated vehicles can produce sensitive information such as location data. [C8] Who may keep it, and for how long? Both sides should propose protections. A privacy concern alone does not show that the technology cannot be developed responsibly.
04 / Research and preparation
Imagine ten families sharing cars instead of each keeping one. That could leave more space for other uses. But a shared car might also travel empty to pick up the next person. An International Transport Forum study used a computer model of Lisbon to explore shared self-driving travel. It found that a smaller fleet could still travel farther in total. Keeping high-capacity public transport and sharing rides changed the results. [C10]
Supporters can argue for developing shared services that work alongside public transport. Critics can ask whether empty journeys and extra car use would make streets busier. The study models possible arrangements; it does not show that every future town will have more traffic or less pollution.
A good average can hide a problem for one group. A UK ethics report asks developers to consider how risks and benefits affect different road users. [C9] Ask how a car behaves around children, cyclists, and people who move slowly. A town should hear from people outside the car as well as its passengers. This is an argument about fair development, not proof that a particular group is currently less safe.
C2 comes from a safety research institute. C11 is research involving the company whose vehicles it studies. Both can be useful. Check their methods, ask what they counted, and look for checks by other researchers. Company involvement is a reason to examine evidence carefully; it does not by itself make a finding wrong.
Fleet A has two bumps during 100 equally long trips. Fleet B has three bumps during 300 equally long trips. B has more bumps in total but fewer for each 100 trips: one rather than two. These are made-up numbers. Before judging safety, you would also ask how serious the bumps were and whether the roads and conditions were alike.
05 / Research and preparation
The school crossing: A town wants to test cars beside a busy school. What evidence should it ask for first? Who should be allowed to say that the test must stop?
The accessible journey: A passenger uses a wheelchair. The car can drive safely, but the passenger cannot enter it without help. Has the journey become accessible? What else must be designed?
The difficult road: A car works well on familiar roads. Its maker wants to use it somewhere with different weather and road layouts. What new tests would you require?
Use about 40 seconds to introduce the question and your position; 90 seconds for your first reason and evidence; 90 seconds for your second reason; 50 seconds to answer the strongest objection; and 30 seconds to explain your conclusion. This is a suggested practice plan, not an additional competition rule.
Build each reason with four parts: your claim, a source or example, why it matters, and a limit or objection. Say the source in ordinary language, such as ‘A 2026 study by the Insurance Institute for Highway Safety found…’. Rehearse aloud and adjust the length to your own speaking pace.
Do not say that self-driving cars never crash, that one crash proves all systems fail, or that cameras automatically make driving safe. Do not assume that a self-driving car is electric or that all disabled passengers have the same needs. Use a grown-up reading partner for technical reports. You can explain an idea simply without making it stronger than the evidence.
06 / Read the original sources
Start with C1–C3 with an adult. Each source explains what it supports and what it leaves unproved. External readings open in a new tab.
All summaries below are paraphrases. The questions and suggested arguments are original teaching material. Start with C1–C3; adults can help select short passages from the longer investigations.
C1
Official explanation of levels of automation. Level 2 requires continuous driver attention; Level 4 can handle driving within specified limits. Use it to define the topic. The different levels should not be treated as a single technology. Read the automation-level section first.
C2
An insurer-supported safety research institute reports its analysis of 2021–2024 crash data. Useful for a qualified safety argument. The comparison is observational, concerns Waymo in four cities, and adjusts for reporting differences; it is not a random trial or a guarantee about other systems. The article also explains monitoring gaps. Read the opening findings and methods explanation.
C3
Describes projects addressing barriers such as locating a car, boarding, securing mobility equipment, and getting help. Useful for arguments about independence and fair access. These are designs and demonstrations, not proof that every service already meets these needs or that every passenger can travel without assistance.
C4
An official completed investigation into the fatal Tempe crash. Use it to examine the interaction of operator behaviour, engineering, and company oversight. It describes a particular historical test system, not a representative sample of current vehicles. An adult should choose suitable passages from What We Found; graphic detail is unnecessary.
C5
A current example of a collision near a school. The agency still lists the investigation as ongoing. Use it to ask about vulnerable road users and evidence. Do not attribute a final cause, imply a fatality, or treat one incident as a fleet-wide safety rate. Recheck the status before publication or competition use.
C6
Official crash reporting information and its limitations. Useful for understanding why raw totals are not reliable comparisons of safety. Reporting thresholds, coverage, and exposure can differ. Read Data and Limitations with an adult; do not rank companies by simply counting reports.
C7
A study of historical human crashes explored which error types automation might address. It argues that better sensing alone would leave other problems, such as unsafe planning. This is a conditional analysis, not a measured ceiling on the safety of modern fleets. Use it to test sweeping promises, alongside the newer C2 findings.
C8
Its privacy discussion identifies concerns about location and other vehicle information. Useful for asking who controls data and how it is protected. The page includes historical voluntary guidance, so do not describe every recommendation as current binding law.
For each claim, write down the source, what it actually says, and one limitation. Label your own imagined examples. Ask whether the writer is reporting a result, making a prediction, or recommending a rule. A surprising number is useful only when you can explain what was counted and what it was compared with.
Topic wording and age division were checked against the live Global AI Debates homepage on 24 October 2026. This guide supports preparation; students should follow the competition’s published submission instructions.
C9
UK policy analysis covering safety, fairness, privacy, and public trust. Useful for asking whether a better average hides an unfair distribution of risk. Read section 3 on fairness, particularly risk distribution and vulnerable road users. It sets out recommendations; do not present every recommendation as current law or a measured finding about today’s fleets.
C10
A computer simulation using Lisbon travel patterns explores shared self-driving fleets with different sharing and public-transport arrangements. Fleet size and distance travelled move differently. Useful to both sides on urban benefits and extra travel. This is a Corporate Partnership Board modelling report, not a real-world trial or a prediction for every city. Read the executive summary and section 4; distinguish fewer vehicles from less traffic.
C11
Peer-reviewed retrospective research linked from Waymo’s research page reports lower rates for selected injury and crash-severity measures compared with matched human benchmarks. Data extend through January 2025. This involves the developer and is not an independent replication of C2. The observation period and outcomes differ; the datasets may overlap. Do not pool the studies or compare their percentages as if they measured the same thing.
Use C2 and C11 for qualified safety claims, C3 for accessible design, C9 for fairness questions, and C10 for the difference between fleet size and travel. None of these alone decides whether to support development. Explain how your evidence connects to the claim you want to defend.
Middle School Speech · Ages 11–13 · Fall 2026
Should some jobs remain human even when AI can perform them equally well? Explore consent, accountability, care, and public authority through balanced arguments and research.
“There are jobs that AI should not be allowed to do, even if it is as good as humans.”
Read with a parent or teacher where helpful. Choose your own position and explain your reasons.
01 / Research and preparation
Suppose an AI can perform a job as well as a human. Would that settle whether it should be allowed to do it? This guide examines whether good performance is enough, or whether some roles carry responsibilities and relationships that should remain human. Use it to develop your own seven-minute speech on either side.
The words ‘even if’ matter. Saying that today’s AI sometimes makes mistakes does not answer the motion by itself. Start by granting equal capability for the job you discuss. Then ask whether a restriction is justified for some further reason. Explain what ‘as good’ includes: accuracy alone, or also fairness, communication, and sensitivity to the person affected? Do not quietly assume that the AI is worse at the quality your argument depends on.
A teacher can explain fractions, notice distress, safeguard a child, and decide when to involve a parent. A judge can search cases, hear evidence, and issue an authoritative decision. AI assistance with one task is different from handing over the whole role or its final authority. Say exactly which arrangement you defend.
Pro supports the motion. It needs at least one convincing job or core responsibility that AI should be barred from performing despite equal capability. It need not ban AI from every workplace. Con opposes the motion. Showing that AI can do some useful jobs is insufficient; Con should explain why the proposed human-only exceptions fail, or why safeguards answer them without excluding AI.
Autonomy means having meaningful control over your choices. Accountability means that someone must answer for a decision and help put wrongs right. Consent means agreeing freely with enough information. Legitimacy asks whether an authority has the right to make a decision. A prohibition is a rule that forbids something.
‘Should not be allowed’ concerns permission. A personal preference for a human doctor or teacher is relevant, but students must explain when it justifies a general restriction. The arguments below are teaching proposals; the cited research and guidance provide evidence and perspectives, not an official competition position. Sources checked 24 October 2026.
02 / Research and preparation
A judge’s role involves exercising public power over another person, not simply predicting an outcome. Pro can argue that final authority should remain with someone who owes duties to the public, gives reasons, and can be called to account. The Council of Europe’s judicial AI charter emphasises rights, transparency, and user control. [M3] This offers a starting point, rather than proof that all AI judicial tasks must be banned.
The challenge: a court or public body could remain accountable for an AI decision. Pro must explain why institutional responsibility and an effective appeal still would not justify giving AI the particular authority it wants to reserve.
Pro might reserve primary caregiving responsibility for children or highly dependent adults for humans. Its argument could be that people are owed an ongoing human relationship, even when a machine can perform the practical work equally well. UNESCO and WHO treat human agency as important in education and healthcare. [M1, M2] Whether this supports a prohibition is a further ethical argument.
Do not rely on ‘AI cannot care’ without defending that claim. If equal performance includes excellent emotional support, explain whether being cared for by a human still matters independently, and why that value warrants a rule rather than a choice.
Pro can argue that decisions to apply force against people should remain human because of dignity and moral responsibility, even with equal targeting accuracy. The International Committee of the Red Cross recommends prohibiting autonomous weapons designed or used to apply force against persons. [M4]
The challenge: equal competence may remove some safety objections. Pro should identify the ethical concern that remains and explain why responsibility cannot be exercised through human rules and oversight. The ICRC’s position is a recommendation for rules, not evidence that this prohibition already applies everywhere.
03 / Research and preparation
Con can ask what is lost when the job is performed equally well. If an AI meets the relevant requirements, a human-only rule needs a reason beyond unfamiliarity. A trial involving 194 eligible students found stronger immediate learning results for a structured AI tutoring arrangement than for comparison lessons in one university physics course. [M5] This challenges simple assumptions about teaching ability, although it does not establish equivalence across the whole teaching role.
Con must still answer claims about public legitimacy or relationships. A test score does not measure every value people might reasonably want a job to preserve.
Some people might prefer an AI service or find it easier to access. Under equal competence, Con can argue that voluntary use should normally be permitted, with clear information and a human option where needed. Pro can respond that children, defendants, or people with no affordable alternative may lack meaningful choice. Con should address those cases instead of assuming every user freely consents.
A hospital, school, or court could have duties to explain decisions, hear appeals, and remedy harm, regardless of which tools it uses. WHO’s guidance combines potential health benefits with accountability and human rights. [M1] Con can propose enforceable institutional duties rather than a prohibition on whole job categories.
Merely naming a human supervisor is insufficient. Elish’s research warns that operators can receive blame despite having little control over an automated system. [M7] Oversight needs authority, information, time, and a real ability to intervene.
If human relationships matter, Con can defend requiring access to human care while allowing AI to carry out other work. But check the concession: if Con agrees that AI must never exercise a particular job’s core responsibility, it may have accepted the human-only exception Pro needs. Be explicit about the boundary you propose.
04 / Research and preparation
Robert and Linda Sparrow argue that replacing human carers with robots can undermine human contact and involve deception about companionship. [M8] Pro can use this to ask what a person is owed beyond practical assistance. But some of their argument depends on limits they attribute to robots. Under this motion, Pro must explain what objection survives equal capability.
John Danaher challenges the view that robots could only fake friendship. He argues that robot relationships could have value and could also support human friendships. [M9] Con can use this to question a human-only assumption. Friendship is not identical to professional care, however: an argument must connect the relationship’s value to the duties of the job.
Bigman and Gray’s nine studies found resistance to machines making moral decisions, including some cases with positive outcomes. [M10] That helps identify a concern people have. It does not establish that the concern justifies a prohibition. Some scenarios described machines as lacking compassion; those are not a clean test of the motion’s equal-capability premise.
Pro can argue that losing public trust has consequences. Con can ask whether informing users, giving meaningful choices, or improving governance could address that concern without banning AI from the role. Both need reasons beyond a popularity vote.
The ILO’s 2025 study assesses exposure to generative AI at the task level. Exposure means some tasks could be affected; it is not a count of jobs that will disappear. [M11] It helps students separate tasks from occupations, but it does not determine which jobs should be reserved for humans.
Name the exact job or responsibility, grant equal competence, identify the value at stake, and test a less restrictive alternative. If you defend a human-only rule, explain why safeguards are insufficient. If you reject it, explain how your alternative protects that value.
05 / Research and preparation
These cases are thought experiments. Assume equal competence on the stated task; do not present the cases as existing deployments.
The judge: An AI and a human reach equally accurate, fair, and well-explained decisions. Both sit within an institution that offers appeals. Is there still a reason the final judgement must come from a human? Identify that reason without reintroducing lower AI performance.
The teacher: An AI teaches and responds to pupils as effectively as a human. A school wants it to take full responsibility for a class. Would you permit it? Distinguish successful instruction from safeguarding duties and the value of sustained human contact.
The caregiver: An adult freely chooses a competent AI care service. A second person has only that service available. Does permission depend on genuine choice? Would a human option address the concern, or should the role always remain human?
Suggested practice timing: 45 seconds for definitions and position; two minutes for your strongest reason; 90 seconds for a second reason or case; two minutes answering the strongest objection; and 45 seconds for your judgement. Use a claim, reason, evidence or example, and explanation of why it matters.
Ask your practice partner: Have I honoured equal competence? Am I discussing assistance or replacement? Does my evidence support this exact claim? Have I shown why a preference justifies a prohibition, or why safeguards remove the need for one?
Research about current errors can reveal what needs checking. It cannot alone prove that equally competent future AI should be excluded. Equally, a useful AI task does not prove that every human role can be handed over.
06 / Read the original sources
Start with M1–M3 with a teacher. Each source explains what it supports and what it leaves unproved. External readings open in a new tab.
These are annotated starting points. All summaries are paraphrases. Distinguish an organisation’s ethical guidance from an experimental result or a binding legal requirement. Begin with the overviews; longer technical passages can be read with a teacher.
M1
Official guidance developed with an expert group. It discusses health benefits alongside autonomy, accountability, equity, and other protections. Useful to either side when defining responsible healthcare. It is ethical guidance, not a comparison trial or a universal ban on AI healthcare. Start with the overview and the report’s six principles.
M2
Advocates human agency and age-appropriate educational uses, with attention to privacy and inclusion. Useful for discussing what students should retain when AI enters a school. It concerns generative AI and does not evaluate every form of automation or settle whether an equally capable AI could occupy a whole teaching role. Read the overview, then sections 5 and 6 of the linked guidance.
M3
An official framework emphasising fundamental rights, non-discrimination, quality, transparency, and user control. It recognises possible improvements to justice while setting conditions. Use it for questions about public authority and meaningful review. It is not proof that AI judges are more or less accurate, and it should not be presented as a worldwide law banning them.
M4
Recommends new legally binding rules, including prohibiting autonomous weapons that target people and regulating other systems. Useful for examining human control and dignity. Some concerns involve predictability; those need reconsidering under the motion’s equal-competence assumption. Its ethical objections require separate discussion. Read recommendations 1–3; this is the ICRC’s position, not a statement that all recommendations have become law.
M5
An open-access randomised crossover study included 194 eligible students and two university physics lessons. A structured AI tutoring arrangement produced higher immediate learning gains than the comparison. Location, pacing, presentation, and feedback also differed, so this tests a teaching arrangement rather than isolating the effect of AI alone. It does not establish long-term outcomes or whole-teacher replacement. Read Methods and limitations. DOI 10.1038/s41598-025-97652-6.
M6
A study found that using healthcare spending to predict need produced racial bias. At the same score, Black patients had greater illness burdens. Use it to ask what a performance metric leaves out. It does not establish that an equally capable future AI must be excluded. The linked author page provides a readable summary under Algorithmic Bias and links to the Science paper. DOI 10.1126/science.aax2342.
M7
A qualitative analysis of accidents and surrounding accounts examines how people can be blamed for systems they scarcely control. Useful to both sides when testing whether oversight is meaningful. It is an analysis of accountability, not a numerical estimate of how often oversight fails or proof that all human–AI teams fail. Read the abstract first. DOI 10.17351/ests2019.260.
Record the author or organisation, title, date, link, the claim you use, and its limitation. Label facts, forecasts, ethical recommendations, and your own reasoning separately. Say a source’s name aloud where it helps listeners judge a factual claim. Do not turn a paraphrase into quotation marks or cite a source for an argument it does not make.
Topic wording and age division were checked against the live Global AI Debates homepage on 24 October 2026. The guide offers competing arguments and research support. It does not prescribe the view students should defend.
M8
A philosophical argument about autonomy, human contact, and deception in robotic care. Useful for a serious Pro case, with limits: it is not an experiment, and its descriptions of technical capability are historical. Separate objections about care itself from assumptions about what robots cannot do. Read the sections on companionship and the conclusion with a teacher. DOI 10.1007/s11023-006-9030-6.
M9
A philosophical defence of possible robot friendships and their value alongside human relationships. Useful as a Con challenge to the claim that worthwhile relationships must always be human. The linked university record provides the author’s abstract and publication details. It does not prove that existing machines are genuine friends or settle professional care duties. DOI 10.5325/JPOSTSTUD.3.1.0005.
M10
Nine psychology studies examine attitudes to machine decisions. Useful for distinguishing public acceptance from moral justification. Some scenarios explicitly contrast a compassionate human with a machine lacking compassion; a further replication removes that explicit wording. Reported preferences therefore do not establish a justified prohibition under fully equal capability. The link opens an author-hosted full paper. DOI 10.1016/j.cognition.2018.08.003.
M11
Task assessments, expert input, and model predictions estimate exposure across occupations. Useful for understanding that jobs contain multiple tasks and may change rather than disappear. Exposure is not observed job loss or a forecast that a whole job will be automated. The study concerns generative AI, not every robot or AI system. Read the overview before the methods.
Global AI Debates · Fall 2026 · Resources
Understand the motion. Read the strongest challenges. Build a case you can defend.
“AIs should be granted the rights of legal persons.”
For the High School Speech and Essay & Debate tracks. For track requirements, see the competition FAQ.
01 / Understand the question
A status through which the law recognises an entity as a holder of rights, powers or duties. Its content depends on the legal system and the kind of entity. Corporations are a familiar example. Read source 01.
Mattering for one’s own sake. A being could deserve protection even if it cannot sign a contract or be held morally responsible. Read source 04.
Consciousness concerns subjective experience. Sentience, as used in welfare debates, concerns experiences such as pleasure or suffering. Intelligent behaviour alone does not settle either question. Read source 06.
02 / Start reading
For each source, write down its claim, its reason, and the strongest objection you can make to it. External readings open in a new tab.
A short introduction to a familiar nonhuman legal person. Corporations can sue and be sued, and their legal identity is separate from their shareholders.
Read to answer: Which legal capacities would an AI need, and who would benefit?
Use with care: Corporate personhood shows that the law can recognise nonhuman entities. It does not settle whether AI should receive that status.
A foundational exploration of whether AI could be a legal person. Solum tests the idea through questions about trusteeship and constitutional rights.
Read to answer: Which objections depend on an AI’s abilities, and which depend on its artificial origin?
Use with care: This is an exploration of possible legal recognition, rather than a blanket endorsement of rights for every AI. It predates modern generative AI.
A sustained case against AI legal personhood. The authors examine whether new legal entities could weaken accountability and protection for people.
Read to answer: If an AI causes harm, who has the money and legal duty to compensate the victim?
Use with care: An argument against a legal framework is different from proof that artificial consciousness is impossible.
03 / Follow your argument
Choose the branch your argument needs. Each reading includes a question to answer and a limit to keep in view.
For arguments about interests, protection and uncertainty.
Compares proposed grounds for moral status, including cognitive capacities and relationships. Start with the introduction, then read the account most relevant to your case.
Read to answer: What would make an AI matter for its own sake?
Use with care: Moral status alone does not specify which legal rights an entity should receive.
Argues that uncertainty about future AI consciousness and agency gives us reason to assess AI welfare and prepare policies for it.
Read to answer: How much evidence should we require before taking protective action?
Use with care: The authors do not claim that AI consciousness is established. Their welfare recommendations do not by themselves establish a case for legal personhood.
For claims about whether AI could have experiences.
Develops indicators of possible AI consciousness from scientific theories. The report’s assessment of the systems it examined in 2023 was negative, while leaving open future possibilities.
Read to answer: What evidence would distinguish fluent behaviour from conscious experience?
Use with care: Date the finding accurately. A 2023 assessment cannot settle the status of every later system.
For arguments about who should answer for AI’s actions.
Opposes the European Parliament’s 2017 proposal to consider electronic personhood. It challenges both the claimed liability gap and the proposed legal models.
Read to answer: Could regulation and human liability solve the problem without AI personhood?
Use with care: The letter expresses its signatories’ position. The proposal it discusses was not a law granting robots personhood.
A broad guide to the field. Use the contents to find “Moral Status, Machine Ethics, Responsibility” and “Autonomous AI Systems.”
Read to answer: How do moral agency, moral status and responsibility come apart?
Use with care: Use the sections relevant to the motion. General benefits or harms of AI need an explicit link to legal rights.
For testing what the comparison with other legal persons can establish.
Explains the settlement framework, the river’s legal recognition, and Whanganui Iwi’s relationship with it. Read “Legal Recognition of Te Awa Tupua.”
Read to answer: What purpose does legal recognition serve, and how close is the analogy to AI?
Use with care: Keep the Indigenous and Treaty settlement context in view. Personhood does not mean every human right applies to every legal person.
For an advanced argument about political personhood.
Explores a Rawlsian route to artificial personhood through capacities for justice and a conception of the good, without making sentience a requirement.
Read to answer: Could a nonsentient system have the capacities relevant to political personhood?
Use with care: The authors do not attribute those capacities to current AI. Political or moral personhood also needs a further argument to establish particular legal rights.
For a concrete discussion prompt alongside the research.
Netflix’s account of its documentary about AI agents and the Hugging Face breach. Use it to frame questions about authorisation, oversight and liability.
Read to answer: Which actor’s decisions would you investigate before assigning responsibility?
Use with care: A dramatic example of agent behaviour cannot establish consciousness or a need for legal personhood. Verify factual claims against underlying records before citing them as evidence.
04 / Test both sides
These are starting points for your own argument. A strong case explains which rights, for which systems, under which conditions.
Limited personhood could give qualifying AI systems a defined role in transactions and disputes. Consciousness need not be the basis of this argument.
Hard question: What problem would this solve better than treating AI as a tool acting for an existing person or company? Explore 01 and 02.
If some AI systems can have welfare or interests of their own, legal rights could give representatives a way to defend those interests.
Hard question: Why would targeted welfare protections be insufficient? What evidence would qualify a system? Explore 04, 05 and 06.
If rights depend on capacities such as agency or a sense of justice, an artificial system that has those capacities may have a claim to recognition.
Hard question: Which capacities justify which rights, and how would you establish that a system has them? Explore 02 and 10.
A company might direct claims toward a separate AI entity that lacks assets. Victims could gain a defendant and still lack a useful remedy.
Hard question: Could mandatory insurance, capital requirements and continued human liability address this concern? Explore 03 and 07.
Persuasive language and apparent preferences do not establish the capacities on which a rights claim depends. Your objection should identify the missing evidence.
Hard question: What should we do when withholding protection might also cause serious harm? Explore 05 and 06.
Safety duties, insurance, human liability and carefully defined protections might address the relevant harms without creating a new legal person.
Hard question: Can your alternative protect an AI’s own interests, if it has them, as well as human interests? Explore 03, 05 and 08.
05 / Turn reading into a case
Copy this for each main argument. Keep the original passage and its context alongside your notes.
CLAIM: What am I asking the judge to accept? REASON: Why should they accept it? EVIDENCE: Author, title, date, URL and page or section. LIMIT: What does this source leave unproved? LINK: How does this support my position on specific legal rights? OBJECTION: What is the strongest reply? RESPONSE: Why does my argument still hold?
06 / Next steps
Fall 2026 speech videos and initial essays are due 21 November 2026. Live finals are online on 5 December 2026. Use the official pages below for track requirements and judging criteria.