Ethics in Technology

optional Theme - Technology

When an autonomous system kills the wrong person, who is responsible?

When content spreads without anyone asserting it, what happens to truth?

The robots can do kung fu

THE PROVOCATION

The machines move in perfect synchrony - sweeping kicks, held stances, coordinated sequences that would take a human years of dedicated practice to approximate. The video is extraordinary. It is also, in a precise sense, a demonstration of nothing: the robots do not know they are doing kung fu. They have no knowledge of what martial arts is, no understanding of what the movements mean, no awareness that they are performing. They are executing motor programs derived from vast quantities of training data. By April 2026, the same trajectory had extended to competitive sporting performance in the physical world. An AI-powered robot beat elite table tennis players - humans who had spent years developing the embodied, tacit knowledge of the game that Polanyi described on the Methods and Tools page. It processed sensor data and generated motor responses. The Methods and Tools page established why this matters epistemically and this page asks what it means when those same AI systems are set up to take human lives.

The movement systems in those demonstration videos - the coordination and real-time environmental response - are the same systems being developed for military deployment. In Ukraine and Gaza, autonomous drone systems with degrees of target recognition are already operational. The tempo of modern conflict exceeds what human decision-making can sustain at every step. Autonomous systems offer speed and scale that no human chain of command can match.

When an autonomous system identifies a target incorrectly and kills civilians, the question "who is responsible?" has no clean answer. These concerns came to the fore in the aftermath of the USA's 2026 strike on Shajareh Tayyebeh girls' school in Minab, southern Iran, which Iran says killed at least 168 people, most of whom were schoolchildren. The suspicion was that this was a result of an AI error. The programmer wrote the algorithm years earlier, in a different country, without knowing the specific conditions of the deployment. The military commander authorised the use of the system in general terms. The political leadership approved the mission. The manufacturer sold a product. No individual made the specific decision to fire at that specific target at that specific moment. The decision emerged from a process - and the process cannot be held accountable in any of the ways that existing legal and moral frameworks require.

This is the provocation that runs through the three big ideas on this page. In each case, the question of who is responsible for the harm becomes genuinely difficult to answer. And in each case, the ethical frameworks developed in the core Lesson 7 and Lesson 8 - what knowing obliges you to do and what moral progress demands - are directly relevant but only partially adequate.

Big idea 1 - From the printing press to AI: what we owe each other as truth-tellers

In 1439, Johannes Gutenberg's printing press made it possible, for the first time, to reproduce text at scale without hand-copying. The first mass information environment followed within decades. Before the press, manuscript production was controlled by monasteries and guilds: the institutions that copied text were also the institutions that certified its credibility. The press destroyed this gatekeeping function. Anyone with access to a press and type could publish anything. The cost of content fell to near zero. The speed of dissemination increased by orders of magnitude. The consequences were both extraordinary and catastrophic, and they arrived together.

As we learnt back in 11e history, Martin Luther's ninety-five theses spread across Germany in printed form within weeks of 1517 - something that would have been impossible a generation earlier. The Wars of Religion that broke out in Europe killed more people per head of population than either of the World Wars and it started with the printing of words.

The Reformation was, among other things, a media event made possible by the press and the pamphlet wars that accompanied it. Where there had been a roughly unified Latin Christendom with a dominant certifying authority, there were now multiple institutions making incompatible claims, each with genuine power in its own territory.

From the reader's perspective, this created a new and disorienting problem: a multiplicity of competing authorities with no agreed way of choosing between them, rather than a single dominant one. A text condemned in Rome circulated freely in Wittenberg.

The ethical concept that connects the Reformation pamphlet wars to AI-generated content in 2025 is one we encountered in Knowledge and the Knower - Lesson 7. The temptation is to stop caring whether a claim is accurate - to circulate content because it confirms what you already believe, or because it generates attention. This is Frankfurt's bullshit: indifference to the question of truth rather than deliberate departure from it.

Yuval Noah Harari's argument in Nexus extends this analysis into new territory. Harari observes that the traditional chain of knowledge accountability runs through the asserter: a claim is made by someone who stands behind it, whose credentials and track record can be examined, who can be asked to justify what they said. Testimony, expertise, reputation, and institutional standing are all mechanisms for evaluating what is claimed and who is claiming it. But AI removes the asserter from the equation entirely. A language model generates text by predicting which words are most likely to follow previous ones - without any human standing behind the output. The content may be accurate. It may be inaccurate. It circulates without the accountability structure that makes the distinction evaluable.

Can AI 'cure' grief?
BBC

One application of AI makes the asserter argument concrete in a way that is difficult to dismiss. AI systems can now recreate dead people with sufficient accuracy - voice, manner, image, apparent personality - that the recreation can hold conversations with the living. The people being recreated have not consented. They cannot consent. AI ethicists have called for urgent regulation. The question for TOK is whether regulation addresses the right problem - or whether the problem is the removal of the asserter, which regulation cannot restore.

The deeper consequence Harari identifies is that in an environment where any actor can produce unlimited quantities of plausible-looking content instantly and cheaply and distribute it to targeted audiences, the tools by which societies have traditionally distinguished credible claims from incredible ones - source evaluation and expert authority - become unreliable faster than they can be rebuilt. This is the Reformation pamphlet problem at a post-industrialised scale, with the asserter removed entirely.

When a student submits AI-generated work, it makes a false claim about authorship - and raises a Frankfurt question: whether the student is indifferent to the truth about their own understanding.

There is no algorithm rating the daily behaviour of the general population.The researcher Jeremy Daum, who has read the primary Chinese-language sources systematically, has documented the gap between the myth and the reality in detail. The myth spread through a combination of genuinely alarming elements (China's surveillance infrastructure is real and extensive), Western journalism that cited other Western journalism rather than primary sources, political rhetoric that found the narrative convenient, and the cultural influence of Black Mirror presenting fiction as anticipation. Nobody lied, exactly. The Frankfurt bullshit category is more precise: the narrative circulated with widespread indifference to whether it was accurate. As a TOK exercise: how would you verify the claim? What sources would be authoritative? And what does the spread of the myth tell you about the assertion problem Harari identifies - including in democratic societies?

Does China really rank its citizens with social credit scores?

As in this 2019 France24 news report Chinese social credit system is widely described in Western media as a unified national scoring system that assigns each citizen a numerical rating determining their access to travel, housing, employment, and public services - a real-world version of the Black Mirror episode Nosedive. This description is largely false. What exists is a collection of separate, largely disconnected systems: a standard financial credit system for individuals and businesses, and sector-specific compliance registries. Court-ordered blacklists for debt defaulters restrict travel for those who have already been through a legal process.

Big idea 2 - From the Luddites to autonomous weapons: who the system answers to.

The Luddites are remembered as workers who smashed machines because they feared change or failed to understand progress. As good history students taught by me, you will know their actual argument was more specific than that. The frame-breakers who destroyed power looms in the English Midlands between 1811 and 1816 were making a specific ethical claim, defined by what the great English historian E.P. Thompson called the moral economy of the crowd. The introduction of machinery was destroying a form of skilled knowledge and a way of life that had taken generations to build, and that this destruction was being imposed without the consent of those who would pay the price of innovation.

A skilled weaver set his own pace. He read the tension of the yarn and the rhythm of the loom, and worked accordingly. The power loom set the pace. The weaver adapted to it. Speed, hours, physical position, and the repetitive demands of machine-timed production were now determined by the machine. The worker became, in a precise sense, an appendage to it - or as Marx put it in Capital, an appendage of the machine. Parliament was fully aware of who was paying these costs and acted deliberately to protect the machines over the workers. The Frame Breaking Act of 1812 made machine-breaking a capital offence. In 1813, seventeen Luddites were hanged at York.

The ethical wrong was that the decision about how to deploy the technology was made by those who would benefit from it, while those who paid the costs had no voice in the decision and no accountability structure to appeal to.

Amazon warehouse employees work against pick-rate targets set and monitored by algorithms. The algorithm calculates how quickly a worker should be able to locate and pack each item and generates the disciplinary record that leads to termination. Workers report that the pace required leaves insufficient time for toilet breaks and that the system does not adjust for human variation in the way a human supervisor might. The designers of the algorithm are identifiable. The company that deployed it is known. Accountability is present. The workers subject to its decisions have no mechanism for contesting the targets it sets for their bodies. See this 2026 multimedia article by Human Rights Watch.

Cathy O'Neil's analysis in Weapons of Math Destruction identifies the same structure in algorithmic decision systems applied to legal and professional life. The COMPAS recidivism algorithm, used in American courts to inform sentencing and parole decisions, takes inputs about a defendant's history and social circumstances and produces a risk score predicting the likelihood of reoffending. Judges use this score. The defendant is not told how it was calculated. The algorithm's designers argue it is proprietary. The inputs include factors that correlate with race, meaning the system reproduces and amplifies existing bias in a form that is opaque to those it affects and legally difficult to challenge. ProPublica identified the designers. The courts that adopted it are known. The legislators who chose not to regulate it are on record. Again, accountability is present throughout this story.

The era of blind faith in big data must end - Cathy O'Neil
TED

O'Neil argues that algorithms function as opinions embedded in code: built from past data and the assumptions of whoever defined success, they reproduce the biases encoded in their design. She calls for public scrutiny and democratic accountability.

Kwame Anthony Appiah's argument about moral progress, from Lesson 8, can again end the analysis. Appiah recognised that moral progress tends to arrive through a shift in what counts as unacceptable rather than through better arguments. The practices that previous generations considered normal - Atlantic slavery and foot-binding - came to be seen as obvious wrongs, and future generations will extend that judgment to practices we now normalise. The Luddites fought for social justice at a time when trade unions were illegal. Workers won the right to representation through trade unions and over time what was considered to be acceptable employment practice evolved. Appiah's question is which of our current certainties will be the foot-binding of 2100. Algorithmic sentencing that reproduces racial bias while structuring itself to exclude the defendant from challenging it is a serious candidate. Autonomous weapons systems that kill civilians in identifiable error patterns while distributing the accountability for each individual death across a chain of decisions too long and diffuse to prosecute is another.

Big idea 3 - In an age of AI what are schools for?

Let me conclude this series of lessons on technology by bringing together three of my favourite themes once more: history, education and science fiction (and a little philosophy).

The form of education provided at each historical stage tracked the economic function of the people receiving it, and extended no further than that function required. This is what is called the instrumental purpose of education. Society trains its young for what society will need. Peasants in pre-industrial Europe did not receive formal schooling because it was not required for agricultural labour: the knowledge needed to work the land passed through apprenticeship and practice. As we have studied in history on various occasions, mass education is a recent invention. Compulsory schooling spread through industrialising societies in the nineteenth century - Forster's Education Act 1870 in England, Bismarck's kulturkampf, universal elementary education in France after Jules Ferry's laws of 1882 - because industrial capitalism needed workers who could read instructions and submit to the discipline of clock-governed work.

After the second world war there was a significant expansion of secondary and then higher education. Economies needed people who could manage and create as well as follow written orders. I was one of those who benefitted from that expansion and I became the first member of my family to go university. But if the sort of work that drove that expansion of mass education is being automated - reading, writing, translation, research, analysis, coding - then the instrumental case for providing that education weakens in exactly the way that made peasant schooling seem unnecessary. At the moment it seems that education can go one of two ways. The first is a new instrumental adjustment: teach the skills needed to direct and evaluate AI systems and focus on human judgement. This is what we have always done historically and what many schools and universities are debating (or panicking about) right now. The second response is to recognise that this is the first moment in the history of mass education when the instrumental justification could collapse entirely - and to ask what state education might become instead.
Before we turn to sci-fi lets quickly look at the philosophy. Aristotle's answer to what education is actually for we encountered in Lesson 8. Eudaimonia - living well, human flourishing - is the activity of a person developing and exercising distinctly human capacities across the indefinitely varied situations of a life. Aristotle called the central capacity phronesis: practical wisdom, developed through practice in real situations. On this account the purpose of education was to develop the kind of person who can think and judge well when the situation does not fit any predetermined template - producing essays and passing examinations were proxies for that development. Lessons in TOK would seem to have a future in Aristotle's vision.

Of course, Marx also made a contribution to this debate. His vision was the abolition of the condition in which a person is reduced to their economic function - a vision of what work does to the person who does it. We studied the concept in history as 'alienation'. "In communist society," he wrote, "it is possible for me to do one thing today and another tomorrow, to hunt in the morning, fish in the afternoon, rear cattle in the evening, criticise after dinner, just as I have in mind, without ever becoming hunter, fisherman, herdsman or critic." The Luddite framework-knitter who lost their craft lost the activity through which they understood themselves. Like Aristotle, Marx's free person does the same activities without being reduced to them.

Marx and Aristotle both provide optimistic visions of the future. In practice, the outcome could go two ways. One possibility is a revival of education as self-development: inquiry, arts, philosophy, dialogue and learning as living rather than training for work. The other is what critics like Illich or Foucault might predict: even "free time" becomes structured by systems of surveillance and algorithmic recommendation, shaping desire as much as labour once did.

The future of education: optimism v pessimism

The two pieces of science fiction I mentioned earlier provide for these optimistic and pessimistic visions. Aldous Huxley's Brave New World chooses comfort over freedom; Gene Roddenberry's Star Trek uses freedom to find meaning.

Brave New World imagines the 'meaning of education' problem solved by removing the desire for it: Mustapha Mond has given up science and literature for stability. The World State is a dystopian post-scarcity techno-totalitarian state. It views human history and deep emotion as unstable diseases. Soma is Huxley's pharmaceutical shortcut to happiness - a drug that eliminates dissatisfaction and the desire to question, with no hangover. Mustapha Mond describes it as 'Christianity without tears.' Human beings are treated as factory products designed to fit a permanent economic machine.

Star Trek, in contrast, offers the optimistic alternative. The Federation is a utopian post-scarcity democracy. It views historical struggles as stepping stones toward maturity. Human beings are free agents who collaborate across species to better themselves and the galaxy. In the Federation, replicators have eliminated scarcity, and like in Marx's German Ideology, Picard makes wine and Data paints - because they choose to.

A third piece of science fiction covers the AI and robot dimension. I couldn't not include a Pixar film. Wall-E's humans have drifted into passivity through accumulated convenience rather than explicit design. That's arguably closer to where our current attention economy argument actually points: nobody decided to make humans unable to look up from their screens. It happened through a series of individually rational choices by individuals and companies that added up to something nobody chose. This connects directly to the Scope page's unintended consequences argument and to Zuboff on the Perspectives page. Varoufakis's technofeudalism is almost a precise description of Buy-n-Large - a single corporate entity owns the means of life and directs all human activity. Of course, this is Disney and the humans win out, led by a robot but achieved through collective rebellion. In the end the Luddites win.

An AI generated image. Left: Huxley's World State - humans conditioned via Soma into contentment. Centre: Wall-E's Axiom - people who drifted into passivity through convenience. Right: Star Trek's Federation - a figure tending vines at dusk, work chosen rather than required.

There is a question that follows directly from the science fiction: if AI does displace human work at scale, what happens to the people whose work disappears? Whether the outcome resembles Wall-E or Star Trek depends partly on policy choices that are already being debated. One of them is Universal Basic Income.

UBI is a policy in which the government regularly pays every individual a fixed amount of money, regardless of employment or income. One example from actual UBI experiments suggests that the assumptions of Star Trek might be more realistic than Brave New World. Cook County, Illinois approved a guaranteed income programme in its 2026 budget, building on a pilot where $500 a month to 3,200 households found 94% used it to address financial emergencies and 70% reported a positive effect on mental health. Rutger Bregman, in Utopia for Realists, documents the wider pattern: people freed from financial necessity pursue education and invest in their communities. Bregman also resurrects a prediction Keynes made in 1930: that by 2000, technology would have reduced the working week to fifteen hours. Keynes predicted the technology accurately; the outcome confounded him. Hours have barely fallen, because work became identity and status as well as survival.

What schools choose to build in students - the Aristotelian capacity to live and judge well, or the narrower set of skills the economy currently needs - is the educational version of the same problem. The film below makes the economic argument. Aristotle and Bregman make the case for what would have to follow from it.

Former presidential candidate Andrew Yang argues that Universal Basic Income (UBI) could protect people from economic instability caused by AI-driven unemployment. Supporters claim UBI could provide security and maintain consumer spending, while critics argue it may discourage work and create major tax burdens. The discussion reflects growing fears that AI could transform society as dramatically as earlier industrial revolutions.

AI's job shake-up is accelerating. Is it time for universal basic income?
CNN

This CNN report argues that AI is already reshaping the labour market. Studies from institutions such as Massachusetts Institute of Technology, Goldman Sachs, McKinsey Global Institute, and the World Economic Forum predict that automation could displace millions of jobs across industries including finance, healthcare, retail, logistics, and administration. Examples include companies such as Amazon and Salesforce increasing automation and reducing staff.

Bringing it together

The point is to change it

Below is the slide from the lesson I first taught of technology in 2023, within weeks of ChatGPT global launch. The then most recent IB TOK textbook published just a few years earlier confidently explained that it is 'difficult to programme chatbots to speak like a human. AI can respond to an individual question, but cannot sustain the thread of a conversation...' Below the extract is the conversation I had with ChatGPT about why AI can't hold conversations. Wherever we are at the moment in the history of AI, we should never underestimate how quickly things can change.

The three big ideas on this page share a structure. In each case, a system or institution accumulates knowledge or capability that creates an asymmetry - it knows more than those it affects, or can act faster than any human chain of accountability can follow. In each case, the question of responsibility for harm becomes genuinely difficult to answer. The ethical frameworks from Core Lesson 7 and Lesson 8 are directly relevant and show that this difficulty is genuine.

The Scope page argued that every deployment of technology produces consequences that were not intended and could not fully be foreseen - we are always inside our own blind spot. Several of the consequences discussed on this page were not fully unintended: the attention economy's harms were known internally before they were public; the accountability gap in autonomous weapons was identified before the systems were deployed. The unintended consequences argument describes genuine ignorance in some cases and a choice not to look in others. Knowing comes with responsibility.

Lesson 8 opened with Marx's eleventh thesis on Feuerbach and returned to it at the close: the philosophers have only interpreted the world in various ways; the point is to change it. The Technology theme ends at the same point. The evidence about what these systems do to knowledge and to learners is now substantial. What institutions and individuals do with that evidence is a separate question - and one that the ethical frameworks on this page make harder to avoid.

Next: Optional Theme 2: Politics

Questions, assessments, films and other stuff.

Questions to think about

The printing press created the first mass information environment and produced both the Reformation and the pamphlet wars. AI is creating another. Is there anything the Reformation experience tells us about how this one will unfold - or is the scale difference decisive?
Frankfurt's bullshit category describes indifference to truth rather than departure from it. Is this a more or less serious ethical failure than lying - and does the answer change when the indifference is structural rather than individual?
The Luddite ethical argument was that the costs of a technological transition were being imposed on those who had not consented to bear them. Is this argument applicable to the deployment of AI in education - and if so, who are the Luddites?
O'Neil argues that algorithmic decision systems reproduce and amplify existing bias while shielding themselves from legal challenge. Is the problem the bias, the opacity, or the removal of human judgment at the point of decision? Could any of these be fixed independently of the others?
Marx argued that free, conscious, creative activity - not just survival labour - is constitutive of being human. If AI takes over creative and cognitive work, does this represent a new form of alienation, or the liberation Marx was aiming for? What would he need to say to answer the question?
Huxley's Brave New World and Roddenberry's Star Trek both imagine a post-scarcity world in which most people no longer need to work. They arrive at opposite conclusions about what happens to human beings as a result. What does the difference between those conclusions depend on - and is it an empirical question or an ethical one?

Exhibition connections
See more exhibition ideas and previous student work here

It is never too early to start thinking about your TOK Exhibition. The ideas in this lesson connect strongly to three of the 35 prompts. Start noticing objects in the world around you that speak to these questions.

Prompt #27: Does all knowledge impose ethical obligations on those who know it?

Suggested object: a smartphone or a social media app - any object that represents knowledge held asymmetrically, where one party knows something that directly affects another party who does not. The argument connects Haidt's evidence about the platforms' internal research to Singer's argument from Lesson 7: if it is in our power to prevent harm without sacrificing anything of comparable moral significance, we are obliged to act. The strongest exhibitions will examine what makes the obligation binding and what it requires in practice - disclosure, redesign, or something else.

Prompt #16: Should some knowledge not be sought on ethical grounds?

Suggested object: an autonomous weapon system interface or a targeting algorithm - any object that represents the application of pattern recognition to decisions with lethal or liberty-affecting consequences. The argument from Big Idea 2 is that seeking the knowledge of how to automate these decisions - possible and technically impressive - creates accountability gaps that no existing legal or moral framework can fill. The exhibition argument asks whether the creation of a technology that by its nature removes human judgment at the point of a consequential decision is the kind of knowledge that should have been sought, and what grounds would make the answer yes or no.

Prompt #11: Can new knowledge change established values or beliefs?

Suggested object: a piece of AI-generated text or a deepfake image - any product of content generation without a human asserter. Harari's argument connects directly: AI-generated content changes the conditions under which all information is evaluated - by making the question "who asserted this and on what grounds?" unanswerable for a growing proportion of what circulates. The exhibition argument asks whether this constitutes a change in an established belief (about how claims should be evaluated) or a change in the conditions that make the belief actionable - and whether there is a meaningful difference.

Feature films
For more see my 10 films for the TOK journey page.

🎬 WATCH — The Social Dilemma (2020)

Jeff Orlowski-Yang

Former engineers and executives from Facebook, Google, Twitter, and Pinterest describe the design choices that made their products maximally engaging and explain, in their own words, why those choices were harmful. The film is useful as primary evidence: testimony from inside the system about the gap between intentions and consequences. It has a point of view and should be read critically as well as seriously. Several engineers describe understanding the harms only after building the systems that caused them - which is the Scope page's blind spot argument applied to the people who built the technology rather than the people who use it. My students can access the film here

🎬 WATCH — WALL-E (2008)

Andrew Stanton / Pixar

Buy-n-Large, the single corporation that owns everything, has solved material necessity and produced something close to Huxley's World State through accumulated convenience. The film connects directly to Varoufakis on techno-feudalism and to Zuboff on the attention economy. The resolution is deliberately optimistic: when humans look up and put their hands in soil, the capacity for meaningful activity turns out to have survived. For Big Idea 3: Wall-E is the argument that the Brave New World outcome arrives through the ordinary logic of systems optimised for engagement and comfort. The question the film leaves open is whether the recovery it shows is realistic, or whether it requires a small robot to arrive first. My students can access the film here

optional Theme - Technology