Movement IV · What It Means for You — Lesson 12
Will AI Change Everything?
The question is badly formed, and the useful response is not an answer but a set of lenses — three of them, held at once, with a stated list of what would move you between them.
Learning objective
By the end of this lesson you can hold three competing framings simultaneously without collapsing into one, state the strongest case for AI's breadth, and — the actual skill — name in advance the specific evidence that would move you between the framings.
12.1 Three lenses, not three positions
Public argument about AI's significance sorts people into camps. The academic literature does something more useful: it treats the framings as lenses that can be applied simultaneously to the same object, each bringing different features into focus. The framework used here follows Makrehchi's Three Lenses on the AI Revolution: Risk, Transformation, Continuity.1
| Lens | Closest analogy | Brings into focus | Blind spot |
|---|---|---|---|
| Risk | Nuclear technology — irreversible and global externalities | Tail outcomes; what happens if capability outpaces control | Median outcomes, which are where almost everyone actually lives |
| Transformation | The Industrial Revolution — a general-purpose technology | Productivity, reorganisation of labour, distributional effects | Timing; transformation lenses are chronically early |
| Continuity | The fifty-year arc from personal computing to internet to mobile | Absorption, institutional adaptation, unglamorous diffusion | Genuine discontinuities, which do occasionally happen |
The paper's central historical claim is worth stating exactly, because it cuts against both extremes: no past transition constituted a strict singularity — disruptive shifts eventually became governable through new norms and institutions — while AI nonetheless carries singularity-class tail risks.1 The conclusion drawn is that AI is both evolutionary and revolutionary: predictable in its median effects while carrying severe worst cases, which argues for coupling pro-innovation strategy with safety governance rather than choosing between them.1
Why three lenses beats one position
A person committed to a single lens makes predictable errors. Committed to risk, you will over-forecast catastrophe and under-prepare for the mundane disruption that actually arrives — Lesson 11's quiet hiring effect is invisible through this lens. Committed to transformation, you will be right about direction and wrong about timing, repeatedly — Lesson 13 explains why. Committed to continuity, you will be right most years and catastrophically wrong in the one year you are not.
One object, three lenses
A single central form — abstract, neither robot nor brain — viewed through three overlapping translucent lenses arranged around it, each tinted differently and labelled Risk, Transformation, Continuity. Through each lens the same object reads differently: through Risk its edges are sharp and a long faint shadow extends off-frame; through Transformation it is shown mid-reconfiguration; through Continuity it appears as one item in a receding row of similar shapes stretching back in time. Critically, the lenses overlap — the composition must argue that they are used together, not chosen between.
Aspect 16:9 · 3200×1800 px · Brand palette: navy #00040c, silver-blue #a8c0d8, column gold #a87848 · Style: conceptual editorial illustration
12.2 The generality argument
The strongest case for “everything” is not about capability. It is about breadth.
Earlier general-purpose technologies began confined — to a sector, an industry, a region — and spread outward over decades, constrained by the physical infrastructure they required. Electrification needed a grid. Containerised shipping needed ports. Generative AI propagated across virtually all industries and geographies at once, because the infrastructure it needed already existed: an internet connection and a browser.2
This is the basis for the argument that AI is better described as a general-purpose cognitive technology — closer in kind to language or writing than to steam — with a range of application as broad as thought itself.2 Take that seriously as the strongest form of the transformation case.
Then apply the discipline this course has been building. That claim is an argument from analogy, published as a working paper, and analogies do not settle empirical questions. It identifies something real about diffusion speed. It does not establish that the effects will be proportionate to the diffusion — and Lesson 13's central finding is precisely that adoption and effect come apart.
12.3 The honest uncertainty
Here is the transferable skill, and it is worth more than any position on the question. Name, in advance and in public, what would change your mind.
A view that no possible observation could disturb is not a forecast; it is an identity. The remedy is to write down the indicators before the evidence arrives, so you cannot retrospectively decide that whatever happened was what you expected. Three worth watching, each of which discriminates between the lenses:
| Indicator | If it moves… | Supports |
|---|---|---|
| Sustained multi-year productivity growth in services | Sustained rise, not a single quarter | Transformation — this is the outcome the GPT framework predicts, late |
| The entry-level employment gap from Lesson 11 | Reverses, or spreads beyond ages 22–25 | Reversal → continuity · spread → transformation |
| A capability jump comparable to 2020's few-shot result | A new behaviour class, not a benchmark increment | Transformation, and raises the weight on risk |
Say which you are watching and why. Then, when one moves, you have already committed to what it means — which is the only reliable defence against updating your reasoning to protect your conclusion.
12.4 Change is not uniform
The final point, and the one with the most human weight. Even the strongest transformation case does not imply that change arrives evenly. Displacement often precedes widespread benefit, and the two do not land on the same people.1
Lesson 11 is this in miniature: aggregate employment rose while a specific cohort in specific occupations saw hiring contract. Both statements describe the same economy. “Things will be fine on average” is compatible with a great many individual outcomes that are not fine, and averages are cold comfort to the person inside the variance. Lesson 13 shows this pattern has a name and a nineteenth-century precedent.
Demonstration protocol
No screen demonstration in this lesson. Use the slot for a tight montage.
- Three domains where change is already measurable — with the measurement shown, not asserted. Ten to fifteen seconds each.
- One domain where the promised change has not arrived. This is the segment that earns the lesson its credibility, and it should be as concrete as the first three.
- Close on the indicator table from §12.3, on screen, and state which one you are personally watching.
Key terms introduced here
- Risk lens
- Framing AI by its tail outcomes and irreversible global externalities. Brings catastrophe into focus; blind to the median.
- Transformation lens
- Framing AI as a general-purpose technology reorganising production and labour. Reliably right about direction, unreliable about timing.
- Continuity lens
- Framing AI as the latest step in a fifty-year computing arc, absorbed through institutional adaptation. Usually right; occasionally very wrong.
- General-purpose cognitive technology
- The claim that AI's range of application is as broad as thought itself, making it closer in kind to language than to steam.
- Falsification indicator
- An observation named in advance that would move you between framings. A view with none is an identity, not a forecast.
Evidence at a glance
| Claim | Specifics | Source |
|---|---|---|
| Three simultaneous lenses rather than three camps | Risk (nuclear analogy) · Transformation (industrial GPT) · Continuity (computing arc) | Note 1 |
| No past transition was a strict singularity | Disruptive shifts became governable through new norms and institutions | Note 1 |
| …yet singularity-class tail risks are argued to remain | Hence: couple pro-innovation strategy with safety governance | Note 1 |
| Diffusion was simultaneous, not sectoral | Propagated across industries and geographies at once via existing digital infrastructure | Note 2 |
| Displacement and benefit are unevenly distributed | Displacement often precedes benefit and lands on different people | Note 1 |
The misconception to kill
“Change is uniform.”
It is not, in two distinct senses that are easy to conflate. Not uniform across time: displacement often precedes widespread benefit, sometimes by decades.1 Not uniform across people: the costs and the gains land on different populations, which is why an accurate aggregate figure can be a misleading description of lived experience — exactly the pattern in Lesson 11.
The related error: treating the three lenses as teams and picking one. Each is a legitimate analytical instrument with a documented blind spot. Someone who can only use one will be surprised in a predictable direction.
Self-check
Answer before revealing.3
Name the three lenses with their analogies and their blind spots.
Risk — nuclear technology; sees irreversible tail outcomes, blind to the median. Transformation — the Industrial Revolution and general-purpose technologies; sees reorganisation of production, unreliable on timing. Continuity — the fifty-year computing arc; sees absorption and institutional adaptation, blind to genuine discontinuity.
What is the strongest case for “everything,” and what is its weakness?
Breadth rather than capability: this technology propagated across essentially all industries and geographies simultaneously because the infrastructure it needed already existed, which is unlike earlier GPTs that spread outward from a sector. Its weakness is that it is an argument from analogy about diffusion, and diffusion is not effect — Lesson 13's whole subject is the gap between the two.
Why is “name what would change your mind” the actual skill in this lesson?
Because a view that no observation could disturb is an identity rather than a forecast, and because committing to the meaning of an indicator before it moves is the only reliable defence against reinterpreting evidence to protect a conclusion you already hold.
Someone says “the aggregate data shows things are fine.” What is the precise problem with that as a conclusion?
Aggregates conceal distribution. Displacement often precedes benefit and lands on different people, so an accurate average can coexist with severe concentrated harm — demonstrably so in Lesson 11, where total employment rose while hiring for one cohort in exposed occupations contracted sharply. “Fine on average” is not a claim about anyone in particular.
Falsification worksheet
Complete this before you next argue about AI's significance. Date it, keep it, and revisit it in six months.
Which lens do I default to, and what does it therefore predictably hide from me?
Indicator 1 — what I am watching, and what movement in it would mean:
Indicator 2 — same:
Indicator 3 — same:
Today's date, and the date I will review this without editing what I wrote above:
Endnotes
Source tiers: Primary peer-reviewed or preprint research; Institutional research body or government; Documentation first-party technical documentation; Journalism established newsroom; Trade vendor or aggregator.
- Primary M. Makrehchi, “Three Lenses on the AI Revolution: Risk, Transformation, Continuity,” arXiv:2510.12859 (v1, 14 October 2025; v2, 13 December 2025). Proposes examining AI through three simultaneous perspectives: risk, which likens AI to nuclear technology in its irreversible and global externalities; transformation, which parallels the Industrial Revolution and treats AI as a general-purpose technology driving productivity and the reorganisation of labour; and continuity, which situates AI within the fifty-year arc of computing revolutions from personal computing through the internet to mobile. Argues that “no past transition constituted a strict singularity: disruptive shifts eventually became governable through new norms and institutions,” while maintaining that AI carries singularity-class tail risks — concluding that AI is both evolutionary and revolutionary, predictable in median effects yet carrying severe worst cases, and that pro-innovation strategy should be coupled with safety governance. Preprint; not peer reviewed at time of verification. arxiv.org/abs/2510.12859
- Primary “Closer to Language than Steam: AI as the Cognitive Engine of a New Productivity Revolution,” arXiv:2506.10281, on generative AI propagating across virtually all industries and geographies simultaneously through existing digital infrastructure, and on the resulting characterisation of AI as a general-purpose cognitive technology with applications as broad as thought itself. Preprint; an argument from analogy, and this course treats it as such. arxiv.org/abs/2506.10281
- Primary H. L. Roediger III and J. D. Karpicke, “Test-Enhanced Learning,” Psychological Science 17, no. 3 (2006): 249–255. doi:10.1111/j.1467-9280.2006.01693.x
Bibliography
Primary research and preprints
- “Closer to Language than Steam: AI as the Cognitive Engine of a New Productivity Revolution.” arXiv:2506.10281. arxiv.org
- Makrehchi, Masoud. “Three Lenses on the AI Revolution: Risk, Transformation, Continuity.” arXiv:2510.12859 (2025). arxiv.org
- Roediger, Henry L., III, and Jeffrey D. Karpicke. “Test-Enhanced Learning: Taking Memory Tests Improves Long-Term Retention.” Psychological Science 17, no. 3 (2006): 249–255.
Referenced from other lessons
- Brynjolfsson, Erik, Bharat Chandar, and Ruyu Chen. Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence. Stanford Digital Economy Lab, revised 12 August 2026. (Lesson 11.) digitaleconomy.stanford.edu
- Crafts, Nicholas. “Artificial intelligence as a general-purpose technology: an historical perspective.” Oxford Review of Economic Policy 37, no. 3 (2021): 521–536. (Lesson 13.) academic.oup.com