AI, Automation, and the Future

Few topics generate more breathless claims — in both directions — than artificial intelligence. "It will solve everything"; "it will end humanity"; "it's all just hype." Scientific thinking is exactly the tool for cutting through this noise: distinguishing what has actually been demonstrated from what is projected, and holding uncertainty honestly.

This is not an AI course (that lives in another subject); it is about thinking clearly about a powerful, fast-changing technology.

Demonstrated versus speculative

The first move is to separate what an AI system has been demonstrated to do — measured on real tasks — from what someone speculates it might do. Impressive demos are evidence of specific capabilities, not proof of general ones. A system that writes fluent text has demonstrated fluency, not understanding or reliability.

The hype cycle and extrapolation

New technologies often follow a hype cycle: inflated expectations, disappointment, then more realistic productivity. A common error is extrapolation — assuming rapid recent progress will continue at the same rate indefinitely. Sometimes it does; often it hits harder problems. Straight-line predictions from a steep curve are frequently wrong.

Reasoning under deep uncertainty

The honest position about AI's longer-term future is uncertainty. Serious, informed experts genuinely disagree about pace and impact. Scientific thinking here means holding that uncertainty — taking possibilities seriously without committing to confident predictions in either direction.

A headline claims an AI "passed a medical exam," implying it can practise medicine. A scientific thinker asks: on what exact task was it measured, under what conditions, and how does exam performance relate to safe real-world practice? Passing a multiple-choice test demonstrates one narrow thing; it does not demonstrate reliable diagnosis with real patients. The claim's reach exceeds its evidence.

Skepticism can overshoot into dismissal. "AI is all hype" ignores genuinely demonstrated, useful capabilities that are already changing work. The mirror error, "AI will obviously do everything soon," outruns the evidence. Both skip the disciplined middle: look at what is measured, note the uncertainty, and update as evidence arrives.

A good example of stating an AI achievement carefully is protein-structure prediction. For decades, predicting how a protein folds from its sequence was a major unsolved scientific problem. In 2020, an AI system (DeepMind's AlphaFold) performed dramatically better than previous methods at a long-running community assessment (CASP), and the results were widely regarded as a genuine scientific advance; the work later contributed to a 2024 Nobel Prize in Chemistry. This is a real, demonstrated capability on a well-defined benchmark, and it has accelerated biology. But note how the careful claim is bounded: it solved a specific, measurable problem well, which is very different from the sweeping headline that "AI has solved biology." The lesson is to celebrate real achievements while stating precisely what was and was not demonstrated — the essence of scientific thinking about technology.

• "An impressive demo proves general ability." It demonstrates specific, measured capabilities only.
• "Recent fast progress guarantees continued fast progress." Extrapolation from steep curves often fails.
• "AI is either all-powerful or all hype." The evidence supports a bounded, uncertain middle.
• "Experts agree on AI's future." Serious experts genuinely disagree about pace and impact.

Find a bold AI headline. Rewrite it to state precisely what was demonstrated, on what task, and what remains unproven or uncertain. Notice how much the claim shrinks.

Think Like a Maester: Ask of every AI claim: demonstrated on what, measured how, and how far does the evidence actually reach?

1. What is the difference between a demonstrated and a speculative AI capability?
2. What is the hype cycle?
3. Why is extrapolating rapid progress risky?
4. How should a scientific thinker treat deep uncertainty about AI's future?
5. How is the AlphaFold achievement stated carefully versus overstated?

Thinking scientifically about AI and automation means separating demonstrated capabilities — measured on real tasks, like AlphaFold's protein-folding advance — from speculation, and resisting the urge to extrapolate steep curves indefinitely. New technologies often follow a hype cycle, and the honest position about AI's longer future is genuine uncertainty. Avoid both hype and dismissal: look at what is measured, state claims precisely, and update as evidence arrives.

• Minds and Machines: The Science of Intelligence
• Making Decisions with Uncertain Science