The Future Has a Long Record of Missing Its Deadlines

People have spent centuries predicting the future with the confidence normally reserved for reading a menu. The usual mistake is not that they imagine change. Change is easy. The mistake is assuming that one visible change will continue in a straight line until it has replaced everything else, including several things that appear to be doing perfectly well.

The Internet Was Supposed to Make Paper Obsolete

The internet age produced a particularly durable forecast: paper would disappear. Offices would become paperless, contracts would become frictionless, and filing cabinets would go quietly into retirement. Instead, many workplaces added email, cloud storage, document-management systems, shared drives, chat applications, electronic signatures, and a printer that develops opinions whenever someone needs it urgently.

The prediction was not absurd. Digital communication did reduce some kinds of paper use and changed how documents move. But it underestimated institutions, habits, regulations, bad software, and the enduring human belief that an important form becomes more official after it has been printed, signed, scanned, emailed, and printed again.

This is the recurring problem with technology forecasts: they often identify a new capability but skip the difficult question of what people, organizations, laws, costs, and existing infrastructure will do with it. A technology can be real, useful, and transformative without producing the neat replacement story attached to its launch.

History Is Full of Futures That Arrived Sideways

Earlier eras made the same error using different machinery. New inventions were expected to eliminate old ones; faster transport was expected to dissolve distance; automation was expected to deliver abundant leisure. Some of those changes did happen in part. The surrounding world, however, had the bad manners to react.

• Railways changed travel and commerce, but did not make geography irrelevant.
• Radio, film, television, and the internet each inspired predictions that an earlier medium was about to expire. Earlier media mostly responded by becoming more specialized, more irritating, or both.
• Labor-saving technologies reduced effort in particular tasks while creating new tasks, new industries, and new ways to answer messages after dinner.
• Predictions of universal leisure consistently overlooked a basic fact: organizations are capable of converting saved time into additional work with remarkable efficiency.

Forecasts also tend to mistake the first use of an invention for its lasting use. A new device is demonstrated doing one thing, and observers assume that thing is the future. Sometimes they are right. More often, the important uses appear later, after millions of people have tried it for reasons that did not fit the demonstration.

AI Makes the Old Problem More Expensive

Artificial intelligence has revived the prediction industry at a useful scale. Some forecasts say AI will soon perform most knowledge work. Others say it will disappoint, plateau, or merely become another software feature with a monthly fee. Both camps can point to real evidence, which is inconvenient for anyone hoping to buy a certainty before lunch.

The hard part is that “what AI will do” is not one question. It is a bundle of different questions: what models will be technically capable of, how reliable they will be, how cheap they will become, which tasks employers will trust them with, what regulation permits, who controls the systems, how workers and customers respond, and what failures become unacceptable after they happen to someone important.

A system may be capable of drafting a report yet unsuitable for approving a loan, diagnosing a patient, negotiating a contract, or running a power grid without supervision. The difference is not simply intelligence. It is accountability. When software makes a consequential mistake, the question is not whether the output looked confident. It is who has to explain it on Tuesday morning.

Why We Keep Doing This

Predictions are useful because planning requires them. Governments need to build infrastructure, companies need to make investments, and people would like some indication of whether learning a new skill is a sensible use of their evenings. The trouble begins when a forecast is treated as a timetable rather than a conditional argument.

We also prefer a future with a single dominant story. “AI will replace everyone” and “AI is all hype” are easier to remember than “some tasks will be automated, others will be reorganized, new work will emerge, adoption will differ by industry, and the results will depend on prices, policy, incentives, and a series of awkward incidents.” The longer version is less satisfying, largely because it resembles reality.

The practical response is not to stop forecasting. It is to forecast with more humility and shorter horizons. Separate what a system can do in a controlled demonstration from what it can do reliably at scale. Ask what must change outside the technology for the prediction to come true. Keep several scenarios alive at once. And leave room for the possibility that the most important effect of AI will be something nobody was paid to predict.