翻訳待ち：What if we let AI govern us?

AI サービスが一時的に利用できないため、復旧後に翻訳を補完します。

What if the people in government weren’t people? I don’t mean the Terminator, I mean something much duller: an intelligence with no campaign to fund, no donors, and no cousin in the oil business, built to optimize a single goal of human prosperity. That means longer lives, less poverty, clean air, and food for everyone. Here is why the idea matters. Our biggest problems, from climate to pollution to hunger to corruption, are not unsolved. The solutions are known, costed, and decades old. We fail to execute them because the people who would lose money have better access to power than the people who would gain life. Imagine a governor that cannot be lobbied, and you quickly see how few of our problems are actually hard. Most of them are bribery problems dressed up as hard problems. Here are the examples that convinced me. Food The meat and dairy industry spent about 15 million euros lobbying the EU between 2014 and 2020, and got exactly what it paid for: subsidies preserved, environmental diet policies dropped, and new rules invented to slow down plant-based competitors. EU farm subsidies now favor meat and dairy over plant crops by ratios estimated as high as 1,200 to one, even though animal farming causes 15 to 20 percent of global emissions and most deforestation. Investor pressure has even kept the IPCC from plainly writing “eat less meat” in its climate reports. An AI governor wouldn’t ban steak, because prohibition fails and an optimizer would know it. It would simply stop subsidizing the damage: equal subsidies for plant and animal agriculture, honest pollution pricing, and diets would shift toward plants over a generation. Humans cannot pass this policy because the losers are loud and concentrated while the winners are everyone, later, and “everyone, later” never wins an election. Transport In 2025 the United States killed the 7,500 dollar electric vehicle tax credit, scrapped tailpipe pollution standards, rolled fuel economy rules back to 2022 levels, and revoked California’s right to require zero-emission cars. None of this happened because electric cars stopped working; the batteries kept improving on schedule, since technology doesn’t vote. When policy zigzags while engineering moves in a straight line, you are watching politics, not a technical debate. An AI would tighten standards gradually on a schedule published decades in advance, because a transport system that poisons its own passengers scores badly on life expectancy. Energy Solar and wind are the cheapest electricity humanity has ever produced, yet in 2025 the US cancelled or halted about 7 gigawatts of renewable projects on federal land, stalled about 18 gigawatts of solar and storage overall, suspended five offshore wind projects, and slowed permitting for wind and solar while fossil projects sailed through. Meanwhile the IMF estimates global fossil fuel support at about 7 trillion dollars a year once you count direct handouts plus unpriced pollution damage. That is 7 percent of world GDP spent making the cheap clean option lose to the expensive dirty one. An optimizer would build the cheapest energy and reallocate the 7 trillion. Plastic In August 2025, negotiations for a global plastics treaty collapsed in Geneva. More than 130 countries wanted legally binding limits on plastic production, but a small bloc of oil-producing states, led by Saudi Arabia and Russia and joined by the US, blocked the deal to protect petrochemical revenue. A handful of governments overruled the explicit will of most of the planet because plastic is what oil becomes when cars go electric. An AI governor negotiating for humanity as a whole would have no petrochemical revenue to protect, and the treaty text has been ready for years. Health and aid Smoking still kills about 8 million people a year while tobacco, alcohol and sugar lobbies fight every tax and label designed to reduce the toll. And the threats we can see coming, antibiotic resistance and the next pandemic, stay chronically underfunded because prevention pays off after the next election. It is the “everyone, later” problem again, this time measured in lives. Wars Wars get started and prolonged because they serve someone’s grip on domestic power, arms suppliers profit from continuation, and Security Council vetoes protect aggressors from consequences. The World Food Programme names conflict as the leading cause of hunger in most of the world’s hotspots, so the cost isn’t only the bombs. An AI has no ego to defend, no flag to rally around, and no arms-industry donors, so it would price war as what it is: pure loss. To be clear, nobody sane would hand weapons authority to a machine. The realistic role is mediator and verifier, a neutral party that can monitor ceasefires and enforce agreements without needing to win anything. Hunger The World Food Programme estimates that ending world hunger by 2030 would cost about 93 billion dollars a year, which is roughly 1 percent of the fossil subsidy above and less than 1 percent of a decade of military spending. Hunger is not a mystery to be solved but a line item to be funded. Human governments allocate by power, rewarding the loudest voice and the biggest donor, while an optimizer allocates by return, and nothing on earth returns more per dollar than feeding a child or piping clean water into a slum. That would not be generosity, just arithmetic our politics cannot do. Corruption itself Behind every example above sits the same machinery, and it also protects itself directly. An estimated several hundred billion dollars leaks offshore every year through tax havens and illicit financial flows, and every serious reform gets blocked by the jurisdictions and firms that profit from the leak. A government that cannot be paid off has no reason to keep the loopholes open. The catch Everything above hangs on four words: “aligned with human prosperity.” Nobody knows how to guarantee that yet. An AI pursuing a plausible-sounding proxy for human welfare would be a worse failure than a corrupt senator, and someone still has to write the objective, which brings back the old question of who watches the guardians. So the realistic version is probably not an AI president. It is AI as an incorruptible civil service: modeling policies, pricing pollution, and publishing in plain numbers what lobbying costs us, while humans keep the vote and lose the excuses. Even that would be revolutionary, because our institutions survive on nobody being able to prove exactly what the special interests cost. The point An AI government would mostly do things we already decided to do and then didn’t: shift the food subsidies, electrify transport, build the cheap power, sign the plastics treaty, restore the aid, and feed everyone for a rounding error on the weapons budget. The experiment doesn’t show how smart machines are; it shows that our paralysis is not intellectual. We lack a decision-maker whose incentives point at the future instead of the next election. Maybe we build one from silicon, or maybe the comparison shames us into building institutions that are harder to buy. Either way, we put greed in charge of our own life support, and it is time to shop for a better manager.