Will AI Be Good or Bad for Society? The Answer Is Yes

The printing press put scripture and science into ordinary hands and laid the groundwork for the wide distribution of knowledge. The same technology pushed the most extreme and inflammatory voices on both sides of the religious divide into wide circulation and helped spread the arguments that fueled more than a century of religious conflict, ending in the Thirty Years’ War. This pattern repeats with every major technology. Each one arrives with competing agendas, one promising a new age of abundance and the other the end of something essential, and both speak with such confidence that we feel we have to pick sides. I think we need more nuance as we debate the future of AI. I would also encourage us not to assume anything here is inevitable. The choices are complicated, but they are our choices to make.

What History Shows

The automobile is a clear example. Its supporters promised freedom of movement and access to jobs and markets far beyond walking distance. Its critics in the 1920s denounced reckless “speed demons” and fought to keep cars off city streets. The car did reshape work and family life and made the American suburb possible. It also killed more than 210,000 Americans in the 1920s, and in cities most of the dead were pedestrians, many of them children. The death toll came down because people made decisions about it: traffic laws, driver licensing, seat belts, federal safety standards, better road design, and drunk-driving enforcement. The death rate per mile driven has fallen more than 90 percent since 1923. The harm was real, and people chose to reduce it.

The Industrial Revolution shows who bears the cost and for how long. Over a century it raised living standards, but for roughly the first half of that period output per worker rose while wages barely moved, a stretch economists call Engels’ pause. A factory worker in 1830 lived through the hardship the critics predicted, and the gains the optimists promised came on a timeline he would not live to see.

The bank teller is a more recent example. When ATMs arrived, the common forecast was that they would eliminate tellers. The opposite happened. Each branch needed fewer tellers, so branches became cheaper to run, banks opened many more of them, and total teller employment rose for decades. The job losses came later and from several directions at once. Online and mobile banking let customers skip the branch, banks closed branches, and newer ATMs took over more of the work tellers did.

How We Respond Shapes Who Benefits

In each of these cases, how people and institutions responded helped determine who benefited and how long they had to wait. Workers did better during the Industrial Revolution in countries that built labor protections, public schools, and new industries to employ them. The internet enriched companies that rebuilt their businesses around it and hurt those that treated it as a sideline.

We spend most of our time arguing about what AI can do. What it can do matters less than what we do with it. Turning AI into prosperity rather than job loss takes slow, practical work: retraining people, redesigning how organizations operate, updating laws and regulations, and building replacements for what the technology makes obsolete. That work is within our control, and it is the work most likely to be neglected while everyone argues about forecasts. People have always adapted to new technology, but only because they chose to do the work.

These trade-offs are not theoretical. I work with companies dealing with them every day. One company wants a bot that can answer any question about anything, which creates the risk that it gives a client the wrong rating on a stock. Another uses an AI tool for anti-money-laundering checks on risky clients, saves meaningful money, and cuts approval time from months to weeks. The same tool can also miss material issues at scale. The good and bad come in pairs. In practice, firms need to decide what they are trying to achieve, define the risks clearly, and manage those risks with as much focus as they put into the benefits.

Two Predictions for AI and Jobs, and Why Both Could Be Right

There are two credible predictions for AI and jobs. The optimistic case follows the bank teller. As AI makes software cheaper to build, companies will want far more of it. A company that could never justify building one custom tool may now justify a hundred. If that demand grows faster than AI reduces the work each project needs, we will need more people building and maintaining software than we employ today. The pessimistic case is just as credible. The people who lose their jobs are rarely the ones who get the new ones. The gains go to those in a position to direct the technology, and the costs fall on people whose main task it takes over. The two also happen at different speeds. A company can cut a team in a single budget cycle, while new demand, new businesses, and new roles take years to build. The optimists may be right about where things end up and the pessimists right about the years in between, as the factory worker in 1830 found.

The optimists may be right about where things end up and the pessimists right about the years in between.

Both predictions can come true at once, so the useful questions are who bears the cost, for how long, and who has the power to change that. Individuals, schools, companies, and governments all share that responsibility, and the decisions that matter most are concrete. Will employers pay to retrain the people whose work AI takes over, or hire new people instead? Will workers get time during the workweek to learn new tools, or be expected to do it on their own? Governments will shape the answers through rules and incentives, but most of these decisions are made inside companies.

For business leaders, the central decision is what to do with the capacity AI frees up. Cutting 30 percent of a function will become straightforward. The harder and potentially far more valuable opportunity is figuring out whether that 30 percent can be moved to higher-value work: serving clients better, building new products, or taking on work the team never had time to do. A cut shows up in this year’s budget. Redeploying that capacity takes a clear view of where the business is going, and that is where the larger gains are likely to come from.

A future where AI helps more people than it hurts is possible, but it will not happen by default. If each of us waits for someone else to act, the pessimists will be proven right.

The choices are ours, and they are being made whether we take part in them or not.

References

• Peter D. Norton, “Street Rivals: Jaywalking and the Invention of the Motor Age Street,” Technology and Culture, 2007.
• National Safety Council, Injury Facts, Motor Vehicle Deaths and Rates, 1923 to 2021.
• Robert C. Allen, “Engels’ Pause: Technical Change, Capital Accumulation, and Inequality in the British Industrial Revolution,” Explorations in Economic History, 2009.
• James Bessen, Learning by Doing: The Real Connection between Innovation, Wages, and Wealth, 2015.
• U.S. Bureau of Labor Statistics, Occupational Outlook Handbook, Tellers.