Rise of the A.I. Manager
In 1933, during the depths of the Great Depression, twenty-five percent of job-seeking Americans could not find work. But if we take projections for the coming century seriously, twenty-five percent unemployment is nothing compared to what is to come. Academics and industry leaders like McKinsey & Company estimate that we could automate forty-five percent of work activities using already demonstrated technology; an Oxford study goes as high as eighty percent.
Today, computational power limits our application of artificial intelligence technology—but this will not be true for much longer. If computing power continues to increase at its current rate, it will reach the level of a human brain by 2030. By 2060, it will equal the combined processing power of every human on Earth. The processing power of multiple human brains could then be applied at the cost of a MacBook.
Automation technologies historically have increased efficiency and promoted economic growth. Proponents of increased automation contend that more automation would free up workers to focus on less mundane tasks, potentially leading to greater equality. But according to the Swiss bank UBS, that will not be the case. Its 2016 report predicts that not only will inequality increase between developed and developing countries, it will also increase within countries themselves as the rise of automation squeezes out unskilled and semiskilled work.
With many of the world’s leading technology and financial companies telling us that the fourth industrial revolution is upon us, it’s time to reevaluate. We need a new economic model.
What if rather than corporations and market forces, an artificial super-intelligence (A.S.I.) controlled the economy?
Imagine a U.S. in which eighty percent of the economy was automated. Robot doctors and nurses fill the hospital, bots make investment decisions, and self-driving cars, ships, and trains cover all transportation. When the economy reaches this high level of automation, the companies that control the automated workforce cede control of the workforce to a decentralized A.S.I., which efficiently deploys resources to meet the basic needs of the population, such as housing, food, clothing, wifi, electronics, and healthcare. Each citizen receives a stipend, consisting of an adequate amount of discretionary income for the purchase of excess and non-automated products. The A.S.I. analyzes mass economic data to meet consumer demand and set the stipend.
Citizens use their discretionary income to purchase luxury goods or services not rendered by the A.S.I., which human entrepreneurs supply. In that sector of the economy, market forces rule. But when it comes to necessities, the A.S.I. controls product type, quantity, and distribution.
Take the above scenario—in many respects not complex enough to be realistic, but suggestive of increasingly relevant issues—as given. It raises timely questions: Can an A.I. make the judgement calls involved in managing a company? Can a non-human entity be an effective executive?
Ceding control of an automated workforce to non-humans seems risky at best and possibly disastrous at worst. But some organizations, such as Bridgewater Associates, one of the world’s largest hedge funds, are already attempting to build automated managers on team and company levels. If an A.I. can manage one company and the millions of automated bots that are already making trades, then it is possible that an A.S.I. could manage a nation’s economy by managing every company in that nation.
This structure raises a political question: If an A.S.I. controlled eighty percent of the economy, who would control the A.S.I.?
There is a way for the A.S.I. to operate independently of government control. Instead of being controlled by a single entity, it could instead operate as a Decentralized Autonomous Organization (DAO). Its data storage and computation would be spread out across tens of thousands of computers across the world, instead of operating from a central location. Cryptocurrencies such as DashCoin are currently experimenting with DAO organization structure. As of October 2017, DashCoin is collectively worth two billion dollars.
Under a DAO model, collectives of developers could decide on changes to the A.S.I. and vote on enacting those changes like an open source project. Any citizen who has completed the free training and certification process could join the development group. The A.S.I.’s behavior and rule-setting criteria—its “A.I. constitution”—could only be changed if enough developers voted in favor, much like the ratification process of a constitutional amendment. This way, a powerful individual, government, or corporation would not be able to enact major changes without convincing a huge number of developers. The A.S.I. would work for the benefit of the majority.
Once the A.S.I. took control of the automated workforce, it would monitor and process an astronomical amount of information every millisecond to ensure that production was efficient, demand predictions was optimal, and human needs were met. Bots would produce goods and services twenty-four hours a day at a rate determined by the A.S.I.’s demand predictions. If a mismatch in market occurred, the A.S.I. would learn from it and update its prediction models.
But how can we ensure that innovation continues in a world where all essential needs are met and work is optional? How will the A.S.I. account for emerging human needs and wants?
We would need a secondary free market. No matter how effective the A.S.I. might be, there likely would be emerging human demands that it would not recognize immediately. To make new discoveries and products and to satisfy the passionate and ambitious, we need innovation. And people need incentives to innovate.
A standard disposable income stipend, calculated and distributed by the A.S.I., could solve this problem. It would be similar to a Universal Basic Income: all citizens would receive a regular, uniform sum of money. In this case, it would serve as the secondary free market’s currency, not money to live on, because survival needs would be satisfied already.
Individuals could trade with human entrepreneurs or save the disposable income to purchase excess and luxury goods. Successful human entrepreneurs could then use their revenue to build automated production lines to meet increased demand for their new products. If a new product met the criteria to qualify as an essential, the A.S.I. would force “buyout” control from the human entrepreneur, similar to an acquisition today. If the production line was not already automated or could be further automated, the A.S.I. would do so.
Technologically, these advancements may be possible in the foreseeable future. Politics will complicate actualization, of course, especially given today’s political climate and right-leaning trends. But if the conjectural future of the A.S.I.-controlled economy were to become reality, it would be a triumph of science and a realization of the best manifestations of socialism.
Artificial intelligence, automation, and computing are only in their infancy. The political pendulum will keep swinging. In the meantime, let’s enjoy the brave new world where a 6-axis industrial barista can serve us coffee without spelling our names wrong, and watch for the moment our economy needs an update.
Jason Huishen Lu ‘20 is a sophomore in Saybrook College.