Focus
- 1 Many technologists predict that by the 30s there will be an artificial intelligence that after two hours of talking to a human, people will believe it is a real person. They can assemble car models and pass other cognitive tests.
- 2 With the increasing likelihood of the emergence of superintelligence, people are beginning to reflect on the impact it will have on people and human society: it may lead to explosive economic growth, or it may lead to zero economic growth.
- 3 In a world where super-AI exists where material needs are met almost entirely by machines, people may abandon labor altogether and turn their attention to inherently enjoyable activities, including entertainment, games, and so on.
- 4 There are some needs that may only be met by humans, even in a world where artificial intelligence is super-powerful. Humans may still play a major role in three areas: areas where the boundaries between work and play are blurred, entertainment itself, and work in which humans still retain some advantage.
Tencent Technology News on May 28, in the movie "Wall-E" released in 2008, humans live in a fully automated future world. Various artificial intelligence robots are responsible for all the production labor, while people only need to sit on the sofa and watch TV. Scottish novelist Iain M. Banks' Culture series goes a step further by envisioning a world where artificial intelligence becomes powerful enough to become superintelligent, operating far beyond current imagination.
These sci-fi titles are a favorite of Amazon founder Jeff Bezos and Tesla CEO Elon Musk. In the world Banks describes, resources are extremely abundant, with AI "brains" directing much of production. Humans, on the other hand, turn to art, explore the vastness of the universe, and indulge in all kinds of entertainment.
Even now, such a future is still a bit far away. But with the explosion of OpenAI chatbot ChatGPT, the rapid development of generative artificial intelligence has prompted many people to take these ideas more seriously. On May 22, three OpenAI executives posted that "it is conceivable that within the next decade, AI systems will surpass the level of experts in most fields, and their output will be comparable to that of today's largest companies." ”
Last summer, forecasters on Metaculus, a favorite online prediction platform for many technologists, thought it would be until the early '40s that an AI would be produced that could make a human think it was a real person, capable enough to assemble a car model and pass a variety of other challenging cognitive tests after talking to a human for two hours.
But with the amazing breakthroughs made in the field of AI over the past year, forecasters on Metaculus now believe that this will become a reality in the early 30s. At the same time, there is no shortage of research funding, and 5 new generative AI "unicorns" (startups valued at $1 billion or more) have been born this year.
The road to artificial general intelligence (AGI) may take longer than expected. However, the growing likelihood of the emergence of super-powerful AI raises the question: when it arrives, what will happen to humanity? Will they become TV fans like they did in Robot Story? The following is a thought experiment guided by economic principles that provides partial answers to the above questions.
Work completely replaced?
Inevitably, such a thought experiment involves a number of rather bold assumptions:
First, we assume that AI will be benevolent, controllable, and very different from humans;
Second, we assume that human culture will not change so radically as a result of technological advances that people begin to like and even worship AI. Instead, we think of AI as a tool: a virtual, superintelligent, extremely cheap robot;
Third, we assume that restrictions on the widespread use of AI, such as energy constraints, will be addressed.
In 2019, three economists, Philippe Aghion, Ben Jones, and Chad Jones, simulated the impact of artificial intelligence. They found that if AI could automate all production, including the research process itself, and thus achieve self-improvement, it had the potential to fuel explosive economic growth.
An almost unlimited number of AIs can work together to solve any given problem, opening up enormous scientific possibilities. However, their simulations also raise an important caveat: economic growth will not be achieved if AI automates most, but not all, of the production or most, but not all, of the research process. As economists say, "It may not be the aspects of our good work that constrain growth, but those that are critical but difficult to improve." ”
This is explained by a point put forward by the late economist William Baumol. In a paper published in 1965, he and his colleague William Bowen studied the wages of workers in the performing arts. They point out that "the output per hour per hour of a violinist playing the Schubert Quartet in the Standard Hall is relatively fixed". While technological advances have increased productivity in other industries, the performing arts have not been affected. Because even if prices rise, people are still willing to spend money on art. Demand is "inelastic," and art accounts for a larger share of GDP, even dragging down overall growth.
Baumol's example points to a broader principle: if the areas where AI can be fully automated are just imperfect substitutes for areas where it cannot fully automate, and the demand for non-automated industries is difficult to change, then the share of the unproductive sector in GDP will rise, reducing overall growth.
Philip Agion, Ben Jones, and Chad Jones point out that, in fact, this has been the case for most of the past century. Technological developments have automated a large number of agricultural and manufacturing industries, driving down the relative price of their output. As a result, more income is being spent on industries such as education, health care, and entertainment, where productivity has not increased at the same level.
In a world where AI is more capable than the most talented humans, would Baumol's research still make sense? If AI doesn't have a concrete image, perhaps because advances in robotics lag behind computer technology, then the answer will be yes.
Many sectors of the economy, including construction and manufacturing, are clearly the real economy. There are countless forms of employment, including many healthcare industries, that require a combination of ingenuity and the ability to traverse the real world. As AI begins to dominate cognitive labor, the importance of these jobs will only become more important. Humans will work in the real world, perhaps under the guidance of AI "CEOs" or "professors."
But what if super-powerful AI also developed superhumanoid robots? Material needs will almost certainly be met by machines. Then, one might expect to give up labor, as in Robot Story. In fact, as early as 1930, another economist, John Maynard Keynes, wrote an article titled "Economic Possibilities for Our Children and Grandchildren."
In it, he speculates that in a century, people will work fewer than 15 hours a week. He predicts that the growth brought about by technology will solve "economic problems" and turn attention to activities that are inherently pleasant. Admittedly, Keynes's 15-hour workweek had not yet materialized, but higher wealth levels reduced working hours. The average workweek in rich countries has fallen from about 60 hours at the end of the 20th century to less than 40 hours today.
Still, there are some needs that perhaps only humans can meet, even in a world where AI is super-powerful. It's also worth noting that things that are inherently enjoyable may include work. Consider three areas where humans may still have a role to play: areas where the boundaries between work and play are blurred, entertainment itself, and jobs where humans still retain some strengths.
Entertainment and games
Let's start where the lines between work and play are blurred. While working hours in this field have decreased over the past 100 years, most of the decline occurred before the 80s of the 20th century, and more and more rich people are working longer hours than poor people. Keynes's article hints at an explanation for this strange development. He divides human desires into two types: one is absolute need, that is, we can feel them regardless of the situation of our companions; The other is relative needs, which we only feel when the satisfaction of those needs makes us feel superior to our peers.
Keynes may have underestimated the scale of this second type of demand. A cynic might say that the entire academic community is caught up in a situation where there is no obvious value to the world, but scholars are fiercely competing for their ingenuity. Economists would say that for many, work has become a "consumer good" that provides far more utility than the income it generates.
The game offers another hint that explains why people may not stop working altogether. Millions of people engage in recreational and sporting activities, vying for influence in activities that some consider irrelevant. Perhaps when robots surpass humans, interest in watching such games wanes. But evidence from sports where humans have been reduced to second-rate players suggests otherwise.
Since IBM's DeepBlue defeated chess grandmaster Garry Kasparov in a chess match in 1997, interest in the activity has grown. Other games "conquered" by AI, including Go and competitive video games, have witnessed a similar pattern. The number of video game players worldwide has almost doubled in the past 10 years, reaching 3.2 billion last year. Nowadays, more and more players make a living from competitive or live streaming.
Artificial intelligence may increase this interest. As Banks speculates, humans may be focused on "the things that really matter in life, such as sports, games, love, studying disappearing languages, savage societies and impossible problems, and climbing mountains without the help of seat belts," as others presumably want to see as well.
Finally, consider the areas where humans have an advantage in providing goods or services, which can be called "human premium". Even in the era of super-developed artificial intelligence, this premium can maintain the demand for labor. One of these aspects could be private information. As long as people are more willing than machines to share their secrets with other people, those who are trusted will selectively reveal this information to the world, ready to be absorbed by machines. Journalists might think that investigative journalists would still have jobs.
The "human premium" may also appear elsewhere. People value history, mythology and meaning. The origin of non-fungible tokens (NFTs) can be verified on the blockchain, and their valuation is often many times higher than images with the same pixels but different history. In areas such as care and therapy, people derive value from the limited time others spend with them, which adds to the sense of interaction. Man-made diamonds have the same molecular structure as natural diamonds, but they are traded at a discounted price. In the future, items with the label "artificial" may be particularly popular.
The human problem
If the "human premium" is large enough, it could even drag down economic growth. We can divide the economic sector into one with a greater human premium and one with a lower human premium: if humans are unwilling to replace human-made goods and services with machine-produced goods and services, the Baumol effect will only deepen. Estimated economic growth may even be zero. In fact, if extremely powerful AI can't accelerate growth, it means that the economy has shifted from the material level to gaming, politics, and other areas. In these areas, what people value most is interacting with others.
Perhaps one day, AI will produce entirely new goods and services that transcend human desire to please and interact with others. Such a contest would reveal something profound: To what extent are humans "social animals"?