There is a growing problem with artificial intelligence that its promoters would rather not discuss. AI is being sold as the technology that will transform the world, making businesses more productive, replacing expensive workers and delivering information faster than ever before. But what happens when the technology begins destroying the very things it needs to survive? What happens when AI drives writers out of business, replaces skilled workers, weakens education and reduces the spending power of ordinary people? There is a name for this sort of problem: an AI doom loop. It is a vicious circle in which the apparent success of AI creates the conditions for its own eventual failure.
Consider the internet, which has provided much of the material used to train modern AI systems. For decades, writers, journalists, researchers and publishers have produced enormous quantities of information, much of it freely available. Their work was supported through advertising, subscriptions and people visiting their websites. Now AI systems increasingly provide answers directly, often drawing upon the work of these publishers without requiring readers to visit the original sources. This may be convenient for users, but it threatens the businesses that produce the information in the first place. If publishers lose their readers and revenue, many will eventually reduce their output or close altogether. The AI companies may then discover that they have helped destroy an important source of the fresh information their systems require.
This is not merely a hypothetical concern. Internal Microsoft documents discussed in copyright litigation have reportedly acknowledged the danger of a doom loop in which AI damages the internet content industry upon which it depends. The problem is easy enough to understand. AI companies benefit from the work of human writers, but if their products undermine the ability of those writers to earn a living, fewer people may be willing or able to produce original material. The internet could become increasingly filled with AI-generated content, much of it recycling information produced by earlier AI systems. Instead of a growing storehouse of human knowledge, we could end up with an enormous echo chamber in which machines endlessly repeat and rearrange what other machines have already said.
The same problem is emerging in employment. Businesses are understandably attracted to technology that promises to reduce labour costs. If an AI system can perform work previously requiring several employees, management may see an opportunity to increase profits. But workers are also consumers. They buy houses, cars, food, clothing and countless other products. If large numbers of people lose their jobs or suffer substantial reductions in income, who will buy the goods and services that businesses produce? One company's decision to replace workers may make financial sense, but when thousands of companies pursue the same strategy, the wider economy could suffer. The savings achieved by individual businesses may be accompanied by a loss of customers across the economy.
There is a further twist. If businesses respond to falling demand by cutting more jobs and adopting more AI, the problem can feed upon itself. Workers lose income, consumer spending weakens, businesses cut costs and still more workers are displaced. Of course, new technologies have always eliminated some jobs while creating others. The Industrial Revolution provides many examples. But it would be foolish to assume that history guarantees a happy ending in every case. AI is being introduced across many occupations at once, including jobs that previously required considerable education and experience. Whether enough new employment will emerge to replace what is lost remains an open question.
Even the process of finding employment is becoming caught in an AI doom loop. Job applicants use AI to produce impressive applications and résumés, sometimes sending out hundreds with little individual effort. Employers, overwhelmed by applications, increasingly turn to AI systems to screen them. The result is a contest between machines, with one producing applications and another attempting to distinguish worthwhile candidates from the flood. Genuine human ability risks becoming harder to identify. The supposed solution to the problem creates more of the problem, encouraging still greater reliance upon automation.
Education presents an equally troubling possibility. Students can now use AI to write essays, summarise books and answer questions that previously required hours of reading and thought. Some teachers and lecturers are also using AI to prepare lessons, assignments and assessments. Imagine an education system in which machines increasingly produce the questions, machines provide the answers and human beings become spectators. Students may obtain qualifications without developing the intellectual abilities those qualifications are supposed to represent. Employers could then find themselves recruiting graduates who look impressive on paper but struggle when required to think independently. The danger is not simply cheating. It is the gradual weakening of the habits of reading, reasoning and sustained concentration upon which genuine education depends.
This raises a deeper issue. AI did not create the knowledge and intellectual achievements upon which it was trained. Human beings did. Generations of scientists, mathematicians, philosophers, historians, writers and artists produced the material that made these systems possible. If AI encourages future generations to abandon the difficult work of learning, investigating and creating, the intellectual foundations of the technology could gradually deteriorate. Machines might become increasingly capable of producing convincing answers while the number of people able to recognise their mistakes declines. That would be a peculiar sort of progress: a society surrounded by apparently intelligent machines but increasingly dependent upon them because its own intellectual abilities have weakened.
There is also the question of who benefits financially. The wealth generated by AI may flow disproportionately towards the companies controlling the technology, the enormous computing facilities and the energy supplies required to operate them. Meanwhile, many of the workers whose labour is replaced may receive little of the resulting prosperity. A society in which a small number of corporations become extraordinarily wealthy while large sections of the population experience economic insecurity is hardly an attractive vision of the future. Nor is it necessarily a stable economic arrangement. Businesses ultimately depend upon customers with money to spend, not merely upon machines capable of producing goods and services cheaply.
None of this means that AI must inevitably produce economic disaster. It has genuine uses, including assisting scientific research, improving productivity and helping people perform difficult tasks. The question is whether society will use these systems to strengthen human abilities or allow them to replace the activities upon which those abilities depend. There is a considerable difference between technology that helps a writer research an article and technology that destroys the writer's livelihood. There is a similar difference between helping students understand difficult subjects and allowing them to avoid learning altogether.
The great irony is that AI could become a victim of its own commercial success. In the rush to replace human labour, capture internet traffic and automate intellectual activity, the industry may undermine the human creativity, knowledge and economic prosperity that made its achievements possible. The immediate profits may be impressive, but profits earned by weakening the foundations of an economy and culture are not necessarily sustainable.
The real danger, therefore, is not the familiar science-fiction nightmare of machines suddenly becoming conscious and deciding to destroy humanity. It is something much more ordinary and potentially more plausible: human beings, pursuing short-term convenience and profit, gradually allowing machines to weaken the economic and intellectual foundations of their civilisation. No rebellion by intelligent robots is required. We could accomplish the damage ourselves, enthusiastically embracing each new technological advance while failing to notice what is being lost along the way.
The AI doom loop is ultimately a warning about human judgement. Technology does not have to hate humanity to damage human society. It merely has to be used in ways that reward immediate gains while shifting the longer-term costs onto everyone else. Unless those incentives change, the greatest threat from artificial intelligence may not be what the machines eventually become, but what human beings allow themselves to become through dependence upon them.
https://futurism.com/artificial-intelligence/ai-is-creating-doom-loops-all-over-the-economy