Artificial intelligence has moved past its cycle of false starts and is now a permanent fixture of economic and social life, according to an editorial marking forty years of the journal AI & Society. Writing in the journal, Abbe Mowshowitz argues that the only real uncertainties are the speed of AI’s growth and how pervasive it will become. His essay concentrates on two of the most consequential societal challenges posed by intelligent artifacts: the effect of AI on employment, and AI’s role in the provisioning and interpretation of information. Both, he suggests, demand serious planning from governments and individuals alike, at a moment when speculation is abundant but reliable knowledge about the near and intermediate future remains scarce.
The employment challenge is framed by a long historical pattern. Just as spinners, weavers and farm workers were displaced by machines during and after the industrial revolution, generating considerable social unrest, intelligent machines are now positioned to replace or sideline human agency across a far broader range of tasks. Robots already abound in factories, and as their capabilities are enhanced by AI they are expected to take over more industrial roles. Intelligent algorithms, meanwhile, are likely to invade offices, homes and other venues where personal and business services are required, extending automation from manual labor into knowledge-intensive work such as teaching and preparing legal briefs.
The historical record offers concrete measures of how deep this displacement can run. In 1826, farming accounted for roughly 75 percent of the total American labor force; by 2026 that share had fallen to between 1.5 and 2 percent. Crucially, the growth of new industries that followed the mechanization of agriculture in the nineteenth century did not create enough jobs to compensate for the collapse in the demand for farm labor. Manufacturing tells a similar story: from 1950 to 2026, employment in manufacturing fell from 31 percent to 7.92 percent of all non-farm employment in the United States, a decline that proceeded fairly steadily at about three percentage points per decade, mostly attributable to advances in technology, beginning with the arrival of computers in the 1950s and continuing with the more recent spread of robots on the factory floor.
Past policy responses to such transitions suggest a possible playbook. Societies have previously met these challenges by keeping children in school longer, reducing the number of hours worked per year, and making paid leave more generous. Whether those instruments will suffice against AI is an open question. The economics of the transition will be governed largely by tradeoffs between the cost of human labor and the cost of machine systems capable of replacing it; with the aid of intelligent artifacts, substitution could prove cost-effective in virtually any workplace, whether the task involves manual labor or expert knowledge work. Assuming the population does not decline enough to absorb the reduced demand for jobs, policies will have to be adopted that compensate for decreased opportunities, and it may prove necessary to lean more heavily on transfer payments to keep people afloat in an altered economy.
A more radical possibility lurks further out: consumption could one day replace work as the principal means of sharing the fruits of production. If and when that occurs, the impact of AI on jobs might cease to be an issue at all. But the editorial is blunt about the interim, noting that such a transformation could take many decades or even generations, offering little help to workers facing displacement now. In the meantime, the author calls for federal, state and local governments to join together in commissioning detailed studies of AI’s likely effects on work requirements across every sector of the economy. Only such detailed knowledge, he argues, can support measures tailored to forestall negative effects in each sector, whether that means retooling education and training to match newly created jobs or compensating for a significant nationwide reduction in labor requirements.
The stakes of delay are illustrated by a pointed historical warning. If timely action on the consequences of AI suffers the same fate as action, or inaction, on Social Security, the stage might be set for serious social upheaval. Individuals, the editorial stresses, should not wait to be saved by government programs. Getting a well-paying job with a high school or even a college degree is no longer assured, which means either that people must find another route to a foothold in the new economy or that institutions providing education from kindergarten to university must radically adapt their programs and courses. Concern about the high cost of university education and the uncertain job market for graduates is already visible in behavior: with surging demand for electric power for AI data centers and massive construction projects to build them, young people are beginning to conclude that a blue-collar vocational path might beat investing huge sums in a degree. There may well be a substantial short-term increase in demand for electricians, plumbers and construction workers, but the editorial cautions that these choices alone cannot deal with possible widespread losses of employment opportunities posed by AI.
The second critical challenge concerns information itself. It has become commonplace for people to use chatbots powered by large language models to answer questions, apply for jobs, write papers, obtain financial advice, vet candidates for public office and assist in countless other tasks of modern life. Yet assessing the reliability and accuracy of information gleaned from websites retrieved by search engines has been difficult since the early days of the World Wide Web. Checking the veracity of a chatbot’s output is even harder for the average person. In essence, a user either accepts or rejects what ChatGPT, Claude, Gemini, Copilot, Perplexity and similar systems provide, often with no independent means of verification.
Dominant actors in the AI arena claim to take safety seriously and have invested resources in identifying issues and securing their products. But the editorial insists that the road to safety is a two-way street. Software engineers and AI specialists can only do so much by placing guardrails in software. They can attempt to build in safeguards against hallucinatory behavior, minimize incorrect or misleading answers, and forestall potentially destructive behavior by AI agents, yet their efforts may not succeed. Users are well advised to keep in mind that assertions of fact may be wrong and that qualifying conditions may be missing. Nor is unintentional error the only risk: an AI system may be designed to misinform or deceive, or may be riddled with biases, whether inadvertent or intended.
The remedy proposed draws on an established safety culture from other industries. Just as natural gas companies issue regular safety bulletins, and few would argue against instruction and training for the safe use of complex and potentially dangerous products like automobiles and construction equipment, the same logic should apply to information produced by AI systems. People need to be taught the dangers of relying uncritically on such output, much as earlier generations were taught to be skeptical of deceptive forms of propaganda. Alongside user education, the editorial highlights a structural concern: developing and deploying large language models requires enormous capital for components and production, and the economics of LLMs favor monopoly or oligopoly, as was the case for search engine companies. Only a handful of multibillion-dollar firms specialize in AI, including OpenAI and Anthropic, alongside hyperscalers such as Google, Amazon and Microsoft, and their enormous power, if unchecked, could result in the deployment of systems that feed false or misleading information to the public. Past experience with excessive concentration in the transportation, communication and electricity industries, which led to the formation of regulated public utilities, offers useful lessons for crafting balanced regulation.
The editorial distills the challenge for governments into three parts: assisting in an orderly transformation of work, crafting balanced regulation of the monopolies and near-monopolies in the AI industry, and supporting educational programs that enlighten the public about the dangers of uncritically accepting AI-provided information. For individuals, the prescription is twofold: plan for careers with an open mind, free from prejudice arising from today’s status distinctions, and adopt a critical approach to information supplied by AI systems. Forty years after the journal’s founding, the message is that the age of intelligent artifacts is no longer a forecast to debate but a condition to manage, and that the window for building the knowledge and institutions needed to manage it is open now.
Subject of Research: Societal challenges of artificial intelligence for employment and information reliability
Article Title: Meeting critical challenges in the age of intelligent artifacts
Article References: Mowshowitz, A. (2026). Meeting critical challenges in the age of intelligent artifacts. AI & SOCIETY. https://doi.org/10.1007/s00146-026-03389-9
Image Credits: AI Generated
DOI: 10.1007/s00146-026-03389-9
Keywords: artificial intelligence, employment, automation, large language models, chatbots, information reliability, AI safety, labor economics, AI regulation, monopoly, education policy, AI & Society
News Source: Denise Maddox. (October 25, 2026). As Intelligent Machines Reshape Work and Information, Expert Urges Planning Now. Scienmag.
