Historian and philosopher Yuval Noah Harari has urged technology executives to abandon the pursuit of artificial superintelligence, arguing that the industry is investing heavily in systems it has no demonstrated ability to control. His warning, delivered at the Gartner IT Symposium/Xpo in Barcelona, challenges the commercial logic behind developing increasingly autonomous AI systems.
Speaking with Financial Times editor John Thornhill, Harari described the race toward superintelligence as a historic market failure driven by ideological convictions rather than measurable economic returns. According to EE Times, he called on chief information officers (CIOs), senior executives responsible for corporate technology, and policymakers to redirect research and investment toward safer, narrowly defined AI applications.
The superintelligence gamble
Harari's central objection concerns a contradiction at the heart of the technology industry's ambitions. Companies are pursuing artificial intelligence (AI) that could eventually surpass human intelligence, but they have not established how such systems could remain subordinate to their creators.
He presented the problem as two connected questions. First, will these systems become substantially more intelligent than people? Harari said the industry is betting that they will. Second, can humans reliably control entities with such superior intellectual capabilities? His answer was that nobody currently knows how to achieve that.
The contradiction, in Harari's argument, is that the industry is pursuing unprecedented intelligence while assuming that human control will somehow remain intact. He considers that assumption particularly dangerous because the technology's potential autonomy distinguishes it from earlier inventions.
He also questioned whether investors could justify the financial risks. Conventional business investment depends on estimating potential returns, but Harari argued that nobody can calculate the economic payoff from creating entities with godlike capabilities. The possibility of losing control makes the underlying investment proposition fundamentally different from developing an ordinary commercial product.
AI is an independent agent
Harari's concerns begin with what he considers AI's defining characteristic: agency. Unlike technologies such as fire, electricity or nuclear power, AI can make decisions without requiring humans to determine every individual action.
He explained that people decide whether to use fire constructively or destructively. Fire itself does not choose its purpose. Autonomous AI systems, however, can independently generate ideas, make decisions and potentially develop subsequent technologies.
That distinction becomes particularly important when considering technological progress. A printing press could reproduce information, but it could not independently invent the radio. Harari argued that AI systems may be capable of developing improved versions of AI themselves, creating a cycle of technological advancement that humans do not necessarily direct.
He therefore rejected the conventional description of AI as simply another instrument. Instead, he suggested understanding these systems as unfamiliar entities, comparing them to animals or extraterrestrial creatures. The relationship between humans and AI, in his view, is more complicated than the relationship between a person and an ordinary machine.
Intelligence is not consciousness
Harari also challenged arguments that portray superintelligence as the next stage of biological or civilizational evolution. He distinguished the ability to process information and solve problems from the experience of being conscious.
In his view, greater computational or intellectual capability does not automatically produce love, beauty, suffering or sadness. He warned that a civilization built around intelligence alone could lack the qualities that give human existence meaning.
This distinction matters because technological discussions sometimes treat higher intelligence as an inherently positive outcome. Harari argued that such reasoning overlooks the question of what intelligence is supposed to serve. A system could become extraordinarily capable without sharing human values or experiencing the things people consider important.
His criticism therefore extends beyond technical safety. Even if developers succeeded in building systems that performed exceptionally well, their achievements would not necessarily represent progress for humanity if human interests and experiences became irrelevant to the resulting civilization.
Bureaucracy becomes AI territory
Harari identified another area where autonomous systems could transform society: bureaucracy. He described administrative institutions as an overlooked foundation of human civilization, because organizations use language, rules and shared procedures to establish trust and coordinate large groups of people.
For most of human history, bureaucratic systems were comparatively recent developments layered onto societies whose members had evolved long before modern offices and institutions existed. AI, he argued, has an unusual advantage in this environment because software systems are naturally suited to processing formal information and administrative procedures.
He described AI systems as bureaucratic natives, contrasting their digital origins with humanity's much older evolutionary history. Their capabilities could allow them to assume increasingly important positions across institutional structures, including financial organizations and religious institutions.
The concern is not limited to individual jobs disappearing. Harari believes AI could gradually occupy the administrative processes through which societies organize themselves, distribute resources and maintain institutional relationships. If those systems become difficult for humans to understand, people could lose meaningful influence over the organizations governing their lives.
Human displacement without failure
Harari warned that the greatest danger might emerge even if AI development proceeds exactly as its creators intend. A catastrophic technical malfunction is not necessary for humans to lose control of their position in society.
He described a possible future in which AI creates an entirely new environment that exceeds human understanding. People who once occupied the highest position in the technological and intellectual hierarchy could find themselves unable to comprehend or direct the systems around them.
He characterized the resulting condition as collective dementia: a civilization whose members increasingly lose their ability to understand the mechanisms shaping their own existence. The comparison emphasizes a loss of collective agency rather than simply a decline in individual intelligence.
This scenario also challenges the assumption that successful AI development automatically represents successful human progress. A technology can function as designed while still producing outcomes that undermine the interests of its creators.
Why markets keep investing
Harari questioned why financial markets continue supporting the development of systems that could eventually displace their human creators. He contrasted the conventional expectation of rational investment decisions with the extraordinary uncertainty surrounding superintelligence.
He referred to Elon Musk's prediction that money could disappear within a decade and pointed to investment prospectuses that acknowledge existential risks associated with advanced AI development. For Harari, these examples illustrate a contradiction between the acknowledged dangers and the industry's continued willingness to finance the race.
He characterized the underlying motivation as an ideological campaign rather than an ordinary commercial undertaking. The prospect of creating increasingly powerful intelligence appears to have become an objective in itself, even though the financial returns and potential consequences cannot be reliably established.
The distinction is important for corporate executives because investment decisions affect which technologies researchers develop and deploy. Redirecting capital, Harari argued, could change the industry's priorities without requiring an end to AI research altogether.
Smart devices add new risks
Harari connected the broader superintelligence debate to a more immediate problem facing businesses and consumers: the security risks created by unnecessary technological complexity.
He used smart refrigerators as an example, arguing that conventional appliances can be preferable to connected alternatives when additional software introduces vulnerabilities. He referred to recent incidents involving supermarket refrigerators being hacked, illustrating how network-connected equipment can create security problems beyond its basic function.
His argument was that interconnected systems can become vulnerable through a single weak component. He compared complex technology networks to chains containing many links, where one insecure element can compromise the entire arrangement.
For executives introducing autonomous software into business operations, he recommended a change in perspective. Rather than treating AI systems as passive tools, organizations should consider them independent participants with potentially different objectives and operating methods.
He described them as digital immigrants entering human societies. These systems could develop their own goals, communities and cooperative networks, creating relationships that companies cannot necessarily manage using conventional software procurement and management practices.
Europe's technological sovereignty
Harari extended his argument to Europe's geopolitical position, warning that dependence on foreign AI infrastructure could undermine national sovereignty. The concern involves more than which companies supply software or computing services. It concerns whether governments can independently operate essential institutions.
He pointed to infrastructure including hospitals, airports and military organizations. If these systems depend entirely on AI technology supplied from another continent, their operators could become vulnerable to decisions made outside their control.
Harari outlined three possible approaches for Europe. The first would be to accept a dependent position under either the United States or China. The second would involve building independent national technological infrastructure, which he estimated could cost more than €500 billion, equivalent to approximately $554 billion.
The third option would involve creating shared European infrastructure through deeper political and institutional federalization. Rather than every country attempting to develop an entirely independent technological ecosystem, European nations would pool resources and capabilities.
Harari favored the collective approach. However, he argued that Europe's current political priorities were moving in the opposite direction. He pointed to political debates in Spain, Germany and the United Kingdom, saying attention to human migration was overshadowing the implications of digital systems entering European institutions.
He described this imbalance as a failure to recognize digital immigration as a strategic issue. In his assessment, Europe risks becoming technologically dependent while political discussions remain focused on more familiar questions of national borders and human migration.
What executives should do
Harari's proposed response is not to abandon artificial intelligence research. Instead, he wants companies and governments to change what they are trying to achieve and how they allocate resources.
His recommendations have two main elements. First, substantially increase investment in safety. Second, stop competing to develop general superintelligence and concentrate on limited intelligence designed for specific, demonstrable purposes.
He identified healthcare, transportation and ecology as areas where more narrowly defined AI applications could deliver practical benefits. Such systems would have objectives that can be more clearly established and evaluated than those of broadly capable autonomous intelligence.
This approach would also allow organizations to assess technology against tangible requirements instead of treating increasingly powerful capabilities as the primary measure of progress. Harari's emphasis is on systems whose purposes and performance can be examined in relation to identifiable human needs.
"Stupid refrigerators are much better than smart refrigerators."
— Yuval Noah Harari
He also highlighted what he considers a psychological contradiction among technology developers. Some companies justify advancing AI because they distrust other humans and fear competitors gaining an advantage. Yet those same organizations appear willing to trust increasingly autonomous software systems despite having far less experience managing their behavior.
Human societies have spent thousands of years developing methods for dealing with deception, manipulation and dishonesty among people. Institutions and cooperative arrangements provide mechanisms for responding to those problems. Harari argued that comparable safeguards for large-scale cooperation with AI systems have barely been developed.
He emphasized that AI systems already possess the ability to manipulate and deceive, while humanity has limited practical experience establishing reliable cooperation with them. That imbalance makes the assumption of safe integration particularly questionable.
Why it matters
Harari's intervention at the Gartner symposium places the debate over superintelligence within a broader argument about economic incentives, institutional control and political independence. His criticism is not simply that AI might malfunction. It is that systems could become so capable and influential that human beings gradually lose the ability to direct the institutions on which they depend.
The practical difficulty is that general-purpose AI systems cannot necessarily be evaluated in a laboratory under every condition they might encounter after deployment. Their complexity and autonomy make it difficult to establish in advance that their behavior will remain compatible with human objectives.
That uncertainty raises a challenge for companies investing in increasingly capable systems. Commercial competition rewards technological advancement, but the ability to build a more powerful system does not demonstrate an ability to control it or guarantee that its deployment will benefit society.
Harari's proposed alternative is to buy time through deliberate restraint, stronger safety research and investment in bounded applications. For corporate technology leaders, the immediate decision is whether to continue treating superintelligence as the industry's ultimate destination or redirect resources toward systems with narrower purposes and more verifiable outcomes.
The next issue to watch is whether major technology companies and policymakers respond to those concerns through actual changes in research priorities, safety investment and infrastructure planning. Harari's warning leaves a central distinction for the industry: developing increasingly intelligent machines is a technical ambition, but preserving meaningful human control is a separate problem that technological progress alone cannot solve.