Since the foundational work of Austrian economist Joseph Schumpeter, research has distinguished invention from innovation by its practical implementation. Whether an innovation spreads, however, depends partly on its social acceptance and how its uses are perceived. Without such acceptance, even a promising technology may remain confined to scientific experiments and investors’ speculation. To gain public support, an innovation needs a compelling narrative capable of generating interest, if not enthusiasm. Artificial intelligence illustrates this phenomenon: while its narrative is still taking shape, its social acceptability is increasingly contested.
In July 2026, Le Monde diplomatique declared that “The AI backlash has begun”. The initial sense of wonder generated by ChatGPT has increasingly given way to broader public unease about AI’s social consequences, from its effects on work, education and information to concerns about privacy, the environment and the growing power of technology companies. In this emerging narrative, AI appears as a “digital Moloch” whose costs are borne collectively while its benefits accrue to only a handful of oligarchs.
In May 2026, when former Google CEO Eric Schmidt addressed nearly 10,000 graduates at the University of Arizona, warning that artificial intelligence would transform every single profession, boos echoed through the crowd. Since then, commencement speeches focused on AI have faced similar jeers at several other US universities. This unrest aligns with data: a Harvard Kennedy School survey, cited by the Associated Press, reveals that roughly 70% of American students now view AI as a direct threat to their career prospects.
In September 2026, OpenAI’s rapid-fire claims of solving long-standing mathematical problems sparked deep division, with some celebrating a historic milestone while others warned that machine-generated solutions could undermine the processes through which new ideas emerge, young mathematicians are trained and mathematical knowledge itself develops.
Narratives that generate hostility
How did a technology promising unprecedented prosperity come to generate such hostility? Part of the answer lies in AI’s real or anticipated effects on society. Another lies in the narratives surrounding it.
Four broad narratives currently shape the discourse on artificial intelligence. They are neither mutually exclusive nor entirely coherent: the same company or executive may embrace several, sometimes contradictory, visions.
Narrative No. 1: ‘AI will save us… if it does not destroy us’
The first narrative presents AI as a civilisational gamble: it could usher in unprecedented abundance and scientific progress, but also trigger catastrophes or escape human control.
In September 2024, OpenAI CEO Sam Altman envisioned AI accelerating scientific discovery, solving complex problems and making capabilities once reserved for specialists widely accessible.
Anthropic CEO Dario Amodei developed a similar vision the following month. He imagined AI systems matching the abilities of the best researchers and compressing decades of scientific progress into a few years, transforming medicine, neuroscience and efforts to address climate change and poverty. The same models could facilitate cyberattacks, assist in the development of biological weapons or, as science fiction has warned for decades (from 2001: A Space Odyssey to The Matrix and The Terminator), become sufficiently autonomous to escape our control.
Elon Musk takes this logic much further
In July 2026, during an interview with The Economist, he compared our future situation to that of chimpanzees faced with humans: when the intelligence gap becomes too wide, it is difficult to imagine the less intelligent species remaining in control. According to Musk, there is a 10% to 20% risk that AI and robots will cause a catastrophe. Nevertheless, he believes that even if an off switch existed, it should not be used, because he is convinced that the most likely outcome will be “incredible abundance for everyone”.
AI and robotics would gradually make work optional, enable a generous universal income and eventually make money itself unnecessary. This is the narrative of the “happy apocalypse”: catastrophe is possible and humans will probably not remain in control of the systems, but acceleration must continue regardless, because it is both irresistible and potentially wonderful.
Narrative No. 2: ‘demystifying superintelligence’
Researcher Yann Le Cun, the former head of research at Meta, advocates a rationalist counter-narrative. In a February 2024 interview with Time magazine, he stressed that large language models have neither a robust understanding of the physical world nor genuine reasoning and planning abilities comparable to those of humans. He considers the scenario in which an artificial intelligence takes control of the world to be implausible. He also supports greater openness in AI models. In his view, just as a democracy needs a pluralistic press, it also needs a plurality of models and assistants.
LeCun has since left Meta to pursue research into “world models”, systems designed to understand and reason about the physical world. His approach shifts the debate from prophecy to engineering: the risks lie not necessarily in machines deciding to dominate humanity, but in how systems are designed, deployed and governed.
Narrative No. 3: ‘AI will augment us, not replace us’
Microsoft has promoted the seemingly more modest narrative of the “copilot”: generative AI assists humans rather than replacing them. By taking on routine cognitive tasks, it promises to free up creativity and improve productivity. The metaphor is skilful: in an aircraft, the copilot assists the pilot without taking away the controls.
Yet Microsoft also promotes AI agents capable of accessing organisational data and executing sequences of actions autonomously. The distinction matters: a copilot augments human work, whereas an agent can replace it. The narrative of augmentation thus risks obscuring questions about employment, skills, power and the distribution of productivity gains.
Narrative No. 4: ‘superintelligence for everyone’
On July 28, 2026, Meta’s CEO Mark Zuckerberg also entered the battle over AI narratives with an opinion piece published in The Wall Street Journal. He argued that the decisive question was no longer whether superintelligence would exist, but who would have access to it. His narrative is one of “personal superintelligence”: everyone could have an assistant that exceeded human capabilities and was tailored to their needs and values. This is not only a technological vision, but also an attempt to make the disruption caused by AI desirable again. Without abandoning the promise of superintelligence, this narrative seeks to make it more acceptable by promising to distribute it to everyone.
Yet democratising its use does not guarantee the democratisation of power. This narrative is consistent with the interests of a global platform such as Meta: billions of people may use a personal assistant while remaining dependent on a handful of companies that own the data centres, define the models and control the data required to train them.
When narratives generate their own consequences
These four narratives do not by themselves explain the apparent rejection of AI, which is also driven by concrete experiences: threats to employment, imposed transformations, errors, data centres, energy consumption and so forth. They do, however, give these concerns a coherent form. Moreover, the focus on the subjugation or extinction of humanity pushes immediate and very real issues into the background: the distribution of gains, copyright, accountability, human decision-making, the preservation of skills and environmental costs.
These narratives are also performative: they do not merely describe possible futures, but shape investment, regulation and competition. That became particularly visible in September 2026, when former Anthropic researcher Jacob Coxon resigned with a public warning that leading AI companies were racing towards systems they themselves believed could become catastrophically dangerous. His intervention rapidly drew support from other researchers, followed within days by calls from Anthropic’s Dario Amodei, OpenAI’s Sam Altman and Elon Musk for a coordinated slowing of frontier AI development.
The episode also exposed how difficult it is to separate safety concerns from commercial and geopolitical interests. President Donald Trump rejected calls for a slowdown, arguing that restraint would benefit China and that the US must maintain its lead in the AI race. Industry leaders face a mirror-image dilemma: while deceleration might mitigate certain risks, doing it unilaterally risks ceding ground to global competitors. Chinese firms—along with European players like France’s Mistral—have narrowed the performance gap, largely by developing open-source, compute-efficient models. Consequently, calls from tech giants for tighter regulatory controls can be interpreted less as altruism and more as a strategy to protect the market dominance of their own closed, resource-intensive systems.
This dynamic illustrates the core paradox of the existential-risk narrative. The leading AI corporations are framed as simultaneously too dangerous to leave unsupervised and too strategically vital to genuinely restrain. As a result, the policy debate becomes trapped in a false dichotomy: either allowing dominant firms to police themselves, or granting governments sweeping powers to shut down systems entirely. In both scenarios, citizens are relegated to passive spectators of a technological transformation presented as entirely inevitable.
Beyond the narratives: taking back the wheel
Each of the four narratives is marked by a more or less explicit desire to protect the interests of established players. The sums invested in AI are so astronomical that, however sincere the convictions may be, they necessarily remain compatible with the objectives of those advancing them. No industry narrative is entirely disinterested, and the vision promoted by a company simply reflects its strategic choices.
The problem may therefore not be to find the “right” narrative for artificial intelligence, but to stop presenting technical, economic and political choices as inevitable. First, we must stop personifying AI. These systems are designed, financed and deployed by human organisations.
Saying that AI wants, decides or threatens tends to conceal those who determine its objectives, choose its uses and benefit from it. We must also remember that the pace of development, the openness of models, their fields of application, the distribution of gains and the protections granted to individuals are not natural phenomena. They are decisions that must remain subject to political debate.
American or Chinese narratives
Finally, prophecies must give way to concrete questions. What do we want to automate? Where should human decision-making be preserved? Who controls the infrastructure and the data, bears responsibility for errors and guarantees access to remedies? How should the effects on employment, skills, the environment and inequality be measured? As Pope Leo XIV recalls in the encyclical Magnifica Humanitas, published in May 2026, innovations are never neutral: the question is who deploys them, under what rules and for whose benefit.
We must also recognise the highly US-centric nature of the four narratives presented here.
China’s official discourse, for example, offers a revealing contrast. It does not herald the arrival of a superintelligence that will either save humanity or bring about an apocalypse. Instead, it describes the methodical integration of AI into the real economy through a proactive industrial policy and under state control. This model raises other questions, particularly concerning political centralisation, surveillance and freedom of expression, but it reminds us that the American imagination of AI is not the only possible one.
AI should therefore be regarded as a powerful but governable technology whose benefits must be demonstrated rather than prophesied. Its proponents cannot indefinitely continue to present it as simultaneously uncontrollable, inevitable, and universally beneficial. Tech giants cannot market a technology as a “copilot” while claiming no-one is at the wheel.
- Professor of Strategy, CentraleSupélec, ESCP Business School
- This article first appeared in The Conversation
