Agents of change—such as unions, consumers, journalists, judges or tech engineers—need the capacity to effect change: they need strategic leverage to have an impact on the objectives that are being maximized by Artificial Intelligence. But agents of change also need values and beliefs that make them want to effect change, in a direction that is beneficial for society at large. Once we start to consider values and beliefs, ideology comes into play. Effective collective governance of AI is undermined by ideological obfuscation of the issues at stake.
In everyday language, we often use the term ideology rather casually to describe the beliefs of people we disagree with. In political theory, ideology is a hotly contested term, but it usually means something more specific than suggested by everyday usage. Correspondingly, I use the term to mean a coherent set of beliefs or arguments that ultimately serves certain interests, and that justifies the preservation of an unequal status quo (without making this explicit). Ideology can exist without anybody actively trying to deceive us. For an ideology to persist, it suffices if certain stories that are more convenient for certain social groups are repeated more often, while stories that are less convenient are repeated less often.
Ideology, in this sense, often does one or more of the following three things:
First, it represents the interests of particular groups as those of society at large. This is often combined with warnings about an external enemy or outsider group. Differences and conflicts of interest among insiders are ignored, and conflicts with outsiders are played up: “We have to stand together to defend against our common enemies.”
Second, ideology represents contingent choices as objective necessities. The very possibility of dissent is thereby forestalled, since it is foolish and dangerous to resist what cannot be otherwise: “There is no alternative” instead of “another world is possible.”
And third, ideology represents social relationships as material or technological ones. This happens especially in the context of discussions of the economy. For example, “Wage inequality is due to technical progress, which cannot be stopped.”
By making arguments about the general interest, the absence of alternatives, and the technical nature of the issues at hand, ideology has the function of preventing change and preserving the status quo. This benefits those who are currently in power. How does this apply to current stories around AI? Can we think about them as part of an ideology?
The Human v. Machine Story
Consider the story of human versus machine. In this story, there is no society. Society is effectively identified with that one human, and all conflicts of interest are effaced. AI becomes the ideal outside enemy, outside humanity at large, which allows humanity to become one homogeneous group. Since there are no conflicts of interest in this story, the interests of humanity at large must be the same as the interests of, say, tech CEOs.
In the existential risk version of this story, where AI becomes exponentially self-improving, this external enemy is a technical necessity. It is not the product of human choices or social conflicts anymore. It is a self-improving force of nature.
Fights of human versus machine, and intelligence explosions, are favorite stories of science fiction authors and of futurologists. Similar ideological functions can be served by some of the more sober stories of computer science and AI ethics—if all problems of AI are interpreted as optimization errors. In these optimization error stories, too, social problems are conceived as technological ones. Discrimination is conceived as a failure to predict accurately or to maximize profits. Value alignment is conceived as a wrongly specified reward. Privacy is conceived as a problem of getting individuals to share their data voluntarily. In each case, technical fixes are proposed for issues that ultimately require social negotiation.
A flip side of the ideological argument that the problems of AI require technical fixes is the ideological argument that only technical experts can provide these fixes. More broadly, a common story holds that AI is very complicated and hard to understand, and only the experts of the tech industry can possibly address the issues of AI. This story forestalls public debate and democratic decision-making. It concentrates decision-making power in the hands of an interested party, and it prevents the development of alternative stories by preventing debate. As this book aims to show, in contrast to this story, the basic ideas of AI are not that complicated and can be made broadly accessible.
A variant of the ideological argument about the technical inevitability of particular choices regarding the course of AI is the ideological argument about the political inevitability of these choices. This argument often invokes great-power rivalries. In the United States it might be “If we don’t do it, then China will.” In Europe it might be “If we don’t catch up, the American tech companies will keep dominating.” Functionally, this political argument, just like the technical argument, implies that there is no alternative.
A structuring set of ideas—an ideology, if you will, or a story—is certainly necessary to understand and interpret the world. But we do not have to stick to the old structuring ideas and stories that I just discussed. We can instead develop alternative stories. A good alternative story makes explicit that there are conflicts of interest and of values in society. It makes explicit that another world is possible—there are social choices to be made, and technology is not fate. And it empowers all of us to collectively make these choices that impact us. My book tries to contribute such an alternative story. An alternative story alone will not change the world, but an alternative story might be necessary to change the world.
—
Reprinted with permission from ”The Means of Prediction: How AI Really Works (and Who Benefits)” by Maximillian Kasy, published by the University of Chicago Press. © 2025 by Maximillian Kasy. All rights reserved.