Is hype about Artificial Intelligence justified? Cristobal Garibay-Petersen, Marta Lorimer and Bayar Menzat argue public discourse around AI has gone beyond what its technical specifications warrant, limiting the space for democratic engagement with and control of the technology.

Since the release of ChatGPT in November 2022, speculation about how a new era of Artificial Intelligence (AI) would change society has been rife. From evangelising claims that AI would save the world, to catastrophist warnings that it could destroy it, to admonishments to remain sceptical, positions on the AI-debate truly run the gamut. In a recent paper, we look at these debates and ask: how is AI being presented in public discourse, and to what political effects?

Testing four claims about AI

Four sets of discourses, made primarily by influential tech commentators, interest us: claims that AI somewhat resembles human intelligence and affects “humanity”; ideas surrounding AI having “agency”; statements about the economic implications of AI; and confident declarations about the need to act “urgently” in response to AI developments.

These claims we find problematic for two reasons: first, they are inaccurate from a technological point of view; second, their wide-ranging implications for democratic governance are being obscured.

Take claims that AI systems are getting closer to matching human capabilities in certain areas and even surpassing them in others. From a technical standpoint, the idea that AI resembles human intelligence in a meaningful way is spurious. Contrary to popular belief, models like ChatGPT do not “learn” continuously.

Because they operate with fixed parameters (weights), interactions do not leave any trace on the underlying model. This is very different from biological learning processes, which involve continuous, dynamic interactions with the environment that produce durable changes in neural connectivity.

A false equivalence

From a critical analytical perspective, the idea that human and machine intelligence can and should be compared is equally problematic because it creates a false equivalence between human decision-making and computer decision-making. This false equivalence can become the basis for a series of dubious practices, such as the replacement of (fallible) human judgement with (allegedly superior, but no less fallible) machine judgement.

Claims that AI is an issue that affects “humanity” are similarly politically problematic because they conceal the way in which different groups are affected by AI. Some jobs are more at risk of others, and some groups (usually, those already marginalised) are more likely to be negatively affected by AI.

Failing to recognise the differential impacts of AI hampers the ability to mobilise these groups politically. Identifying a constituency is essential for the articulation of political problems and solutions. Suggesting that “humanity” is in this together negates the very existence of the kind of divisions one could mobilise around.

Problems of agency

One should also be wary of claims that AI technologies are about to acquire the ability to make autonomous decisions. These claims frequently present conflicting views of the role of humans in the development of AI.

On the one hand, AI is frequently presented as self-developing and at constant risk of escaping human control. On the other hand, the development of AI also becomes an opportunity to reassert human agency and dictate its future. Ascribing agential qualities to AI is questionable from a technological perspective because it overestimates its ability to produce truly novel content or generalise outside of training data.

From a political perspective, this complex mix of machine agency and human reaction is also troublesome. The attribution of agency to AI conceals the ways in which humans are implicated in its development.

At the same time, presenting human agency as purely reactive promotes a form of politics that leaves limited scope for political choice. Because it takes the political agenda as externally set, it removes it from the realm of democratic deliberation: the technology is developing in a certain direction and politics, policymakers and citizens need to accept it and respond to it in pre-determined terms.

AI and the economy

Then there are the economic claims that are made about AI. Discourses on AI have tended to portray the development of the technology as essentially linked to a specific conception of what economic reality is and ought to be. These assumptions are problematic to the extent that they uncritically reflect key pillars of a certain form of liberal capitalism and its depoliticising tendencies.

For example, the idea that competition-driven markets should act as the benchmark against which success, legality or rightfulness ought to be measured reduces the scope to consider what other criteria one might want to keep in mind when developing new technologies. Likewise, that a very specific conception of what constitutes “good” or “responsible” economic practice is set in stone is problematic to the extent that economic practice and theory, like everything else, are subject to change over time.

The myth of urgency

The final, and perhaps most important, set of claims that concerns us are the temporal assumptions that are made about AI. Public discourse about AI points towards a sense of acceleration, urgency and temporal linearity. AI is presented as a force in motion, which is already changing societies and will do so even more radically in the future. The claim of a sudden acceleration prepares the ground for demands to act urgently to speed it along or to halt it in its tracks.

Many of these claims are not fully warranted from a technical standpoint because they by and large overstate the speed at which AI is developing and create narratives of certainty where scientific knowledge is significantly less certain.

For example, many claims concerning AI’s ability to improve seem to be based on the notion that augmenting the size of AI models (“scaling”) might address various extant challenges in the field. Reality, however, presents a less optimistic picture, particularly given the huge computational resources needed for the kind of scaling that is envisaged.

Instead, the confident timelines and perceived urgency for action seem to be driven by political – ideological, even – attempts to move the present towards the future they predict. This future, for better or for worse, is one with AI.

AI and democracy

This occupation of the future, and the appeals to urgency that go with it, should worry us. The reduction of the future to one where AI will take over suggests that only one future is possible. From a democratic standpoint, this reduces the space to choose between political alternatives. The appeal to urgency further reduces the space to imagine, let alone bring about, alternative futures.

Framing something as a problem in need of an urgent response reduces the space for deliberative, “slow” decision processes and for democratic choice itself. AI discourses pointing to the need to urgently address the problems or opportunities that AI causes reduce the space to collectively define what the problem that needs addressing is, how it should be addressed and who is best placed to address it.

We should, therefore, remain critical of AI-hype. AI has become more than what its technical specification should warrant. It is now a mobilising political concept: an idea that is used to push political development in a certain, not necessarily understood and not necessarily chosen, direction. Unfortunately, this is being done in a way that limits the space for democratic engagement with, and control of, the technology itself.

For more information, see the authors’ recent paper in Big Data & Society.

Note: This article gives the views of the authors, not the position of LSE European Politics or the London School of Economics.

Image credit: naimurrahman21 provided by Shutterstock.