Title

Josh Simons. 2023. Algorithms for the People – Democracy in the Age of AI. Princeton, NJ: Princeton University Press. 303 pages. ISBN 9780691244006

Josh Simons has divided his book into ten parts starting with the obligatory introduction. This is followed by The Politics of Machine Learning. Simons prefers machine learning over the rather nebulous term artificial intelligence (AI) that still has questions like is artificial intelligence reality intelligent? hanging over its head. Chapters two, three and four are about fairness, discrimination, and political equality, respectively. On the IT side, Chapter five is about Simons’ prime examples of Facebook and Google. Chapter six is about Infrastructural Power. Chapters seven and eight are about Democratic Utilities and Regulating for Democracy, respectively. A conclusion ends this readable and well-argued book.

One the whole, about 2/3 of the book illuminates the pathologies of AI or, to use the author’s term, “machine learning”, while the other 1/3 of the book is about Democracy for the People. Among the many known and not so known cases, Simons mentions a case of child abuse and the use of machine learning to send the right officers to the right house for the right reasons. This could – once the decision-making power is handed over to a machine – go horribly wrong.

With these warnings in mind, Simons says, “my aim is to explore how to make democracy work in the coming age of machine learning” (p. 5). By doing that, the author focuses on “the political character of machine learning” (p. 6). While announcing “a political theory of machine learning” (p. 9) in its introduction, the book never develops this in full. This would, in any case, be a book of its own.

Way more achievable is to have outlined what machine learning is and, on this, Simons says, “machine learning hold two fundamental promises for decision making: the promise of efficiency and the promise of fairness” (p. 15). Efficiency is one of the most favourite legitimising ideologies of management. It is used to justify virtually anything management does – from corporate environmental vandalism to criminality and beyond. Management presents efficiency as an eternal quest behind to which all workers must obey. Meanwhile, fairness is an utterly human, if not philosophical, concept that can hardly be transferred to a machine. In short, machine learning will incur many problems on both issues.

Rather than the false promise of fairness, “machine learning systems reflect historic inequalities” (p. 23). For one, the models used in machine learning “often involve trade-offs between complexity, accuracy, and error rates” (p. 25). Worse, “machine learning both amplifies and obscures the power of the institutions that design and use it” (p. 29). Machine learning not only structures communication, but it also controls communication. It controls communication so that what is seen and, often more importantly, what is not seen supports corporations and capitalism. In other words, machine learning is inextricably linked to communication, capitalism, and control.

To smokescreen this reality, “the politics of machine learning is often buried in […] technical details” (p. 31). Yet, machine learning is ‘about society, not mathematics’ (p. 47). Beyond that, machine learning is here to turbo-charge the rather “checkered history of systematic racism in the US criminal justice system” (p. 47). On this, for example, machine learning does three things:

1. it creates a “pernicious feedback loop” (p. 56) that reinforces racism, poverty and inequality;

2. it ”justifies more policing” (p. 56) even though after 400+ years of prisons and the unabated continuation of crime, perhaps prisons are not the answer to crime;

3. lastly, and this might be the worse part, machine learning is a tool that ”projects the imprint of injustice into the future” (p. 56).

Beyond that still lies the profit motive of capitalism. Simons writes, ”Facebook uses machine learning to power the advertising system that distributes ads to its 2.9 billion users” (p. 58). This is not just the very point where billions of dollars come in, but it also testifies to the economic and political power of Facebook’s monopoly. Worse, it also explains why Mark Zuckerberg is treated like the president of a country at international meetings. Zuckerberg can reach more people than any president of any country.

While the latter is truly impressive, the key to all this lies in the first part. Simons notes, ”companies find machine learning useful because it accurately predicts something genuinely useful for making a profit” (p. 66). In that undertaking, the aforementioned injustice and racism are mere – and often rather welcomed – by-products. On the latter, Simons notes, ”we can’t write an algorithm that’s going to solve racism” (p. 99). True, but what is often done is that such algorithms are written in such a way so that they – accidentally or deliberately – worsen racism. In other words, ”data is not really neutral. In fact, it’s the opposite of neutral” (p. 99).

On the equally dangerous side of ideology, Simons correctly emphasises that ”by building system that shape who sees what, when, and why, Facebook and Google mould the minds of billions of citizens and shape the public spheres of democracy across the world” (p. 105). The raw power of both monopolies – Facebook for social networking and Google to structuring information – can be seen in the fact that ”over 70% of all internet traffic goes through sites owned by Facebook and Google” (p. 106). To spice up the power of Facebook even more, ”Facebook’s most important system is newsfeed” (p. 107). Its algorithms select what consumer see and perhaps more importantly what they do not see.

It remains imperative to understand that ”machine learning models…prioritise some interests and values over others” (p. 110): profits over people, climate change denial over global warming, corporate interest over trade union interests, Donald Trump over Kamala Harris, etc. In other words, ”if people see too much lying, racism, pornography, and abuse [and Donald Trump], it is because Facebook built a ranking system that distributes and amplifies them” (p. 112).

The same applies to Google: ”what makes Google unique is PageRank, an algorithm that ranks the relevance of websites to a query” (p. 117). It ”encodes a kind of judgement” (p. 117). Worse, ”Google exercises control by defining concepts like equality” (p. 124). To keep their dominance concealed, ”Facebook and Google hide their power behind anodyne techno-babble” (p. 127) designed to ”obscure the politics of its machine learning systems” (p. 128).

This gives both ”infrastructural power to structure our public sphere” (p. 135). Worse, this power is ”unilateral, subject to neither meaningful economic competition nor effective democratic oversight” (p. 135). Just like Google, ”Facebook is the problem, not the solution” (p. 137) as it ”weaponizes us against ourselves” (p. 146).

On the old question of “what is to be done?”, Simons suggest that ”corporations [like Google and Facebook and many others] should be subject to public oversight and democratic governance” (p. 158). It is rather unsurprising to see that well-meaning, liberal and Harvard-trained authors like Simons suddenly discover that ”competition has been conspicuously absent” and that corporations have ”a monopoly position” (p. 163). Didn’t Karl Marx tell us about this in his Das Kapital about 160 years ago? Shock and horror! Google and Facebook are monopolies.

Yet, Simons argues what all good liberals argue when saying ”this book argues that to regulate Facebook and Google, this is precisely what we should do” (p. 167). In other words, regulate capitalism and corporations and all will be fine. For that, Simons makes a surprisingly good suggestion. He advocates that Google and Facebook, and many other IT corporations should be regulated like a public utility – like water, the postal service, the sewage, public waterways, etc. (p. 181). In his final chapter – Regulating for Democracy – Simons lays out a detailed plan on how this can be achieved.

In this conclusion, Simons tells us correctly that ”democracy cannot be automated” (p. 221). Beyond that, transferring Internet corporations into public utilities would also mean that such a move ”will require reforming the administrative state and changing how we think about policy making itself” (p. 219).

Overall, Simons has written a highly readable and exquisitely argued book. However, it re-tells all too many known cases where artificial intelligence and machine learning have gone badly wrong – from police racism to infamous and criminal Cambridge Analytica Ltd. So much for corporate social responsibility and business ethics.

Yet, the idea of transferring Internet corporations into public utilities or at least treat them as such under a state issued regulatory framework is sensible. After all, they hold monopolistic power. Google and Facebook can spread misinformation (accidentally) and disinformation (deliberately). Under the ideological justification of networking millions, if not billions of people, Facebook feeds users advertising and the news its algorithm deems relevant or not.

Both corporations, more or less, decide what, when, and why (p. 105) we see or read things. More importantly, they also decide what we do not see and when we would need to see it. Worse, they also have the power to hide things from the general public. This alone demands regulation and democratic oversight as Simons argues in his exceptionally well written book.

About the Author

Thomas Klikauer

Born on the foothills of Germany’s Castle Frankenstein (https://www.youtube.com/watch?v=iuqb0VSS9Ow), Thomas Klikauer (MA Bremen and Boston, PhD, Warwick) is the author of over 1,000 publications including a book on Media Capitalism (https://doi.org/10.1007/978-3-030-87958-7). https://klikauer.wordpress.com