The democracy of algorithms

Diego mir

Alan Turing's idea of ​​the computer as a "universal machine" does not mean that it can solve any problem. Algorithmic governance is well suited to improving certain aspects of the political process, but it is of little use to others; can correct human deficiencies and biases, it serves to identify certain preferences, to measure impacts, but it is inadequate for those dimensions of the political process that are not susceptible to computation and optimization, areas that do not have an easy quantification and measurement, that is, to the genuinely democratic moment in which the criteria and objectives that technology can subsequently optimize are decided.

The reason that algorithms are politically limited lies in their instrumental character. Algorithms serve to achieve predetermined objectives, but they do little to determine those objectives, a task inherent to political will, democratic reflection and deliberation. The role of policy is to decide the design of algorithmic optimization strategies and always maintain the possibility of altering them, especially in changing environments. In a democracy, everything must be open to moments of re-politicization, that is, to the possibility of questioning the established objectives, priorities and means. This is what politics is for and what algorithms are not for.The algorithmically optimized government does not have the capacity to resolve the strictly political conflicts or the political dimension of those conflicts, that is, when the frameworks, ends or values ​​are in question. As Lucy Suchman said in another context, robots work very well when the world has been arranged the way it should be arranged.

Algorithmic governance is oriented to achieve objectives that have not been discussed, that it does not establish or put into question. However, democratic politics is not a mere processing of information, but its interpretation in a context of guaranteed pluralism; It is not about how best to achieve certain goals but how to decide them. The resolution of problems of an administrative nature is very different from politics understood as the conflict of interpretations about reality, where it is not about optimizing results but about establishing them.

There is a big difference between how algorithmic systems learn and how democratic decisions are made. Automatic decision-making systems process information to achieve certain objectives as well as possible, while democratic politics, in contrast, does not try to optimize predefined objectives but above all to find out what those objectives should be. The political begins there where it is necessary to debate about what algorithms must satisfy, what values ​​they must fulfill, what conception of fairness they must serve.

There is an everyday example of the difference between deciding goals and implementing them inadvertently, with which one and the other are confused. I refer to that common place according to which we should not care who governs, whether it is from the right or left, but who manages well, as if that management could be valued without resorting to ideological estimates (we could call it "the Bertín Osborne Principle") . This topic is plausible only to the extent that, indeed, right and left are no longer what they were and, as rigid categories, they explain less and less. But those who defend it are not usually wanting a de-ideologized politics but a depoliticized politics.

As in politics in general, also when we speak of algorithmic governance, the idea of ​​producing better decisions with the help of machines requires that there be previously a criterion about what is a good decision. It is true that artificial intelligence serves to inform decisions and optimize results, but although some economists have tried to quantify and measure aggregate well-being, there is no predefined and incontestable notion of what is a satisfactory policy result. Democracy is a political system that starts from ignorance about what could be a good decision, which is suspicious that someone pretends to know it and sets in motion collective learning procedures to overcome that perplexity.

The meaning of the institutions of mediation in a democracy consists in establishing a distance between the immediate will and the political decision.

The procedure for this is the opening of spaces in which something like a deceleration of decisions is possible to allow the free exchange of opinions and points of view.

A democracy requires this capacity when it comes to satisfying diverse preferences and interests, which not infrequently pose crazy demands.

In this regard, the presence of the people in algorithmic democracy is more of the

will of all

than of the

general will,

to use Rousseau's terminology, of aggregation than of configuration, of sovereignty than of democracy: our starting preferences are taken into consideration, of course, but we are deprived of the moment of deliberative construction in which those preferences are no longer merely aggregated. rather, they interact with others. The problem with algorithmic governance is that thanks to algorithms we intervene in the expression of preferences and interests, but not in the construction of a desirable social totality that would have allowed us to eventually modify them. Our presence in the algorithmic democratic process would be to put our traces and traces at the disposal of decision systems,but not to intervene in the dialogue in which these data are weighed and the idea of ​​a desirable society is discussed based on them. In an algorithmic democracy, being a citizen would consist of having the right to emit wishes but not to weigh them with those of others and even modify those own wishes.

The great promise of algorithmic governance is that optimal results make us forget about desirable procedures.

It is a type of governance that seems to prefer effectiveness even at the price of being excluded from decision-making (or reduced to a minimal, implicit and individual presence, in the form of requirements and preferences present in our fingerprints).

But if the citizenry cannot in any way supervise or control algorithmic decisions, we cannot call that self-government of the people.

Since the results of the government are very important, what defines democracy is more the procedure than the result.

Democratic government is not about providing certain

outputs,

but about guaranteeing certain

inputs,

specifically those that ensure equal freedom for all citizens to take part in the process of shaping political will and in decision-making processes. Algorithmic governance can only be democratic when its objectives and procedures have been expressly authorized by the people in an act of a political nature. This governance has at least three weaknesses from the democratic point of view: that we think that by issuing digital signals we have already sufficiently expressed what we want, that we have done so without explicitly internalizing the compatibility of our will with that of others, that we do not notice the deep heterodetermination that this supposes.A governance that seems to legitimize itself because it does not impose, but rather pleases, runs the risk that we are so satisfied that we stop worrying about the conditions in which that satisfaction has occurred.

Daniel Innerarity

is professor of Political Philosophy, Ikerbasque researcher at the University of the Basque Country and professor at the European Institute of Florence. @daniInnerarity