What If We Left Voting to AI?

Last week, partly out of curiosity and partly to test my own decision, I asked the AI I use a question: considering Arya, the ecosystem we are part of and the city we live in, who should I vote for in the upcoming chamber of commerce election? It assessed the candidates, their projects, their track records and whatever information it could reach. Then it named someone. What was interesting was that the person it recommended was not the one I supported. My first reflex was to try to find where the AI had gone wrong. Then I stopped. What had I based my own choice on?

When we choose someone, we don’t look only at their projects. Knowing them, knowing how they have behaved in the past and whether they inspire our trust are all part of our decision. Even the fact that they once paid us a visit says more than we think. When someone else we trust supports them, we can leave what we don’t know directly about the candidate to that person’s judgment. Friendship, obligation, belonging and relationships built over years are part of the decision too. None of these is wrong in itself. But when the AI reached a different conclusion from mine, another scenario came to mind. What would happen if we took the person out of the picture for a while?

Let’s hide the candidates’ names, photos, companies and who backs them. Everyone presents their projects in the same format. What will you do, why will you do it, who will benefit, what will it cost, what risk does it carry, and how will we measure success? Voters first choose the future they want by looking only at the projects. Then the names are revealed and we move on to the second question: whom should we give the authority to make that future happen? If we chose the idea first and the person second, would we still vote for the same people?

Such a system would not only separate the person from the idea. It would also push voters to ask “what do I actually want?” before “whom do I support?”. And it would force those seeking to represent us to move beyond general rhetoric and explain what they will do, how, with what resources and to what end. Today, wildly different promises can live comfortably inside the same sentences. In a system where we chose projects first, both making promises and voting would become a little harder. Perhaps its most important contribution would not be finding the right project, but bringing both voters and candidates into a clearer discipline of decision-making. After all, when we vote for a candidate, we are deciding at least four separate things: what kind of future we want; which projects we think will take us there; who we judge capable of delivering them; and whose judgment we will trust when a problem we don’t yet know about appears tomorrow. We answer four different questions with a single name. Sometimes the problem with elections is not that we choose the wrong person, but that we squeeze different decisions into a single vote. When the person, the idea, trust, relationships and the future all go into the same package, it becomes hard to tell what we are actually voting for.

Now let’s go back to the first question and take AI with us to the ballot box. A system that knows me, my company, my sector and the decisions I have made over the years can read hundreds of pages of project documents far faster than I can. It can see that a project benefits my company while working against another sector. It can compare what a candidate says today with what they have done in the past. It can put contradictory promises, and alternatives I hadn’t noticed, in front of me. I would like to use such a system. In fact, I already have. I would even like everyone to try it. Ask the AI you use to assess the candidates with you, your company, your sector and the city you live in in mind. Then, before you look at its answer, look at your own first reaction. When it recommends the same person, do we look for why it is right, and when it recommends someone else, for why it is wrong?

What we must not overlook in this experiment is that AI sees not the world, but the traces the world leaves behind. One candidate may have done excellent work for years, yet left only a faint digital trail. Another has a strong communications team, hundreds of news stories and videos, even beautifully prepared project documents. The AI finds more evidence about the second candidate. But more visibility is not more merit. And an absence of data is not proof of failure. So is AI finding the good candidate, or the candidate AI finds easier to read? What is more, candidates will learn that voters are using these systems. From then on, they will start producing digital traces designed to persuade not only people but also the models that assess them. The system we use to decide will change the behaviour of the person trying to be chosen.

The idea that we entrust the person we elect with more than today’s projects is not new. In his 1774 speech to the electors of Bristol, Edmund Burke said that a representative owes the voters not only his industry but his judgment. A representative was not merely someone who carried out voters’ current demands, but someone who would use his own judgment on matters that could not be foreseen at the time of the election. Some 250 years later, we are adding a new actor to the same problem. Alongside the voter and the representative, there is now another intelligence that can calculate more than either of them can see. That forced me to look into my own decision again.

In an election we are not choosing only a person. With a single vote, we choose the future we want, the projects we believe will take us there, the person who will deliver them, and the judgment they will use on problems we don’t yet know about. AI may compare all of this better than I can, show me contradictions I can’t see, and even make me question my own choice. But its calculating better than I can what is right for me will not mean I have given it the right to decide what is right.