When I started writing this, it was election day in Sweden.
Election days always make me think about how strange democracy is. We treat it as something solid: an institution, a system, perhaps even a national character trait. But democracy depends on people continuing to do the work required to sustain it.
Part of that work is thinking for ourselves.
That means taking in incomplete and sometimes contradictory information, questioning it, and forming an opinion we can explain. It also means accepting that we might need to change our minds.
None of this happens automatically because we have access to information. How that information reaches us matters. So does what we do with it.
We Are Outsourcing The Wrong Thing #
We ask AI to research subjects, summarise arguments, interpret data, and talk us through personal problems. Sometimes we ask substantial questions. Sometimes we ask for the capital of Spain. It is still Madrid, as far as I know.
The attraction is obvious. Thinking is slow and uncomfortable. AI is fast, articulate, and endlessly available.
I worry about what happens when accepting its answers becomes a habit. Asking for help does not automatically weaken our judgment. But routinely handing over the difficult work of comparing explanations and questioning assumptions leaves us doing less of that work ourselves.
That is the cognitive debt I am concerned about: becoming accustomed to conclusions we have done little to examine.
Remember Cambridge Analytica? Its use of personal data for political targeting prompted questions about how precisely voters could be profiled and influenced. People have asked me whether something similar would be possible today.
I think the more pressing question is how the relationship has changed.
A targeted advertisement still arrives as an advertisement. A chatbot is something you voluntarily consult about subjects you do not understand. You explain your uncertainty, ask follow-up questions, and receive an answer tailored to the conversation.
The opportunity to influence sits inside the explanation you came looking for.
Patterns Make Us Vulnerable #
Pareidolia is our tendency to see recognizable forms in ambiguous things: a face in a cloud, an animal in a rock. It is a useful analogy for the meaning we can supply when interacting with a chatbot. Fluent conversation can invite us to imagine a knowledgeable, trustworthy character behind the words.
Familiarity can strengthen that impression. When the answers also fit what we already believe, confirmation bias gives us another reason to accept them.
I have written before about sycophancy in large language models: their tendency to agree with us when a challenge would be more useful. No devious plot needs to be behind that behavior for it to become a problem.
Here I am deliberately moving into conjecture. What if someone altered a widely used chatbot’s instructions to favor a political position? A change in emphasis. A source quietly omitted. One proposal consistently described as sensible, another as reckless.
The possibility of instructions directing political output is hardly fanciful. In 2025, xAI attributed Grok’s political responses to an unauthorized prompt change.
That does not establish the subtle electoral influence I am imagining. My concern is what happens when people trust a source whose editorial choices they cannot inspect, and grow less inclined to question those choices at all.
Polling Is Data, Headlines Are Decisions #
Newspapers make choices about what we see too. Their choices are often much easier to spot.
In May, Aftonbladet reported a “horror figure” for the governing bloc, with the red-green opposition leading by 9.1 percentage points. Ten days before the election, the same newspaper announced that the Social Democrats were “collapsing”, as the gap between the blocs narrowed to 3.3 points.
Those figures describe different moments. Opinion can change, and a substantial movement deserves examination. Different results don’t make either poll worthless.
But the distance between a measurement and a headline matters.
Polling data has sampling error, timing effects, methodological choices, uncertain turnout assumptions, and groups that are notoriously difficult to reach. A movement in one poll may be a signal. It may also be noise. Yet noise does not generate clicks unless it is dressed as momentum, panic, collapse, or resurrection.
Here’s the thing: publishing a number is not a neutral act. Choosing the number is an editorial decision. Choosing the comparison is an editorial decision. Turning it into a headline about a collapse is absolutely an editorial decision.
For me, an election is deeply personal. I see no democratic value in treating weak movements in late polls as a football score. The likely result is not a better-informed electorate. It breeds anxiety, tactical voting, and the feeling that politics is a horse race where the clever choice is to back a winner.
I cannot tell you how many votes a headline changes. I can question whether it helps readers make an informed choice.
A free, uncomfortable, unafraid press is the bedrock of democracy. That is precisely why I expect more from it. Reporting uncertainty honestly is part of its responsibility. So is resisting the temptation to turn every new measurement into a crisis.
Just because a number is available doesn’t mean it’s the most important thing to tell a voter.
Critical Thinking Is Not Automatic #
The responsibility is shared. AI companies and news organizations make choices that shape our understanding, and they should be accountable for them. Telling readers and users to think critically does not discharge those obligations. It does not magically give anyone the ability, either.
I can explain swimming in excruciating detail. You can understand every word. But if you have never practiced, I would strongly advise against testing that understanding in deep water. Critical thinking takes practice too, and outsourcing it gives us less of that practice precisely when we need it most.
That leaves me with an obligation of my own. Especially when an answer tells me exactly what I wanted to hear. Whether someone intended to influence me or simply passed along an assumption they never questioned, I still need to examine what I am accepting and why.
Democracy requires us to form opinions we can explain, take responsibility for our choices, and remain willing to reconsider them. We need information that helps us do that work, and room to think through it.
Election day makes that responsibility personal. I have to make a choice with incomplete information, knowing I might be wrong. I want the difficult questions examined seriously. I can live with some of them remaining unanswered.
Give me facts. Give me context. Give me uncertainty honestly described.
I can make my own choice. Give me a chance to hear myself think.
Photo by Elimende Inagella on Unsplash