The political Oracle

Can an AI Assistant Help Us Navigate the Flood of Information?

In our previous contribution, “Politics Needs to Become Boring”, we argued that serious political work does not need daily theatre.

A members’ party of part-time political adults cannot live from outrage alone. It needs a programme. It needs a sober working culture. It needs reliable information. It needs the ability to distinguish signal from noise.

But this brings us to the last and perhaps most difficult question in this series:

The problem is not that information is unavailable.

The problem is that information is everywhere.

Reports, speeches, studies, statistics, videos, podcasts, press releases, expert opinions, rumours, disinformation, propaganda, commentary, outrage, counter-outrage, half-truths, old facts in new costumes, new facts without context, and confident nonsense dressed as certainty — all of this reaches us every day.

The individual citizen has never had access to so much information.

And yet it has rarely been so difficult to know what to trust.

So we ask our seventh and final question:

Our answer is deliberately cautious:

What: An oracle, not a ruler

Let us start with the most important distinction.

• ✘We do not want an artificial leader.
• ✘We do not want a machine that decides our programme.
• ✘We do not want a digital chairperson.
• ✘We do not want an automated propaganda engine.
• ✘We do not want an algorithm that tells members what they must think.
• ✘We do not want a system that replaces democratic judgement.

What we imagine is much more modest — and precisely for that reason more useful.

• ✘The oracle should not decide.
• ✔It should clarify.
• ✘It should not command.
• ✔It should answer.
• ✘It should not replace members.
• ✔It should help members become more capable.

In Nick Bostrom’s terminology, this would be closest to “oracle mode”: a system designed to answer questions rather than act autonomously in the world. That is an important safety distinction.

But it does not remove all danger.

Even a system that only answers questions can influence people profoundly. It can frame issues, select sources, simplify trade-offs, hide uncertainty, flatter users, reinforce bias, or produce answers that sound more certain than they are.

Therefore, the political oracle must be treated not as a miracle, but as a constitutional problem.

Why: Members need orientation

In earlier articles, we argued that politics should not be left to career politicians.

• We argued that members should become part-time political adults.
• We argued that programme should matter more than personality.
• We argued that politics should become sober, predictable and serious.
• But if we mean all this, then we owe members the tools that make it possible.

A citizen who comes home after a long working day cannot read every report, every legal draft, every scientific paper, every party statement and every policy analysis. A member who wants to contribute cannot be expected to master climate science, European law, budget rules, migration statistics, AI regulation, defence policy and agricultural subsidies at once.

Without support, the demand for citizen responsibility becomes unfair.

Worse: it becomes selective. Those with more time, education, money and confidence participate. Others remain silent.

That would reproduce the very problem we wanted to overcome.

A political oracle could help lower the threshold.

• It could summarise.
• It could translate.
• It could explain jargon.
• It could compare positions.
• It could show which parts of the programme are relevant.
• It could separate facts from assumptions.
• It could present arguments for and against a proposal.
• It could identify what is known, uncertain and contested.
• It could preserve institutional memory.
• It could make the reasoning of the party accessible to members.

Used well, it would not make politics less democratic.

It could make democratic participation more realistic.

How: Programme-guided, source-based, transparent

The political oracle must not pretend to be neutral in the sense of having no worldview.

No political instrument is neutral.

The decisive question is whether its worldview is hidden or declared.

Our oracle would be explicitly programme-guided. It would assess political questions in light of the values, priorities and criteria laid down in our party platform.

But that does not mean it may invent facts to fit the programme.

On the contrary. – It must distinguish carefully between four things:

• facts,
• assumptions,
• interpretations, and
• programme-based evaluations.

If a member asks, “What should we think about this proposal?”, the oracle should not simply answer, “We are for it” or “We are against it.”

It should explain.

• What is the proposal?
• What are the known facts?
• Which sources support them?
• Which claims are disputed?
• Which programme principles are relevant?
• Which values conflict?
• Who is affected?
• What are the likely consequences?
• What would be the strongest argument in favour?
• What would be the strongest argument against?
• Where does uncertainty remain?

Only then should it say what seems to follow from the programme — and even then only as a recommendation for human judgement, not as a final decision.

The dangers: hallucination, dependency, bias

We must be sober about the risks.

First, hallucinations.

Generative AI can produce false answers with great confidence. We knew this already in 2023. We know it even better today. A political oracle that invents sources, misquotes laws or confidently summarises a non-existent report would be worse than useless. It would be dangerous.

Second, dependency.

If members rely too heavily on the oracle, their judgement may weaken rather than improve. The tool meant to empower citizens could become a new authority they obey.

Third, bias.

Every AI system reflects data, design choices, training methods, filters, prompts, retrieval sources and human decisions. Bias cannot be abolished by pretending it does not exist. It must be made visible, tested, challenged and corrected.

Fourth, opacity.

If members cannot understand why the oracle gave a certain answer, they cannot trust it responsibly. A black box is not a democratic institution.

Fifth, manipulation.

An AI system used in politics could easily drift from explanation to persuasion, from persuasion to targeting, from targeting to manipulation. That road must be blocked from the beginning.

Sixth, legal and ethical exposure.

In Europe, AI systems that influence elections or voting behaviour may fall into high-risk categories. Political advertising, data protection, transparency obligations and democratic safeguards all matter. A party that wants to use AI responsibly must not treat law as an afterthought.

What must come first

The oracle cannot be the beginning. – It can only be built on foundations already laid.

First, there must be a serious programme. Otherwise the oracle has no legitimate criteria.

Second, there must be a trusted knowledge base. Otherwise it has no reliable sources.

Third, there must be governance. Who decides which sources are authoritative? Who corrects errors? Who audits the system? Who can challenge its answer?

Fourth, there must be transparency. Members must know what the oracle can and cannot do.

Fifth, there must be human responsibility. No answer from the oracle should automatically become party policy.

Sixth, there must be patience. The first version should be modest: an internal knowledge assistant, not a political brain.

The order matters.

1. Programme first.
2. Knowledge base second.
3. Governance third.
4. Technology fourth.

Not the other way round.

A realistic path

The first stage should be simple.

The oracle should answer questions about our programme, statutes, previous articles and approved background materials. It should cite its sources. It should say when it does not know. It should avoid speculation. It should invite correction.

The second stage could add selected external sources: EU institutions, scientific bodies, courts, public statistics, parliamentary documents and high-quality journalism. Then it could produce issue briefings and argument maps.

The third stage could support member deliberation: summarising discussions, identifying areas of agreement and disagreement, preserving minority views and helping prepare decisions.

Only much later should more ambitious functions be considered.

And some functions should remain excluded entirely.

• ✘The oracle should not issue binding decisions.
• ✘It should not secretly profile members.
• ✘It should not micro-target voters.
• ✘It should not generate manipulative campaign material.
• ✘It should not replace human deliberation.
• ✘It should not become the hidden government of the party.

A political oracle worthy of a democratic party must be weaker than the members it serves.

That is its virtue.

What follows: An intellectual companion for political adults

So, do we need such a system?

Perhaps not today in its full form.

But we already need the function it would perform.

• We need help navigating information overload.
• We need help preserving institutional memory.
• We need help translating complexity into understandable choices.
• We need help comparing arguments fairly.
• We need help applying the programme consistently.
• We need help resisting propaganda, panic and manipulation.
• If citizens are to become politically mature, they need more than exhortation.

They need instruments.

The political oracle is one possible instrument.

• It is still partly utopian.
• It is technically immature.
• It is legally sensitive.
• It is ethically dangerous.
• It could fail.
• It could mislead.
• It could be abused.

But if designed humbly, transparently and democratically, it could also help solve one of the central problems of modern politics:

Our answer is deliberately cautious:

That is why we should not worship the oracle.

But we should begin building the conditions under which such an oracle might one day be useful.

This series began with the question: Why a political party at all?

It ends with a different question:

Our answer is deliberately cautious:

Our answer is now visible.

1. A party as vehicle.
2. A European mission with a practical national beginning.
3. Members instead of saviours.
4. Part-time political adults instead of career politics.
5. Programme over person.
6. Sober work instead of theatre. And,
7. Perhaps one day, an oracle that helps members understand the world without claiming to rule it.

That would not yet be perfect politics. – But it would be a beginning.

Annotated APA 7 references

1. Bostrom, N. (2014). Superintelligence: Paths, dangers, strategies. Oxford University Press.
• Bostrom’s distinction between oracles, genies, sovereigns and tools is essential for framing the political oracle safely. An oracle is a question-answering system, not an autonomous decision-maker. This supports the proposed limitation: the party should not build an AI ruler or political actor, but a constrained assistant that answers questions and leaves final authority with humans.

2. Bostrom, N., & Yudkowsky, E. (2011). Thinking inside the box: Controlling and using an oracle AI.
• This earlier paper is useful because it warns that even oracle-style systems can be dangerous. A system that only answers questions may still manipulate users, reveal harmful information or influence decisions in unintended ways. This supports the article’s insistence that “oracle mode” is safer only in relative terms and still requires strong governance.

3. Council of Europe. (2024). Framework Convention on Artificial Intelligence and Human Rights, Democracy and the Rule of Law.
• The Council of Europe convention is the central democratic-governance reference. It is described as the first legally binding international treaty on AI and aims to ensure that AI systems remain consistent with human rights, democracy and the rule of law. It supports the article’s claim that a political oracle must be treated not only as a technical project, but as a democratic institution requiring safeguards.

4. European Union. (2024). Regulation (EU) 2024/1689 laying down harmonised rules on artificial intelligence.
• The EU AI Act is directly relevant because it treats AI systems intended to influence election outcomes or voting behaviour as high-risk, while distinguishing some administrative and logistical campaign uses. This supports the article’s recommendation that the oracle should begin as an internal member-support tool and avoid voter manipulation, micro-targeting or automated campaigning.

5. Landemore, H. (2013). Democratic reason: Politics, collective intelligence, and the rule of the many. Princeton University Press.
• Landemore provides the democratic-theory foundation for the idea that politics can benefit from the collective intelligence of citizens, especially through inclusive deliberation and majority rule. Her work supports the oracle concept only if the AI strengthens collective judgement rather than replacing it.

6. Landemore, H. (2024). Can artificial intelligence bring deliberation to the masses? In Conversations in philosophy, law, and politics. Oxford University Press.
• This source is highly relevant because it directly examines whether AI can scale deliberation. It supports the optimistic side of the argument: AI may help broaden participation and make large-scale deliberation more manageable. But it also points towards the need for careful institutional design rather than technological enthusiasm.

7. OECD. (2026). Artificial intelligence and the future of citizen participation. OECD Publishing.
• This is one of the most directly relevant current sources. It identifies nine AI application types for citizen participation: information development, sense-making, translation, transcription, virtual assistance, moderation, facilitation, simulation and architecture. It also discusses risks such as bias, lack of transparency and weak deliberative quality.

8. OECD. (2025). Governing with artificial intelligence: The state of play and way forward in core government functions. OECD Publishing.
• The OECD’s broader government AI report is useful for public-sector implementation discipline. It notes that AI can improve decision-making, forecasting, fraud detection and public servants’ learning, but that benefits depend on managing risks such as skewed data, lack of transparency, overreliance, digital divides and propagation of errors.

9. Small, T., Tucker, J., & Tseng, Y.-S. (2023). Sorting a public? Using quali-quantitative methods to interrogate the role of algorithms in digital democracy platforms. Information, Communication & Society.
• This article is useful for understanding vTaiwan and Pol.is as digital democracy tools. It shows that algorithmic platforms may help structure participation and reveal nuanced positions, but also raises questions about how platforms format publics and represent issues. It is a balanced reference for the promise and limits of AI-assisted deliberation.

10. Tessler, M. H., Evans, G., Bakker, M. A., Gabriel, I., Bridgers, S., Jain, R., Koster, R., Rieser, V., Dragan, A., Botvinick, M., & Summerfield, C. (2026). Can AI mediation improve democratic deliberation? arXiv.
• This recent paper is directly aligned with the political-oracle question. It explores whether LLMs can enhance democratic deliberation by scaling participation, supporting equality and improving the quality of discussion. Its value lies in its balanced stance: it identifies real democratic possibilities while stressing unresolved empirical, technical and normative challenges.

11. Novelli, C., Formisano, G., Juneja, P., Sandri, G., & Floridi, L. (2024). Artificial intelligence for the internal democracy of political parties. Minds and Machines.
• This is one of the most relevant party-specific references. It argues that AI can enhance the measurement and implementation of intra-party democracy, while also placing party organisation, membership and leadership under new technological pressures. It supports using AI for member engagement and internal democracy, but not without governance safeguards.

12. Reuters Institute for the Study of Journalism. (2024). How generative AI chatbots responded to questions and fact-checks about the 2024 UK general election.
• This source provides a warning about current AI limitations in political information contexts. It reports that popular chatbots sometimes produced inaccurate, fabricated or inconsistent political information. This directly supports the article’s caution that a political oracle cannot rely on model memory alone and must be source-based.

13. Tseng, Y.-S. (2022). Algorithmic empowerment: A comparative ethnography of two open-source algorithmic platforms — Decide Madrid and vTaiwan. Big Data & Society.
• Tseng is useful as a critical reference. The article examines algorithmic platforms not simply as empowerment tools but as systems that shape participation and consensus. It warns against treating algorithmic mediation as politically neutral. This supports the article’s call for transparency and human oversight.