Futures of Maths

Frequently asked questions

What is a community consultation?

A community consultation asks a whole community what it thinks about a shared question, and lets the community shape the questions as well as answer them. Here the community is research mathematicians, and the question is how AI is used, and how it should be governed, in mathematics.

Unlike a standard questionnaire, the list of questions is not fixed in advance. It works like this:

  1. Launch statements. The research team wrote a first set of short statements about AI in mathematics, such as claims about research, teaching, publishing and credit.
  2. Rating. Each participant rates statements from "strongly disagree" to "strongly agree", or skips them, and can mark any statement as exceptionally important.
  3. New statements. Participants can post their own statements. Once a content moderator has checked them, they join the feed and everyone else can rate them. Over time, the statements come to reflect what the community itself thinks matters.
  4. Analysis. After the consultation, we measure how much participants agree with each statement and how important they find it. We also use statistics to find groups of participants who rate statements in similar ways, the statements that divide these groups, and the statements that all groups agree on. Tools such as Pol.is use a similar approach.

The aim is to map where the community agrees, where it disagrees and what it considers most important, rather than to produce a single vote.

Who can take part?

Researchers in mathematics or a related field who are 18 or older and have at least one article published in the last 3 years in a mathematics journal, meaning a journal indexed by zbMATH Open.

The quickest way to confirm this is to sign in with ORCID: the article must be on your ORCID record, added by Crossref. If that does not work for you, for example because your article is accepted but not yet published, you can ask the independent verification moderator for a manual review. You will need to show the article and that you are its author, from your institutional email address. Manual review may take a few days or longer.

Who is running the consultation?

  • Principal investigator: Jon Roozenbeek (VU Amsterdam and University of Cambridge).
  • Research team: Francesco Fournier-Facio (Heriot-Watt University), Yara Kyrychenko (University of Cambridge) and Julian Wykowski (KIT). The research team wrote the launch statements and analyses the responses.
  • Content moderators: mostly members of the research team. They check every statement that participants post before it appears in the feed, and may lightly edit its wording. They see the statement but never who wrote it, and they have no part in verification.
  • Verification moderator: someone independent of the research team. They check manual review applications from people the automatic ORCID check misses, and nothing else. They never see consultation responses, and the anonymous pass means they cannot link anyone to their answers.

The study is led from Vrije Universiteit Amsterdam.

Is it really anonymous?

Yes. The study uses two separate websites. The verification website checks that you are eligible and gives your browser a one-time pass. It signs the pass without ever seeing it, using a cryptographic technique called a blind signature. The consultation website accepts the pass but never learns your name, email address or ORCID iD. So nobody, including the research team and the verification moderator, can connect your answers to you.

We do not record IP addresses, and the consultation stores dates only to the month. The privacy statement gives the full details.

Where is my data stored, and who can access it?

Both websites and their databases are hosted in the European Union, by Scaleway SAS in Amsterdam, the Netherlands, and all connections are encrypted. Only the research team can see the consultation data (content moderators see the statements, but not who wrote them), and it is stored under a random code, never under your name, email address or ORCID iD.

Members of the research team and the verification moderator may work from the United Kingdom and access the data from there. The European Union recognises the United Kingdom as protecting personal data to a standard equivalent to its own (an adequacy decision, renewed in December 2025), so this is allowed under the GDPR. If you sign in with ORCID, your ORCID iD is also sent to ORCID in the United States to read your public record. The privacy statement gives the full details.

How much time does it take?

You can take as little or as much time as you like answering statements. You do not have to finish in one go: you can stop at any time and come back with your personal link until the consultation closes. New statements from other participants keep arriving, so it is worth checking back.

How does the statement feed work?

The Statements page shows a feed of statements, a few at a time, with more loading as you scroll. For each one, you choose a rating from "strongly disagree" to "strongly agree" or skip it, and you can tick it as exceptionally important.

  • Results. Once you have answered, you see how everyone has rated the statement so far. To protect everyone's privacy, results appear only once at least 5 people have answered it.
  • Select-all questions. Some cards, labelled "Select all that apply", list several options under one opening, for example the reasons you limit your use of AI in research. Tick every option that applies, or tick "None of these" (some questions word it differently), then press "Submit". You can also skip the question. Once you have answered, you see the share of people who ticked each option, once at least 5 people have answered. These questions appear in the feed among the statements and are chosen in the same way.
  • Ranking questions. After every few statements comes a ranking question, marked with its own label, which asks you to put options in order, for example declarations on AI in mathematics by how closely they match your views. Tap the options in order, starting with your first choice; tap again to remove one, or use the arrows to swap. The "i" button shows a short description of an option and a link to it. You can rank fewer options than asked, or skip the question. Nothing is saved until you press "Submit ranking". After that, you see everyone's combined ranking, once at least 5 people have answered. Skipped questions are listed under "Answered by you", so you can come back to them.
  • Pinned statements. A few statements that the research team would like everyone to answer are pinned at the top of the feed.
  • Topics. You can filter the feed by topic, and switch to "Answered by you" to see the statements you have answered and how everyone has rated them.
  • Posting. You can post your own statement of up to 280 characters with at least one topic. If similar statements already exist, they are shown so that you can vote on one instead. A content moderator checks every statement before it appears. They may lightly edit its wording or tags, and they never see who wrote it.

How is the order of statements chosen?

The order is chosen partly at random by a fixed set of rules, which we call the routing algorithm. Its aim is that every statement gets enough answers to be useful, and that you spend your time on the statements where your answer adds the most. Each time a new statement is added to your feed, it is drawn at random from the statements you have not yet seen, with these chances:

  • Launch statements first. While some launch statements remain unseen, about 40% of draws come from them alone, favouring the ones with the fewest answers so far.
  • Otherwise, from all statements, with more weight given to statements that:
    • are new or have had few answers (this extra weight halves every 10 answers);
    • divide opinion, either because ratings are spread out or because groups of participants who rate in similar ways disagree about them;
    • many participants marked as exceptionally important;
    • every group of participants agrees on, so that common ground is confirmed.
  • Less weight for statements that most people skip.
  • A share of every draw (10%) is spread evenly over all statements, so that every statement keeps a chance of being shown.

Select-all questions are drawn in the same way, as one card each. For the weights, each option counts as a statement that people who ticked it agree with and people who answered without ticking it disagree with, and the question takes the average over its options. As they have no importance box, they get the average importance of the other statements.

The algorithm uses only the ratings everyone has given so far. It never reads the wording or topic of a statement, and the order is not tailored to how you have answered. Because each draw is random with known chances, we record the chance that each statement had of being shown to you, and correct for it in the analysis, so that statements shown more often do not count for more.

What are karma points?

Karma points, shown on the "My session" page, are a light-hearted way to show your contribution. You earn them by answering statements (skipping counts the same as rating, so points never push you towards an opinion), by answering pinned statements and when statements you posted are published. No other participant can see your karma, there is no leaderboard, and points have no value outside the consultation.

What happens to the results?

Results are published in academic articles, reports and presentations as summaries, such as percentages and averages. A de-identified dataset may also be published openly so that other researchers can check and reuse it. It never shows who wrote which statement. Statements that participants post may be quoted. Section 6 of the participant information describes each step.

Does this consultation use AI?

No AI system, such as a large language model (LLM), is used on these websites for anything relating to participants or their data. Verification, the order of the feed, the suggestions of similar statements and the results you see all work by fixed rules and standard statistics. Every statement is checked by a human content moderator.

The websites themselves were built with the help of AI coding tools. The research team reviewed and tested the code and the text.

Is the consultation only available in English?

The original version of the consultation is in English. You are free to use your browser's translation feature to read it in your preferred language, but we cannot guarantee the quality of the translation. If you post a statement, please write it in English so that everyone can rate it.

Can I withdraw?

Yes, at any time while you have your personal link. Open "My session" and choose "Withdraw my answers". This permanently deletes your ratings, select-all answers, rankings and answers about yourself. Statements you wrote stay in the study, with nothing linking them to you.

Who can I contact with questions?

Contact Jon Roozenbeek at futuresofmaths@tutamail.com. Questions about your personal data can also go to the VU Amsterdam Data Protection Officer at functionarisgegevensbescherming@vu.nl.