Pisa 2035 – the panel wants the change but doubts it will come in time

This report is not an analysis but an AI-assisted summary of the results. The actual analyses are carried out with the DAE analysis tools.

Summary

• Finding 1: All three futures were seen as desirable (medians 8, 7 and 6 on a scale of 0–10), but none was seen as probable (medians 5, 6 and 5). The gap between wanting and expecting was widest for equity: 7 of 25 panelists rated closing the gaps highly desirable but unlikely.
• Finding 2: No item polarised the panel. All six ratings fell into the consensus categories of the analysis, and the expertise groups did not differ from one another in a statistically meaningful way.
• Finding 3: On the future of PISA, desirability and probability went together (Spearman ρ = 0.50, p = 0.015). Those who wanted the change also expected it, which the analysis reads as a sign of wishful thinking rather than of a shared forecast.
• Strongest disagreement: whether lifting the weakest takes something away from the strongest. Some panelists saw a zero-sum trade-off of resources, others rejected the framing altogether.

Why this question

PISA 2022, published in December 2023, confirmed a broad decline in learning outcomes across most OECD countries. In Finland it deepened a long downward trend, although Finland remained above the OECD average. The influence of home background on outcomes grew, the equalising effect of the school system weakened, and the gap between girls and boys in reading remained exceptionally wide. The results also showed growing anxiety in mathematics and weaker engagement with school.

The study asked what could follow from this by 2035. The horizon of about ten years is long enough for a change in school culture to become visible, but short enough that today's decisions still shape it. Each thesis was opened with figures from PISA 2025: for example, the gap in science between students with and without an immigrant background is 88 points in Finland, against an OECD average of 43.

Pisa 2035 is part of an experimental series of short Delphi studies run by a community of Delphi developers. Each study has one to three theses, and a new topical question is put to the panel every month, in the same rhythm as the Friday–Saturday workshops of the Delphi method training. It is also the first Delphi study carried out on the Delphi Pronoia platform (www.pronoia.fi); the series' earlier studies were run in eDelphi (www.edelphi.org).

How the study was done

• Method: Real-Time Delphi with a dialectical (D) orientation. One open window in which the panelists answered, saw the others' reasoning straight away, commented on it and could revise their answers. Each thesis was rated on two criteria, desirability and probability, on a scale of 0–10, and every answer carried a written reason.
• Panel: 52 human panel members, of whom 25 took part (48 %). 22 answered all three theses, 2 answered two and 1 answered one. By expertise: 13 experiential experts, 8 from pedagogy and 4 who joined through open registration. The panel was also built along a second axis, inside and outside the school, but one of its cells had fewer than 3 participants, so results are not reported along that axis. No AI panelists were used.
• Window: responses from 21 September to 8 October 2026, dialogue until 10 October 2026. The material was frozen in the Pronoia archive on 12 October 2026.
• Material: 71 responses, 55 written arguments, 30 comments and 20 revisions. Nine panelists wrote comments.
• Environment: Delphi Pronoia. The quantitative analysis is Pronoia's own (agreement index A by van der Eijk, with thresholds calibrated to the panel size). The study was planned with the Metodix planning modules. No simulated arguments were used, and the archived panel data contain no AI-generated content.
• AI in this report: the report is an AI-assisted summary of the results, drafted from the archived data and checked by the author. It is not an analysis: the interpretive analyses (argument, dialogue and discourse analysis, weak signals, scenarios) are carried out with the DAE tools. Every figure comes from the archive record dsr-2026-pisa-2035-kolme-teesia.
• Consent: panelists took part with consent. Only aggregated results and anonymous quotes are reported, and no quote carries a panelist code.

Results

The theses are given in the original Finnish, followed by an English rendering.

Thesis 1: Lifting the weakest

Peruskoulussa asetettiin Pisa 2025 -tulosten jälkeen etusijalle heikoimman neljänneksen (tausta, sukupuoli, kieli) osaamisen nostamisen, ja se on kuronut eroja umpeen vuonna 2035.

(After the PISA 2025 results, basic schools gave priority to raising the skills of the weakest quarter – by background, gender and language – and by 2035 this has closed the gaps.)

Desirability: median 8 (interquartile range 6–9), n = 25. Probability: median 5 (4–6). This was the most desired thesis and the one with the widest gap between wanting and expecting. In the joint distribution, 10 panelists rated the thesis both desirable and probable, and 7 desirable but improbable.

The arguments for the thesis appealed to social cohesion and to children. The arguments against were not against equity as such but against its price and its route: resources taken from the more motivated students, a protest from the active middle class, and trends that already widen the gaps.

Erojen kurominen umpeen ei voi tapahtua vetämällä parhaimpia alaspäin.

(Closing the gaps cannot happen by pulling the best down.)

Se, että autetaan heikommassa asemassa olevia, ei ole pois paremmin menestyviltä.

(Helping those in a weaker position is not taken away from those who do better.)

Thesis 2: Peer and community learning

Vertais- ja yhteisöoppimisen lisääntyneet menetelmät ovat vuoteen 2035 mennessä syrjäyttäneet yksilösuoritukseen nojaavan koulukulttuurin valta-aseman.

(By 2035, the growing methods of peer and community learning have displaced the dominance of a school culture built on individual performance.)

Desirability: median 7 (6–8), n = 23. Probability, assessed on the condition that the intervention is carried out: median 6 (4.5–6.5). Almost half of the panel (11 of 23) rated the thesis both desirable and probable, and no one rated it undesirable.

The panel did not question the value of learning together. It questioned the word displaced. Many argued that individual and collaborative learning are not opposites, and that assessment, examinations and selection for further study still rest on measuring the individual. The most common view was that collaborative methods grow alongside individual performance rather than in its place. A sharp exchange in the dialogue concerned resources. One panelist argued that scarce resources make the change unlikely; another replied that the change is a question of leadership and courage:

Resurssit eivät ole vastaus mihinkään.

(Resources are not the answer to anything.)

On kuitenkin epätodennäköistä, että se jo vuoteen 2035 mennessä syrjäyttäisi yksilösuoritukseen nojaavan koulukulttuurin valta-aseman.

(It is unlikely, however, that by 2035 it would already displace the dominance of a school culture built on individual performance.)

Thesis 3: PISA changes

PISA sellaisena kuin sen tunnemme – valvottu yksilökoe ilman apuvälineitä – on vanhentunut vuoteen 2035 mennessä, ja mittaristoa on kehitetty tekoälyavusteisen ongelmanratkaisun ja ajattelun prosessin arviointiin.

(PISA as we know it – a supervised individual test without aids – is outdated by 2035, and the measures have been developed to assess AI-assisted problem-solving and the process of thinking.)

Desirability: median 6 (5–8), n = 23. Probability: median 5 (4–7.5). This thesis had the widest spread of the three and the weakest agreement on desirability. It was also the only one in which the two criteria were linked: the more a panelist wanted the change, the more likely they thought it.

One line of argument held that if AI becomes a normal tool, learning cannot be measured only in a situation where that tool is taken away. The other held that an institution with long time series will not give up comparability, and that the value of measuring what a person can do without a machine is only growing.

Koulutuksen onnistumista ei pitäisi mitata vain tilanteessa, jossa keskeinen työväline poistetaan käytöstä.

(The success of education should not be measured only in a situation where the key tool is taken away.)

Epäilen vahvasti, että noin institutionaalinen putiikki hukkaisi mahdollisuuden kerätä vertailukelpoisia tuloksia yli ajan.

(I strongly doubt that such an institutional outfit would give up the chance to collect results that are comparable over time.)

Where the panel disagreed

In the figures, the panel agreed: all six ratings fell into the consensus categories, and no item polarised. The disagreement lay in the reasons. On equity, the panel split on whether priority for the weakest is a zero-sum trade-off of resources or a question of using existing resources more wisely. On PISA, it split between those who see the change as inevitable and those who see the institution's inertia as stronger than the technology. In Delphi terms, the study produced a consensus of direction and a dissensus of route.

Of the 55 arguments, 31 supported the thesis they answered, 19 challenged it and 5 opened a new perspective. The classification of the arguments' knowledge bases (empirical, experiential, normative and so on) has not yet reached sufficient reliability between coders, so it is not reported here.

Weak signals and tensions

The futures phase of the DAE analysis pipeline (weak signals and tension mapping) was not run for this study. The arguments contain observations that could be examined as weak signals, for example a parent's observation of growing class sizes and of schools giving up on weaker pupils, but they have not been analysed as such.

What follows

For education policy, the message is that the direction is not in dispute, but faith in its realisation is weak. The panel wants equity, collaboration and an assessment that keeps pace with AI, yet it does not believe that the school will get there on its own within ten years. The gap between wanting and expecting is where decisions are needed: on how resources are allocated, on how individual and collaborative learning are combined in assessment, and on what is measured when the tools change.

Limitations

• Response rate: 25 of 52 panel members took part. Those who did not respond may differ from those who did.
• One window: a Real-Time Delphi has no second round in which a revised set of theses could be tested. Of the 20 revisions, only 3 changed a rating; the rest added or refined arguments.
• Order of the criteria: the order of desirability and probability changed between the first thesis and the other two, which two panelists pointed out in their answers. At least one revision looks like the correction of a reversed rating. The figures may contain a small number of answers given in the wrong order.
• Dialogue: the comments were concentrated on a few active panelists, so the dialogue does not represent the whole panel.
• Condition on thesis 2: the probability of thesis 2 was asked on the condition that the intervention is carried out, so its probability is not directly comparable with the other two.

The study as teaching material

Pisa 2035 is an authentic Delphi carried out in connection with method training. It is suitable for teaching in at least three ways. It shows how a Real-Time Delphi works in practice: one window, immediate feedback and revision. It shows why two criteria per thesis are informative: a single scale would have hidden the gap between wanting and expecting. And its limitations are themselves material: the order of the criteria, the response rate and the concentration of the dialogue are design questions that a student can learn from. A separate teaching-case article will describe the design decisions, the course of the window and exercises.

Acknowledgements

Thanks to the panel.

How to cite

Linturi, H. (2026). Pisa 2035 – the panel wants the change but doubts it will come in time. Delphi Pronoia / Metodix. Study record dsr-2026-pisa-2035-kolme-teesia.