Featured answer: AI can widen horizon scanning, cluster signals and expose weak connections, but it should not decide which future is important. Strong foresight combines machine-assisted discovery with source traceability, alternative scenarios, structured challenge and explicit links between signals, assumptions, decisions and trigger points.
This evidence-led analysis by Jonas Adam Mohamed Osman Abdelghafour examines what changed, where uncertainty remains and how leaders can turn the issue into practical decisions without overstating what current technology can do.
Key takeaways
- Use AI to expand the field of view, not to collapse uncertainty into one forecast.
- Retain source provenance and publication dates for every material signal.
- Measure scenario diversity and decision sensitivity, not only extraction accuracy.
- Separate evidence, interpretation, assumption and speculation in every output.
- Human challenge is most valuable where models converge too quickly on a familiar narrative.
Foresight is not prediction
Strategic foresight explores plausible futures so decisions remain robust when uncertainty resolves differently. It is not a machine for producing a single most-likely headline. AI can search large bodies of text, classify developments, translate material and suggest relationships, but fluency can make a fragile interpretation look settled. The faster the synthesis, the more important it becomes to preserve disagreement and source context.
The OECD's work on AI in strategic foresight examines opportunities and risks in applying AI to foresight practice. A useful reading is that AI changes the economics of scanning and synthesis, while the institutional task remains fundamentally human: define the decision, choose the time horizon, test values and trade-offs, and decide what evidence would justify action.
A foresight team should therefore begin with a decision question rather than a technology demonstration. 'What might change?' is too broad. 'Which developments could make our 2030 skills strategy fail, and what no-regret actions remain valuable across those futures?' creates a boundary for evidence and judgment.
A traceable signal pipeline
The pipeline should capture each signal with its original URL or document identifier, publication date, institution, geography, sector, evidence type and extraction time. AI-generated summaries must link back to the source. Store direct quotations sparingly and preserve surrounding context. If a source cannot be retrieved later, the signal should be downgraded rather than allowed to become an unsupported fact through repeated summarisation.
Next distinguish signal strength from strategic impact. A weak signal may have low present evidence but high potential consequence. A mature trend may be highly evidenced yet already reflected in strategy. Score both dimensions separately, then add relevance to the stated decision and reversibility of available responses.
Duplication is another hidden problem. Ten articles repeating one press release are not ten independent signals. Cluster sources by underlying event and estimate source diversity. Weight primary evidence, peer-reviewed research and official statistics differently from commentary, while retaining minority views that reveal alternative mechanisms.
Technical framework for scenario diversity
Let each scenario s be represented by a vector of key assumptions xₛ. A simple diversity measure is the mean pairwise distance D = 2/[S(S−1)] Σᵢ<ⱼ d(xᵢ,xⱼ). The distance function should respect variable types and should not be mistaken for plausibility. Its purpose is diagnostic: near-zero diversity may reveal that every scenario is a paraphrase of the base case.
Decision robustness can be expressed as the proportion of scenarios in which an option meets agreed outcomes and constraints. Complement this with regret: the gap between the outcome of the selected option and the best option within each scenario. AI can calculate and explain these comparisons, but leadership must set outcomes, constraints and acceptable regret.
Backtest the process rather than pretending to backtest the future. Ask whether the pipeline found relevant historical signals before a known shift, whether sources were traceable, whether analysts challenged the dominant cluster, and whether trigger indicators would have changed a decision in time.
Hypothetical example: the future of professional education
A hypothetical university considers three strategies: add AI modules to existing degrees, build stackable short courses, or redesign assessment and curricula around human–AI collaboration. The scanning system identifies model capability, employer practice, regulation, energy constraints and student trust as drivers. Analysts construct four scenarios rather than asking the model for one 2032 forecast.
Under rapid capability growth and strict assurance, curriculum redesign performs best. Under slower progress and low employer adoption, modular courses avoid excessive sunk cost. The no-regret actions are stronger assessment integrity, staff AI literacy and a credential architecture that supports smaller units. The example does not forecast education; it shows how scanning supports staged commitments and trigger-based decisions.
Governance and responsible use
A foresight register should show prompts or analytical instructions, approved data sources, model version, retrieval date, analyst edits and unresolved uncertainty. Teams should test for geographic and language bias because an English-language corpus can overrepresent particular markets and institutions. Sensitive strategic material should not be sent to an unapproved external model.
Create a red team that tries to falsify causal stories, not just fact-check sentences. Ask which actor benefits, what capacity constrains adoption, which regulation could alter incentives, what second-order effect is missing and what observation would make the team abandon the scenario. This makes human judgment visible and reviewable.
AI output should never become evidence merely because it is repeated in a briefing. Each material claim needs an attributable source; each inference should be labelled; each scenario assumption should be explicit. The final decision remains with accountable leaders.
Practical implementation and monitoring
A practical operating cycle has four moments. Quarterly scanning collects and clusters evidence. A semi-annual synthesis identifies changing drivers and updates scenario assumptions. Decision workshops test current commitments against the revised scenarios. Trigger reviews determine whether observed indicators justify acceleration, delay or abandonment. The cadence should match the decision: cyber and AI-capability signals may need monthly review, while demographic or infrastructure pathways may change more slowly.
Quality assurance should sample the journey from source to board paper. Review whether dates and institutions were preserved, whether AI summaries changed meaning, whether one event was counted repeatedly and whether dissenting evidence was removed during synthesis. Analysts should document why a signal was promoted, downgraded or rejected. This audit trail protects the process from both automation bias and retrospective claims that an obvious future was always anticipated.
The portfolio should reserve resources for learning. Small experiments, option contracts, partnerships or modular designs can create information without committing to one future. Record the cost of the option, the question it answers, the expiry date and the trigger for scaling. Foresight becomes strategically useful when uncertainty changes the sequence and reversibility of action, not when it produces attractive descriptions of distant worlds.
How to read the current evidence
Evidence about emerging technology has several layers. Binding law and official implementation dates describe obligations; standards and guidance describe expected practices; peer-reviewed or metrology research tests particular mechanisms; institutional scenarios explore conditional futures; and vendor claims describe products under stated or unstated conditions. These layers should not be blended. A result established in one experiment is not proof of sector-wide readiness, while a scenario is not a prediction. Decision papers should label the evidence type, date and relevant jurisdiction beside each material claim.
Uncertainty should be carried into the decision rather than removed through confident prose. Teams should show which conclusion depends on adoption, performance, cost, infrastructure, regulation or behaviour; identify the observation that would change the conclusion; and set a review date. Where evidence is contested, compare sources and methods instead of taking an average of incompatible claims. This discipline is especially important for futurism and foresight, where technical capability, implementation capacity and social acceptance can move at different speeds.
Limitations and interpretation
This article separates enacted requirements and cited institutional evidence from the author's analysis. Hypothetical examples are explicitly illustrative. Technology performance, legal duties and implementation choices depend on context and may change after 16 August 2026; organisations should confirm current requirements and test claims in their own environment.
What decision-makers should do now
- Frame each exercise around a decision and time horizon.
- Retain source-level provenance for every material signal.
- Cluster duplicate reporting around the underlying event.
- Measure scenario diversity and option regret.
- Use trigger indicators for staged commitments.
- Red-team causal narratives and neglected futures.
Conclusion
AI makes it cheaper to collect and reorganise information, but foresight quality depends on disciplined uncertainty. The strongest system does not produce the longest trend report. It reveals assumptions, preserves alternative futures and helps leaders choose actions that can adapt. Machine assistance expands attention; accountable judgment determines meaning.
Frequently asked questions
Can AI predict the future?
No. AI can analyse data and generate scenarios, but structural breaks, human choices and unknown events prevent reliable single-path prediction.
What is horizon scanning?
Horizon scanning is the systematic search for emerging signals, trends, risks and opportunities that could affect a defined decision or system.
How should AI-generated foresight be validated?
Validate source provenance, extraction accuracy, signal diversity, scenario assumptions, decision sensitivity and human challenge.
What is scenario diversity?
It is the extent to which scenarios differ on material assumptions and mechanisms rather than merely using different wording.
Who owns a foresight decision?
The accountable executive or public authority owns the decision; AI and analysts support but do not replace that responsibility.