Futures, Not Forecasts: How AI Redraws the Boundaries of Foresight

Forecasts are comforting because they pretend the future has already filed its paperwork.

A number arrives. A probability. A trend line. A neat dashboard arrow pointing upward, downward, or toward whichever strategic conclusion the executive team secretly preferred anyway. This is why forecasting tools sell so well: they reduce uncertainty into something that looks like management.

The problem is that important futures rarely behave like quarterly demand estimates. Climate policy, geopolitical instability, energy transition, AI regulation, public health, supply-chain redesign, urban planning, demographic change, and institutional trust do not simply “arrive.” They are shaped by choices, incentives, narratives, constraints, and feedback loops. In other words, they are not only things to predict. They are things to design around.

That is the core move in María Pérez-Ortiz’s position paper, From Prediction to Foresight: The Role of AI in Designing Responsible Futures.¹ The paper coins and frames the idea of responsible computational foresight: the use of AI and computational modelling to support human-centered future design, not to replace judgment with machine prophecy. It is not an empirical benchmark. There are no leaderboard tables where one foresight model humiliates another by 3.7 percentage points, thank heavens. Instead, the paper builds a conceptual map of how AI can support foresight when the goal is not to guess one future, but to explore multiple possible futures responsibly.

That distinction matters for business. Most organizations still treat AI strategy tools as upgraded forecasting engines. The paper’s more useful argument is that AI should become a strategic rehearsal system: a way to test assumptions, generate scenarios, simulate consequences, surface weak signals, involve stakeholders, and keep decision-makers from mistaking a model output for destiny. A crystal ball with an API is still a crystal ball. It just bills by token.

The paper’s real contrast is prediction versus foresight

The easiest way to misread this paper is to think it asks whether AI can forecast the future better than humans.

That is not the central question. The paper accepts that prediction is useful. Weather forecasting, traffic modelling, climate projections, epidemic modelling, financial risk estimation, and demand planning all depend on predicting how a system may evolve. Prediction is not the villain. It is just too narrow for many strategic problems.

The paper separates two modes of future-oriented thinking:

This is why the paper repeatedly pushes against the idea of one definitive future. Prediction narrows attention toward probability. Foresight widens attention toward possibility, consequence, desirability, and agency.

For business readers, this is the first useful translation: forecasting helps you prepare for what may happen; foresight helps you decide what should be made more or less likely.

A retailer forecasting next month’s inventory needs prediction. A government designing an AI regulation regime, a bank assessing long-term climate exposure, or a manufacturer deciding whether to restructure its supply chain around geopolitical fragmentation needs something broader. These decisions are not only about estimating a future state. They are about acting in ways that influence future states.

That is where AI becomes interesting, and also slightly dangerous.

AI is useful because foresight has become too cognitively large

Responsible foresight sounds noble until someone has to operationalize it.

The paper’s definition is demanding. Responsible foresight must be methodologically rigorous, grounded in high-quality data, flexible, ethically oriented, inclusive, transparent, sustainable, and alert to intergenerational consequences. Fine. A modest Tuesday.

The reason AI enters the picture is not because humans suddenly forgot how to think. It is because the systems involved have become too large, too interconnected, and too data-heavy for unaided reasoning. Public policy is the paper’s main domain, but the same pattern applies to corporate strategy. Decision-makers must now interpret unstructured data, social sentiment, scientific research, regulatory signals, market behavior, operational constraints, public legitimacy, and second-order effects across multiple systems.

The paper gives examples from policymaking: AI can synthesize large datasets, detect emerging public needs, analyze social sentiment, support urban planning, help evaluate policies, and simulate potential impacts. These are not merely faster versions of old spreadsheets. They allow institutions to see patterns that would otherwise remain scattered across documents, platforms, disciplines, and agencies.

In business terms, AI helps with foresight because it can perform four practical functions:

1. Signal aggregation: collecting weak signals from data streams, reports, markets, public discourse, and operational systems.
2. Scenario expansion: generating plausible alternatives beyond the default “base case, optimistic case, pessimistic case” ritual.
3. System simulation: testing how interventions may ripple through interconnected domains.
4. Deliberation support: helping humans compare assumptions, values, risks, and trade-offs.

Notice what is missing: “AI decides the future.” That omission is not accidental.

The paper’s argument is that AI should close parts of the decision loop without owning the loop. It can help identify needs, formulate alternatives, test consequences, monitor outcomes, and support revision. But the judgment about what is acceptable, legitimate, fair, and desirable remains human. This is not sentimental humanism. It is a systems requirement. Many foresight problems contain values that cannot be optimized without first being debated.

The toolkit is not a menu; it is a division of labor

The paper reviews a wide toolkit: superforecasting, prediction markets, world simulation, digital twins, surrogate modelling, simulation intelligence, scenario generation, participatory futures, futures literacy, hybrid intelligence, and human-computer interaction.

A weak summary would list these tools one by one. Useful, perhaps, for insomnia. The stronger interpretation is to ask what job each tool performs in a responsible foresight system.

This is the operational heart of the paper. Responsible computational foresight is not a single model class. It is an architecture of complementary functions.

Forecasting tools narrow uncertainty. Simulation tools test consequences. Scenario tools expand imagination. Participatory tools widen legitimacy. Hybrid intelligence keeps humans inside the loop not as rubber stamps, but as active challengers of assumptions.

That last part is crucial. The paper uses a memorable distinction between foretell and forsay: AI should not simply announce a future; humans must retain the ability to question, contradict, and reshape the futures AI helps produce. Without that, the organization gets what the paper neatly calls “Garbage In, Gospel Out.” This phrase deserves a place on the wall of every AI strategy meeting, preferably above the dashboard everyone treats as divine revelation.

The practical upgrade is from prediction dashboards to strategic rehearsal rooms

The business relevance is not that every company now needs a “responsible computational foresight platform.” That phrase alone could frighten procurement.

The practical shift is simpler: organizations should stop treating future-oriented AI as a tool for producing one confident answer and start designing it as a rehearsal environment.

A forecasting dashboard asks:

What is likely to happen?

A strategic rehearsal system asks:

What could happen, what would it affect, who would be exposed, what assumptions drive the outcome, what choices could change the trajectory, and who must be involved before we act?

That difference changes the design of AI systems.

A bank using AI for climate risk should not only forecast asset exposure under a chosen scenario. It should test alternative transition pathways, regulatory shocks, insurance responses, borrower adaptation, regional inequality, and portfolio strategy. A logistics firm should not only predict port congestion. It should simulate supplier diversification, political disruptions, fuel constraints, labor bottlenecks, and customer service impacts. A city government should not only estimate traffic flows. It should model housing, health, emissions, equity, public acceptance, and long-term maintenance.

The paper is written for policymaking, but business strategy has quietly become policy-like. Large firms now make decisions that affect labor markets, energy systems, urban patterns, public trust, cybersecurity, and information ecosystems. Pretending these are purely internal optimization problems is efficient in the same way ignoring the fire alarm is efficient.

The evidence is conceptual, but not empty

Because this is a position paper, its evidence does not work like a controlled experiment. There are no ablation studies, no benchmark comparisons, no appendix tables testing alternative parameter settings. The paper’s evidence is a synthesis of existing methods, examples, and conceptual principles.

That means readers should interpret its claims carefully.

This matters because the article should not oversell the paper as if it discovered a new algorithm. Its contribution is a framing contribution: it organizes scattered developments into a coherent field and attaches governance principles before the machinery becomes normal.

That is valuable. Conceptual work is often underrated because it does not arrive with a GitHub repo. But in emerging fields, naming the system correctly can prevent years of bad architecture. The wrong name invites the wrong product. “AI forecasting” leads buyers to ask for accuracy. “Responsible computational foresight” pushes them to ask about assumptions, stakeholder inclusion, feedback loops, contestability, and long-term consequences.

Less sexy. More useful. Tragic, really.

Responsible foresight has principles, not just tools

The paper’s principles are not decorative ethics. They are constraints on how AI foresight systems should be built and judged.

The principles include sustainability, equity, intergenerational justice, precaution, ethical consideration, inclusion, empowerment, accountability, transparency, systems thinking, adaptability, continuous monitoring, scientific rigor, and data integrity. That is a long list, but it can be reduced into three business design questions.

First, whose future is being modeled?

A foresight system that only reflects the assumptions of executives, regulators, investors, or technical experts will miss lived constraints. This is especially important when AI tools are used in urban planning, labor policy, healthcare access, financial inclusion, or infrastructure decisions. Stakeholder exclusion does not merely create moral discomfort. It creates bad models because it hides information.

Second, which feedback loops are visible?

Many strategic failures come from treating an intervention as a one-step action. A price change affects customer behavior. Customer behavior affects competitors. Competitors affect supply chains. Supply chains affect regulation. Regulation affects reputation. Reputation affects adoption. Somewhere in this loop, a consultant adds a slide called “Key Takeaways,” and everyone pretends the loop has ended.

Responsible computational foresight insists that decisions sit inside systems. AI can help map those systems, but only if the organization asks it to look beyond immediate outputs.

Third, can the model be challenged?

Transparency here does not mean dumping model documentation into a folder nobody opens. It means decision-makers can inspect assumptions, test alternatives, question data, and understand why one scenario appears preferable. The paper’s emphasis on hybrid intelligence and HCI is important because the interface determines whether humans actually deliberate or simply click “approve” with better typography.

The business value is better strategic judgment, not automatic certainty

A responsible computational foresight system can create value, but not always in the tidy ROI language preferred by budget committees.

Some value comes from avoided mistakes: not overcommitting to one scenario, not ignoring vulnerable stakeholders, not selecting a policy or strategy that optimizes one metric while damaging the system around it. Some value comes from better preparedness: identifying early signals, building adaptive plans, and rehearsing decisions before conditions become urgent. Some value comes from legitimacy: involving stakeholders earlier and making assumptions visible.

For companies, the most realistic pathways are:

The key business inference is this: AI foresight is not mainly a prediction product; it is a decision-quality product.

That distinction changes how success should be measured. Accuracy still matters when forecasting is part of the task. But for foresight, better metrics include assumption diversity, scenario coverage, stakeholder representation, traceability of evidence, speed of strategic iteration, quality of contingency planning, and whether decision-makers update plans as new information arrives.

A forecast can be wrong and still useful if it reveals which assumptions matter. A scenario can be unlikely and still useful if it exposes a vulnerability. A simulation can be imperfect and still useful if it prevents a catastrophic oversimplification. This is the annoying truth of strategic work: usefulness is not the same thing as certainty.

The paper’s boundary: it is a map, not a machine

The paper’s main limitation is straightforward: it is a conceptual synthesis, not an implementation study.

It does not prove that responsible computational foresight improves policy outcomes. It does not provide a software architecture. It does not quantify the marginal benefit of participatory AI scenario generation versus traditional expert workshops. It does not resolve conflicts between values, such as speed versus inclusion, transparency versus complexity, or precaution versus innovation.

That does not weaken the paper so much as define how it should be used.

Use it as a lens for designing foresight capability. Do not use it as proof that buying an AI simulation platform will make strategy responsible. Responsibility is not installed. It is institutionalized through process design, governance, incentives, participation, review, and the ability to revise decisions when reality refuses to cooperate.

The paper also leaves open a hard question: how should organizations integrate these tools without producing foresight bureaucracy? Too little process turns AI into a prediction toy. Too much process turns foresight into a workshop industry where everyone produces scenarios and nobody changes decisions. The useful middle is still underdeveloped.

For Cognaptus readers, that is the practical frontier. The future of AI foresight will not be determined only by model capability. It will be determined by whether organizations can build workflows where AI-generated insights actually change agendas, investment choices, risk controls, and stakeholder conversations.

What leaders should take from this paper

The paper’s strongest message is not “AI can predict the future.” It is almost the opposite.

AI becomes more strategically valuable when organizations stop asking it to collapse uncertainty too early. The future is not a single hidden answer waiting to be extracted from data. It is a contested space of possibilities, constraints, incentives, and choices. Prediction has a role inside that space, but it should not own the space.

A mature AI foresight system should therefore do five things:

1. Generate forecasts where the question is forecastable.
2. Build scenarios where the question is open-ended.
3. Simulate interventions where systems are interconnected.
4. Invite human contestation where values and assumptions matter.
5. Monitor outcomes so foresight becomes iterative rather than ceremonial.

The paper’s phrase “responsible computational foresight” may sound academic, but the underlying business idea is refreshingly concrete: AI should help organizations rehearse the future before they spend real money, political capital, institutional trust, or human lives on a strategy they barely understood.

That is not prediction. That is preparation with agency.

And yes, it is less glamorous than a machine that tells executives exactly what will happen next. But on the bright side, it has the advantage of not being nonsense.

Cognaptus: Automate the Present, Incubate the Future.