Futures Thinking
Written by
Sandeep Ozarde
Written by
Sandeep Ozarde
Sustainable Design
In a recent discussion on The Future of Intelligence, Demis Hassabis makes a familiar claim within the sciences: that, in principle, everything that exists in the universe is computable. It reflects a long intellectual tradition in physics and formal sciences, where progress has historically depended on abstraction, formalisation, and proof.
Yet there is an important distinction that often goes unexamined in contemporary AI discourse: the difference between what can be modelled mathematically and what can be lived, interpreted, and acted upon by humans.
This distinction matters because modern AI research increasingly treats intelligence as a problem of formal completeness, something that will emerge once models become sufficiently large, sufficiently accurate, or sufficiently grounded in physical data. The implicit assumption is that intelligence is primarily a property of representation rather than a property of situated engagement. That assumption deserves scrutiny.
This distinction helps explain a puzzle that has become increasingly visible in contemporary AI research: why systems trained on vast corpora of human knowledge still struggle to produce genuinely new discoveries.
As Yann LeCun has argued, large language models excel at pattern completion within linguistic space, but lack grounded world models. They manipulate symbols, not situations. They generate plausible continuations, not new frames of understanding.
Discovery, however, rarely emerges from recombination alone. It requires the ability to re-frame a problem, to notice that what has been formalised so far may not be the right abstraction, or that the question itself is misplaced. Historically, many breakthroughs occurred not because someone solved an equation better, but because they realised a different equation was needed — or that the phenomenon could not yet be expressed mathematically at all.
Crucially, these reframings often happen before formalisation.
They occur at the level of intuition, analogy, mental simulation, and situated judgment — capacities that operate prior to mathematical expression. Mathematics enters later, once the problem space has been stabilised enough to be formalised.
This is not a weakness of mathematics. It is a category boundary.
Cognitive science has long shown that human reasoning is not primarily deductive or formal. Humans rely on heuristics, mental models, embodied experience, and contextual cues. They reason approximately, revise beliefs mid-stream, and tolerate ambiguity where no formal solution exists.
We do not first construct equations and then act. We act, observe consequences, revise our understanding, and only sometimes formalise what we have learned.
Even in scientific practice, mathematics typically follows insight rather than precedes it. The equation is the product of understanding, not its origin.
This matters because many AI systems today invert this sequence. They begin with formal representations — tokens, vectors, objectives — and attempt to scale their way toward understanding. What is missing is not more computation, but the epistemic layer that precedes formalisation: the framing of what matters, what counts as success, and how humans will meaningfully interact with the system.
Modern discussions about intelligence, human or artificial, often default to mathematics and physics as the ultimate problem-solving tools. This is understandable. Mathematics gives us precision, formalism, and predictive power. Physics grounds those abstractions in constraints imposed by the world. Together, they have enabled extraordinary progress: from space travel to semiconductor fabrication, from climate models to protein folding.
Yet this success has quietly produced a category error.
We increasingly treat all problems as if they were formal problems, problems that can be fully specified, optimised, and solved within mathematical space. In doing so, we obscure an uncomfortable truth: most problems that matter in human systems were never purely formal to begin with.
They are ambiguous, contextual, social, value-laden, and interpreted differently by different people.
Mathematics is indispensable for solving formalised problems. But it does not explain how problems become formal in the first place—or whether they should.
That work happens elsewhere.
Before a mathematical model can be written, several non-mathematical decisions must already have been made.
What is the problem? What boundaries define it? Which variables matter—and which are ignored? What counts as success? Who is affected by failure?
None of these questions are mathematical. They are interpretive.
In real systems—healthcare, finance, governance, education, climate policy—the hardest work is not optimisation. It is framing. Two teams can use identical mathematical tools and reach radically different outcomes depending on how the problem was initially defined.
This is why debates about AI “reasoning” or “discovery” often go astray. We ask why systems trained on vast mathematical machinery and statistical regularities cannot generate meaningful scientific or social breakthroughs. The implicit assumption is that discovery itself is a computational act.
But discovery is not merely the solution of an equation. It is the recognition that a different equation is needed, or that the question itself was misframed.
Mathematics allows us to model the world. Design allows humans to inhabit those models.
A financial risk model may be mathematically sound, yet unusable by advisors and misunderstood by clients. A medical decision system may optimise outcomes statistically while undermining clinician trust or patient dignity. A climate model may be scientifically rigorous yet politically inactionable. In each case, the failure is not mathematical. It is relational.
Design operates in the space between formal models and lived experience. It translates abstractions into interfaces, workflows, narratives, and decisions that humans can understand, contest, and adapt. It negotiates trade-offs that cannot be resolved by optimisation alone because they involve competing values rather than competing variables.
Another quiet assumption worth challenging is the idea that intelligence itself is fundamentally mathematical.
Human intelligence does not operate by solving equations. It relies on heuristics, analogies, embodied experience, social cues, and context-dependent judgment. We navigate uncertainty not by optimisation, but by satisficing; not by formal proofs, but by meaning-making.
Mathematics often enters human reasoning after understanding has been achieved—when we seek to formalise, communicate, or stabilise insights that were initially grasped in non-formal ways. This is why early scientific breakthroughs are often described qualitatively before they are formalised quantitatively. The formalism follows the insight; it does not generate it.
Expecting AI systems to “think mathematically” as a path to general intelligence risks repeating the same mistake: confusing the tools of formalisation with the source of understanding.
Framing this as “math versus design” misses the point. The relationship is not adversarial but complementary.
Mathematicians formalise relationships. Physicists model constraints. Engineers operationalise systems. Designers make them usable, navigable, and meaningful in context. Each role addresses a different layer of reality. Problems emerge when one layer is mistaken for the whole.
When mathematics is treated as sufficient for human systems, we get technically correct solutions that fail socially. When design is reduced to aesthetics or branding, we lose its deeper role as a discipline of sense-making under uncertainty.
The future of intelligent systems—human or artificial—depends on recognising this division of labour, not collapsing it.
In contemporary discussions of intelligence and artificial intelligence, progress is often explained primarily through mathematics, optimisation, and physical constraints. This emphasis is understandable: formal models are essential for precision and scalability. What is striking, however, is not the presence of mathematics, but the near-absence of design as an explicit intellectual discipline in these accounts.
Mathematical models never operate in isolation. They are framed, interpreted, communicated, and acted upon within human, cultural, organisational, and ethical contexts. Design—beyond aesthetics—shapes how problems are defined, whose perspectives are included, how uncertainty is handled, how explanations are made intelligible, and how responsibility is distributed between humans and systems. Yet this work of framing and sense-making is routinely treated as secondary or implicit.
The issue, therefore, is not whether mathematics is sufficient, but why the design of problem framing, human understanding, and sense-making is so often treated as secondary or implicit. Without this layer, technically correct systems routinely fail in practice—not because the models are wrong, but because the problems were never adequately framed in human terms. Design is not an optional layer added after intelligence; it is the condition that makes problem-solving possible in the first place. Mathematics may describe the world, but design determines how that description becomes usable, interpretable, and ethically situated.
