Over the last few years, AI has moved from a "maybe one day" promise into the spotlight almost overnight. Practical systems and broad societal conversations now revolve around questions of capability, trust, and impact in ways that would have been hard to imagine a decade ago.
I see three major reasons why this rapid rise happened.
First, AI now perceives the environment in a way that resembles human action cycles: observe the state, make a decision, and take an action. Modern systems loop through perception, representation, planning, and execution, and that iterative structure is a robust recipe for adaptive behaviour. Second, AI evolved from a narrowly functional, code-centric approach, where programming was primarily the domain of specialized developers, to a semantic approach that allows non-programmers to interact meaningfully with systems. In this sense, Andrej Karpathy's quip resonates: "The hottest new programming language is English." Language interfaces and large pre-trained models have democratized access to powerful computational tools. Third, the architecture of modern AI helps us focus attention on what matters. The transformer family, whose name is summed up by "Attention Is All You Need," shifted the design of models toward mechanisms that weigh and integrate contextual signals selectively. Attention lets models prioritize relevant information and suppress noise, which in turn enables more effective reasoning over large inputs. These three elements — perceiving the world as we do, speaking our language, and helping us focus on what matters — make AI's success a natural one. Like Tolstoy's happy families, modern AI systems share a common recipe: they observe, communicate, and attend in ways that align with human cognition. This alignment is what makes them feel so seamlessly integrated into our workflows and conversations. At this point the story could continue as a success. They lived happily ever after. And here, the story should end. They lived happily ever after. But beneath that success are three persistent challenges, each summed up in a single word: resources, privacy, and comprehension. These are not abstract concerns. In projects, lectures, and workshops I regularly hear practical questions about energy consumption, data confidentiality, and the brittleness of contextual understanding — each of which directly limits how, where, and whether deployments can scale responsibly.
“Resources: The Price of Thinking Machines”
Classical computation is intrinsically dissipative: irreversible gates erase information and incur thermodynamic costs. Quantum computing, by contrast, is inherently reversible at the logical level. In theory, reversible operations allow computation to approach thermodynamic limits (Landauer's principle) and promise orders-of-magnitude improvements in energy efficiency over classical architectures, even if full realization remains an engineering challenge. Beyond energy, quantum algorithms sometimes require dramatically fewer steps to reach a result, because they can explore amplitude patterns in complex Hilbert space that are not available to algorithms constrained to real-number computation. Practically, this suggests a future in which certain classes of inference or optimization can be executed with far less energy and in fewer operations — an attractive prospect as models and data sets continue to grow.
“Privacy: The Architecture of Trust in Intelligent Systems”
Privacy concerns in AI are not merely regulatory; they are structural. Classical cryptographic techniques secure communication and storage today, but many rely on hardness assumptions that could be undermined by future quantum algorithms such as Shor's. Meanwhile, quantum information primitives offer fundamentally different guarantees: Quantum Key Distribution enables secure shared keys with security rooted in quantum measurement, quantum teleportation provides ways to transfer states without exposing intermediate information, and emerging ideas in counterfactual communication point toward protocols where information transfer need not follow ordinary channels. Combined with nascent concepts of a Quantum Internet, these primitives could enable new privacy-preserving architectures for distributed learning, model sharing, and cross-institutional research in ways that are guaranteed by physics rather than by computational assumptions alone.
“Comprehension: Beyond Optimization Toward Understanding”
Perhaps the most philosophically ambitious challenge is comprehension. Classical AI systems operate within ordered algebraic structures where comparisons like "greater-than" and local optimization dominates. From a mathematician's perspective, models expressed entirely in real numbers are inherently tied to order relations and local optima; this shapes what they can represent and how they generalize. The complex Hilbert space of quantum mechanics lacks a natural global order comparable to the reals as two complex numbers cannot be meaningfully compared with ≥, yet it supports smooth analytic structures and interference phenomena that have no classical analogue. If intelligence is partly about synthesizing global context, disambiguating overlapping signals, and exploiting non-local correlations, then the mathematical freedoms of complex spaces might provide tools for richer, more fluid models of context and meaning.
“Practical Implications: Paths Toward Implementation”
Practical Implications: Paths Toward Implementation Quantum computing offers natural paths toward resolving these challenges: reversible and amplitude-based computation for resource efficiency; quantum cryptographic primitives and quantum-internet concepts for physics-backed privacy; and the richer structure of complex Hilbert space for new models of context and comprehension. Quantum computing offers natural paths toward resolving these challenges: reversible and amplitude-based computation for resource efficiency; quantum cryptographic primitives and quantum-internet concepts for physics-backed privacy; and the richer structure of complex Hilbert space for new models of context and comprehension.
Is intelligence, in the deepest sense, complex or real?
This is not a philosophical question to be settled by argument alone. It is a practical matter, and the answer will emerge from our work — from the systems we build, the experiments we run, and the benchmarks we establish.
I will continue to explore this question at the interface of theory and practice, and invite others to join the conversation: to test, to build, and to scrutinize the answer together.
