AI systems trained primarily on human cognitive products may be bounded by existing human conceptual frameworks.
Computation alone (the argument goes) cannot instantiate or discover novel conceptual primitives without an experiencing agent to map physical reality to symbols and back.
This is the most philosophically loaded section of the report. It is also one of the cleanest arguments for why current AI may plateau at human-level even if scaled indefinitely.
"What would be the capability of a modern foundation model if it were trained on the same vast quantity of tokens, but the content were restricted to the scientific knowledge of pre-industrial, pre-Newtonian times? It seems highly improbable that the system could reason its way to the laws of general relativity, let alone quantum mechanics, while lacking the conceptual primitives of calculus, universal gravitation, or electromagnetism."
This is the abstraction-barrier argument in one paragraph. Current models recombine — they don't seem to invent new concepts from raw sensory data the way humans did over centuries.
"Current models lack a mechanism to discover the concepts of 'force' or 'causality' from scratch. They inherit these by successfully ingesting large amounts of data generated by an intelligence (us humans) capable of extracting novel concepts from non-language data."
If true, this caps any single AI model's intelligence near human level — but not necessarily collective ASI via 09 - Pathway 4 — Multi-Agent Collectives. Scaling and group formation could still produce ASI even with individual instances capped at human concepts.
If novel concept discovery requires interaction with physical reality (not just static data), this introduces a physical, linear slowdown into any recursive self-improvement loop:
This is the most concrete way the intelligence explosion could be quietly throttled.
"...the transition to ASI may require a shift toward systems that extend current capabilities by forming novel abstractions directly from raw sensor data and refining world models through active, grounded interaction with the physical environment."
I.e., paradigm-shift territory (07 - Pathway 2 — Paradigm Shifts): interactive learning + RL grounded in real or simulated physics, not language-bounded pretraining.
The report quotes Demis Hassabis directly:
"If we went back to the time of Einstein in 1900, early 1900s, could an AI system actually come up with general relativity with the same information that Einstein had at the time? And clearly today, the answer is no [...] there's still something missing."
This is Boden's transformative creativity — discovering new conceptual spaces, not just new elements within existing ones. See 15 - Is Superintelligence Super-Creative.
Most discussion of AI limits is quantitative (data, compute, energy). The abstraction barrier is qualitative — it's a claim about a structural limit on the kind of intelligence the current paradigm can produce.
If it's right, scaling alone never gets to ASI in the transformative sense, regardless of how much compute you spend.