Cinematic Strawberry

 Logo

The Deep Understanding Principle

For decades, the quest to develop human-level artificial general intelligence (AGI) has been framed primarily as an information processing challenge. Could we create systems capable of taking in data, reasoning over representations, and producing intelligent outputs in ways that match or exceed human cognitive capabilities? This has driven approaches ranging from symbolic logic systems to machine learning statistical models over vast datasets. However, a groundbreaking new philosophical perspective argues that such approaches, even those using extremely complex brain-inspired architectures, fundamentally miss what is essential for genuine understanding and meaningful cognition. The ability to make qualitative judgments about meaning, rooted in qualitative first-person subjective experiences or "qualia", may be an indispensable requirement that information processing models alone cannot satisfy. The Deep Understanding Principle captures this insight.

The Principle in Plain Terms

Picture someone who has read every cookbook ever written and has never eaten. Ask what goes with lamb and the answer is excellent, because the answer is in the books. Ask whether this particular sauce has gone too sharp and there is nothing to consult. The first question is settled by relations among words. The second needs a taste.

The Deep Understanding Principle holds that a system trained only on relations among symbols stays in the first position permanently, and that part of what people mean by understanding is the second thing. Searle (1980) built the original case around a person in a sealed room following rules for arranging Chinese characters, producing answers indistinguishable from a speaker's while understanding nothing. Harnad (1990) gave it its sharpest form: a dictionary defines each word by other words, and no route through it reaches the world unless some symbols are pinned to something that is not a symbol. Bender and Koller (2020) restated the point for systems trained on text alone, arguing that form carries no path to meaning however much of it there is.

The principle takes a side in a long dispute. Dennett (1988) argued that qualia, understood as intrinsic properties of experience available to direct inspection, dissolve under examination and can be redescribed in terms of dispositions and reactions. The Deep Understanding Principle holds the opposite: the qualitative dimension is what fixes meaning, and a system without it manipulates form alone.

Qualitative Experience as the Cornerstone of Understanding

The principle states that for any system or entity to achieve authentic understanding beyond mere data manipulation, it must possess the capacity for qualitative judgments grounded in qualitative phenomenal experiences. Specifically: 1) Grasping meaning and true comprehension requires the ability to make qualitative assessments about semantic content, conceptual relationships, and contextual significance - not just formalistic rule-following. 2) Such qualitative evaluations of meaning depend on having qualitative subjective experiences - a felt, first-person perspective that imbues representations with meaning rather than just processing abstract symbols. 3) Purely syntactic operations following formal rules are insufficient for genuine understanding, since they lack the qualitative grounding needed to fix intentional content.

Point three is the oldest part and the most contested. Searle (1980) argued it from the room, and the standard reply, which he anticipated and rejected, is that understanding belongs to the whole system rather than to the person inside it. Harnad (1990) reframed the dispute in a way that suggests a research programme instead of a stalemate: symbols acquire content by being anchored in sensorimotor categories learned from interaction with the things they name. That version is testable, and it does not require phenomenal experience, which is where it parts company with points one and two.

Exploring AI and Qualia

Universe 00110000

Implications for Artificial Intelligence

Any computational model claiming general intelligence and natural language understanding needs an account of how qualitative experience, grounded representation, or functionally equivalent semantic grounding enters its architecture. Conversely, progress on what Chalmers (1995) named the hard problem, why physical processing is accompanied by experience at all, could clarify what artificial systems need for real world understanding beyond narrow capabilities or behavioral mimicry. The requirement therefore divides, and the division is what makes it workable. Sensorimotor grounding is directly testable: a system either anchors its symbols in categories learned from interaction with the things they name, or it does not, and the difference shows up in performance on cases its training text never covered. The qualitative half is where Nagel (1974) applies, since what it is like to be a bat stays out of reach of any description framed from outside, which marks it as the open frontier rather than a defeater for the testable half. Without such an account, there remains a serious explanatory gap in how a system genuinely comprehends semantic content, context, and conceptual relationships in the way humans do naturally. Information processing alone may be insufficient. This places qualitative subjective experience among the central unsolved problems for achieving human-level artificial intelligence.

Visual representation of The Deep Understanding Principle

Universe 00110000

Conclusion

The Deep Understanding Principle puts qualitative judgment at the centre of comprehension. A system that manipulates symbols by rule alone has no way to fix what those symbols are about. Searle (1980) made the case from inside the room, Harnad (1990) sharpened it into the symbol grounding problem, and Bender and Koller (2020) restated it for systems trained on text, where form supplies no route to meaning however much of it there is.

The principle turns that into a requirement. Any architecture claiming general understanding has to say where grounding enters, and the two live routes are sensorimotor anchoring of symbols in the things they name, and the qualitative first-person dimension that Chalmers (1995) named the hard problem and Nagel (1974) argued no third-person description reaches.

This moves qualia from the margin of AI research to its centre. The systems that matter next are the ones built to answer that question rather than to route around it.

References

❮
❯