A Structural Analogy 

Contemporary autonomous driving systems—especially those celebrated as “advanced” due to their reliance on large-scale neural networks and implicit world models—are often evaluated through a misleading anthropomorphic lens. They are described as learning to drive, understanding the road, or approaching human-level intelligence. Such language obscures a fundamental truth: these systems do not possess a human-like cognitive structure, nor are they converging toward one.

A more accurate analogy would be the following:
today’s leading autonomous driving systems resemble an illiterate, socially impaired veteran driver—highly skilled in execution, yet fundamentally incapable of understanding rules, norms, or social meaning.

This analogy, though provocative at first glance, is not rhetorical exaggeration. It is a precise description of the structural characteristics of implicit-model-based autonomy.

---

1. Illiteracy Without Blindness: The Absence of Symbolic Understanding

To call such a system “illiterate” is not to claim it cannot see. On the contrary, modern autonomous vehicles possess perception systems far exceeding human sensory bandwidth. Cameras, LiDAR, radar, and sensor fusion provide a rich, continuous representation of the environment.

What is missing is symbolic literacy.

Implicit world models encode correlations in high-dimensional continuous space. They do not represent traffic rules as explicit propositions, norms, or symbols. A red traffic light is not understood as “a legally binding prohibition enforced by social institutions” but as a visual pattern statistically associated with braking behavior in training data.

The system therefore does not know traffic rules—it merely behaves as if it does, within a learned distribution.

This is illiteracy in the strict sense:
not a lack of perception, but a lack of rule-based, communicable, and reason-giving understanding.

---

2. Why It Cannot Pass the Written Driving Test

Human driving licenses require written examinations for a reason. The written test does not measure motor skill; it measures normative comprehension:

• knowledge of rules

• understanding of exceptions

• reasoning about responsibility and liability

• interpretation of ambiguous scenarios

Implicit-model-based systems fail this test categorically—not contingently.

They lack:

• deontic representations (ought / ought not)

• counterfactual legal reasoning

• explicit distinction between lawful, unlawful, and socially tolerated but illegal behavior

Even if such a system drives flawlessly for millions of kilometers, it does not understand why a behavior is correct, nor can it justify or explain it.

It cannot pass the written test not because it is undertrained, but because it lacks the representational substrate the test presupposes.

---

3. Social Blindness: Driving Without Shared Intentionality

Real-world driving is not governed solely by formal rules. It is saturated with social negotiation:

• eye contact at intersections

• subtle gestures of yielding

• mutual prediction of intent

• informal cooperation and exception handling

Human drivers constantly operate within a shared social-cognitive field. They model not only the physical environment, but the beliefs, expectations, and intentions of others.

Implicit world models do not possess this capability. They treat socially meaningful deviations as statistical anomalies rather than intentional acts. As a result, they exhibit a form of functional social blindness.

This is why the analogy to autism—used here in a strictly technical, not clinical or moral sense—is appropriate:
the system can act competently in structured environments, yet fails to participate in socially grounded meaning-making.

---

4. Why Such Systems Are Still Called “Advanced”

The persistence of the “advanced” label reflects an engineering success, not a cognitive one.

These systems are:

• computationally powerful

• data-efficient at scale

• highly optimized within bounded distributions

• economically deployable

They represent the peak of operational intelligence, not interpretive or normative intelligence.

The confusion arises when performance is mistaken for understanding, and when human cognitive terminology is applied to non-human architectures without qualification.

---

5. Not Immature—But Structurally Limited

It is tempting to say these systems are “on the path” toward human-like understanding. This framing is incorrect.

The limitation is not developmental but architectural.

Implicit world models are not incomplete versions of human cognition; they are a different category of system altogether. No amount of additional data or training time will spontaneously yield symbolic reasoning, normative awareness, or social intentionality unless the system’s representational structure is fundamentally altered.

They are not students who have not yet learned the rules.
They are entities incapable of knowing what a rule is.

---

Conclusion

The analogy of the illiterate, socially impaired veteran driver is not a critique of autonomous driving technology’s usefulness. It is a correction of the narrative surrounding its nature.

Modern autonomous driving systems:

• can drive skillfully

• can outperform humans in constrained scenarios

• can operate safely under defined conditions

But they do not understand traffic, law, or society—and they are not becoming the kind of systems that do.

The real danger lies not in what these systems cannot do, but in mistaking their competence for comprehension, and in projecting human cognitive categories onto machines that do not—and structurally cannot—inhabit them.

Recognizing this distinction is not pessimism.
It is the first step toward responsible design, governance, and deployment.