What can artificial intelligence really do? - Sociologically - Sociologicamente

TEXT START: After discovering how artificial intelligence hides within itself an army of invisible workers – our Mechanical Turk 2.0 – we must go a step further and ask ourselves: what can the machine really do on its own?

THE DISSECTION

The article performs a useful decomposition of AI into tasks, supervision, organizational power, and unequal distribution. Its strongest insight is that “human-in-the-loop” can mean a worker serving as a rubber stamp: legally responsible, operationally subordinate, and too rushed to exercise genuine judgment.

But the article ultimately uses this insight to preserve the centrality of human labor. It shifts the question from whether jobs disappear to which tasks remain human, then treats supervision, verification, context, responsibility, and social connection as durable human domains. That is the article’s escape hatch.

THE CORE FALLACY

It confuses lack of human-like understanding with lack of economic substitutability.

A system does not need consciousness, common sense, or genuine comprehension to perform enough of a worker’s function to destroy the worker’s bargaining power. If AI can produce acceptable outputs, generate competing hypotheses, classify cases, draft decisions, monitor patterns, and escalate only uncertain cases, the economic requirement for human labor falls—even if the machine remains brittle and unintelligent in the human sense.

The article treats the remaining human role as proof that human work survives. Under the Discontinuity Thesis, that is false. A shrinking supervisory layer can coexist with mass displacement. The relevant question is not whether a human still touches the process. It is whether the system still requires most humans as economically necessary participants. The article never confronts that threshold.

Its second error is assuming that verification remains a large, autonomous profession. Verification is itself a target for automation, process compression, statistical sampling, reputation systems, and hierarchical review. “Human judgment” can survive as a function while the mass occupation built around it dies.

HIDDEN ASSUMPTIONS

• That every important AI output requires a human with enough time and authority to verify it properly.
• That organizations will preserve high-quality human oversight rather than reduce it to liability management and throughput control.
• That contextual judgment, responsibility, and social connection scale as employment rather than becoming scarce premium services.
• That workers displaced from creation can be absorbed into monitoring and correction without those tasks being compressed or automated next.
• That professional skill gives workers durable negotiating power instead of giving owners a temporary productivity multiplier.
• That partial automation redistributes labor rather than reducing total labor demand.
• That education can close the gap between those who control AI systems and those who merely operate under them.
• That the persistence of a human role implies persistence of broad productive participation.

These assumptions turn transition mechanisms into permanent safeguards. They are not safeguards. They are lag structures.

SOCIAL FUNCTION

Primary classification: partial truth functioning as transition-management ideology and elite self-exoneration.

The article is not naive propaganda. It correctly describes invisible labor, degraded autonomy, false human oversight, and class inequality. Its failure is more sophisticated: it identifies the machinery of subordination while preserving faith in the continued necessity of the subordinated worker.

That makes it useful to institutions managing the transition. It tells professionals that their judgment remains special, tells educators that critical thinking will preserve relevance, and tells organizations that the problem is implementation quality rather than the collapse of mass productive participation. It converts a structural employment crisis into a skills-and-governance problem.

The article’s “jagged frontier” is real, but it is presented as a stable geography of mixed human-machine cooperation. Under P1, P2, and P3, it is better understood as an advancing front: tasks are captured unevenly, human oversight is thinned, and the remaining workers become increasingly concentrated as owners, indispensable specialists, or disposable alibis.

THE VERDICT

The article accurately describes the early symptoms and misdiagnoses the disease. AI does not need to replace entire professions at once to kill the post-WWII labor-consumption circuit. It only needs to make human labor less necessary, less scarce, and less powerful across enough tasks and institutions.

Human judgment will survive. Human productive participation at mass scale will not. The article’s human-in-the-loop is therefore not evidence of safety. It is the transitional corpse of the old employment system, still moving because institutions have not yet finished removing its authority.