Artificial intelligence as a test of agency: from the outsourcing of cognition to the infrastructure of agency in light of A. Grayling's The Challenge of the Future

• • 🇵🇱 Polski
Artificial intelligence as a test of agency: from the outsourcing of cognition to the infrastructure of agency in light of A. Grayling's The Challenge of the Future

Introduction

Artificial intelligence has ceased to be merely a supportive tool. Today, it is becoming an infrastructure of agency, assuming interpretive and decision-making functions.

This article analyzes the risk of losing human subjectivity through the excessive outsourcing of cognition. You will discover why AI's technical proficiency does not equate to moral autonomy.

We will examine the problem of alignment, the trap of nominal agency, and ways to protect the constitutive functions of being human in the era of automation.

AI is Shifting from Cognitive Support to an Infrastructure of Agency

Unlike calculators or search engines, modern AI systems do not simply provide data; they synthesize knowledge and plan actions. This represents a transition from outsourcing labor to delegating cognitive processes.

The key difference lies in the shortening of the control chain. An agentic system can independently manage email or finances, increasing the risk that errors will materialize before human intervention occurs.

An example is the distinction between an AI suggesting text and a system executing multi-step operations on behalf of a user. This shifts the focus from questioning the machine's intelligence to questioning the scope of our delegation.

The Alignment Problem as a Conflict Between Metrics and Values

Alignment is the problem of ensuring a system's goals are consistent with human interests. This is not merely a technical challenge, but a profound normative and axiological conflict.

Systems optimize for measurable metrics without understanding values. AI may increase sales through manipulation, fulfilling a formal objective while violating the user's ethical assumptions.

This problem stems from human normative chaos. Before we can program a machine, we must precisely define concepts such as justice or well-being—a task that often proves impossible.

Gradual Cognitive Delegation Weakens Human Agency

Daily reliance on AI leads to automation bias. When a system rarely makes mistakes, we stop verifying its output, which erodes our vigilance and capacity for critical judgment.

This leads to the phenomenon of the hollowing out of responsibility. The human remains in the loop only ceremonially, signing off on decisions they do not fully comprehend.

In the public sphere, simply labeling content as synthetic is insufficient. We need a pluralistic infrastructure of trust and education to distinguish fluid AI synthesis from factual reality.

Conclusion

The greatest challenge is not building a safe machine, but protecting the constitutive functions of humanity: the ability to understand, to question, and to take responsibility.

We must define inalienable boundaries that cannot be outsourced. Otherwise, we risk becoming mere nominal approvers of processes steered by algorithms.

Our subjectivity will be determined by where we categorically refuse to let AI replace us.

📚 Based on

The Challenge of the Future

👤 About the book's author

A C Grayling

Northeastern University London

Anthony Clifford Grayling (born 1949) is a British philosopher, author, and public intellectual. He is the founder and former Master of the New College of the Humanities (now Northeastern University London) and a supernumerary fellow of St Anne's College, Oxford. Previously, he served as Professor of Philosophy at Birkbeck, University of London. Grayling's academic work encompasses epistemology, philosophical logic, humanist ethics, and the history of ideas. A prominent public defender of secularism, human rights, and democratic principles, he has written and edited more than thirty books and contributed widely to newspapers, journals, and broadcast media. His scholarship bridges technical philosophy and public discourse, exploring fundamental moral dilemmas and philosophical traditions to address contemporary global issues.

Mind map: Artificial Intelligence as a Test of Agency

📖 Glossary

Problem alignmentu
Wyzwanie polegające na zapewnieniu, aby cele i zachowania systemu AI były zgodne z intencjami człowieka i wartościami społecznymi.
Autonomia operacyjna
Zdolność systemu do wykonywania zadań bez ciągłej interwencji operatora, niepowiązana ze świadomością czy moralnością.
Wydrążenie odpowiedzialności
Sytuacja, w której człowiek formalnie podpisuje decyzję AI, ale w rzeczywistości nie rozumie jej przesłanek i nie może jej realnie zweryfikować.
Jaźń nominalna
Podmiot, który jedynie imituje cechy osoby (w przypadku AI) lub człowiek, którego sprawstwo zostało zastąpione przez algorytmy.
Proweniencja informacji
Możliwość prześledzenia pochodzenia danych i treści, co pozwala odróżnić fakty od syntetycznych halucynacji AI.
Capability approach
Podejście oceniające technologię nie przez pryzmat jej wydajności, lecz przez to, czy realnie zwiększa ona możliwości i wolność człowieka.

Frequently Asked Questions

How does modern artificial intelligence differ from previous technological tools in the context of human action?
Modern artificial intelligence differs from previous tools by shifting the scope of delegation from the outsourcing of strength and simple calculations to the outsourcing of knowledge synthesis and decision-making. AI systems do not only assist in searching for information, but participate at the center of human agency by interpreting, planning, and executing multi-step actions on behalf of the user.
1. Translate each of the 10 numbered fragments into ENGLISH. Return a JSON {"translations":[...]} with EXACTLY 10 elements, in the same order (element i = translation of fragment i). Preserve the meaning.
2. The alignment problem consists of the risk of divergence between the AI system's goal and human interests, resulting from a lack of consistency between intention, formal goal, system behavior, and its social consequences. It is not merely a technical challenge because it stems from human normative chaos and the fact that the capacity for optimization is not identical to the capacity for valuation.
3. In what way can daily use of AI lead to a loss of real control over thought and decision-making processes?
4. Daily use of AI leads to a gradual shift in agency, where humans consciously delegate individual cognitive tasks to the system. As a result, excessive reliance on the tool's high effectiveness weakens vigilance and the ability to verify content, which can lead to the uncritical acceptance of results and a loss of autonomy in evaluation processes.
5. Why does simply keeping a human in the decision loop (human in the loop) not guarantee real responsibility for AI actions?
6. Simply keeping a human in the loop may be merely ceremonial if the formally responsible person lacks the time or competence to independently verify the AI's decisions. This leads to so-called 'responsibility hollowing,' as true control requires the ability to understand the premises and challenge the result, rather than just approving the outcome.
7. Is simply labeling content generated by AI enough to protect us from disinformation and the loss of subjectivity in the public sphere?
8. No, simple labeling is not enough because a label informs about the origin of the material, not its truthfulness. To protect the public sphere from disinformation and the erosion of trust, parallel fact-checking systems, media education, and a pluralistic architecture of institutional verification are essential.
9. How can law and institutions keep pace with the rapid development of AI to effectively protect citizens from various types of threats?
10. To effectively protect citizens, the law should become more adaptive, relying on risk and function principles, update mechanisms, and specialized institutions capable of continuous learning. It is necessary to correct the asymmetry of incentives by introducing responsibility norms, audits, safety standards, and transparency obligations. A reliable policy must simultaneously address different types of threats: from current abuses and failures to systemic risks and hypothetical future scenarios.
How can AI threaten human freedom beyond direct coercion, and how can this be countered systemically?
AI threatens freedom by creating a digital panopticon, where the awareness of being observed and profiled leads to the internalization of norms and the manipulation of behavior below the level of consciousness. This can be countered systemically by introducing legal barriers, ensuring informational autonomy, and shifting business models toward fiduciary agents who are legally obligated to act in the user's best interest.
Does universal access to AI equalize cognitive opportunities, and what determines whether this technology actually empowers humans rather than limiting them?
Universal access to AI does not automatically equalize cognitive opportunities, as effective use of the tool requires domain knowledge and the ability to critically evaluate results. Technology empowers humans when it increases their capacity for action without simultaneously stripping away their understanding of the goal, control over the means, the ability to contest the outcome, or the option to opt out of the system.
How can we prevent AI control from being seized by a narrow group of experts and corporations?
Technical authority must be separated from political authority, so that experts establish facts but do not independently define the values and principles governing the use of AI. Preventing monopoly requires creating a pluralistic ecosystem based on independent audits, oversight by regulators and social organizations, and institutions that ensure the correctability and transparency of decisions.
What specifically should be protected from automation to prevent humans from becoming merely nominal approvers of AI decisions?
Functions constitutive of the person must be protected, such as the ability to understand and question, the authorship of one's own goals, and responsibility. It is also important to preserve mutual recognition and participation in real relationships.

Related Questions

🧠 Thematic Groups

Tags:

More in: Szkatułka kosztowności

 Content is created by Fundacja Dobre Państwo.
Edited and published by APA ONE, the Foundation's own AI-based editorial system.