University of Crete (UOC)
The quest for artificial consciousness and agency: a philosophical inquiry
Abstract
dc:descriptionThis dissertation investigates whether, and under what conditions, artificial systems may be attributed consciousness and agency, and what ethical and governance consequences follow from such attributions. It argues that contemporary debate too often moves directly from behavioural sophistication to claims about consciousness, from consciousness to agency, and from agency to liability. In response, the thesis distinguishes phenomenal consciousness, access-consciousness, and self-consciousness. It uses access-consciousness in Block's technical, functionally specified sense: a content has the relevant access profile when it is available beyond its local processor to multiple decision-relevant functions and is causally capable of altering control. A warranted attribution is therefore a defeasible, truth-directed attribution of this functional organisation, not knowledge of phenomenal experience. The dissertation critically evaluates leading theories of consciousness, including first- and higher-order representationalism, Integrated Information Theory, Global Workspace Theory, Recurrent Processing Theory, and Orch OR. It distinguishes the constitutive functional criterion of global, multi-system availability and counterfactual causal efficacy from supporting architectural indicators, including recurrent integration, temporal maintenance, metacognitive regulation, and principled reportability. Interpretability, reproducibility, system-boundary specification, and intervention or ablation evidence are treated separately as conditions of epistemic warrant rather than as constituents of access-consciousness. The thesis further argues that consciousness does not by itself determine agency. Minimal artificial agency can be attributed where systems exhibit goal-directed control, adaptive revision, and structured action selection. Planning agency may operate under externally supplied ends. Policy-forming agency requires endorsement-control alignment: the system must be able to represent goals as candidate commitments, compare, retain, suspend, revise, or reject them for reasons, and allow that evaluative stance to govern later conduct. Responsibility-grade agency additionally requires critical normative competence: the ability to distinguish applying a norm from justifying it, to recognise reasons that challenge the governing framework, and to revise foundational policies when such reasons are judged sufficient. This competence does not guarantee moral correctness. Nor does it by itself establish that the system is a fitting target of blame, sanction, or legal liability. In this dissertation, “responsibility-grade agency” denotes a demanding threshold of eligibility for those further judgements, which remain relational, institutional, and defeasible. Applied to representative AI families, including classical planners, model-predictive controllers, reinforcement-learning systems, and LLM-based tool agents, the framework suggests that current systems may satisfy conditions for instrumental or planning agency while still falling short of warranted attribution of a robust functional access profile, organisational goal ownership, and critical norm-answerability. The dissertation concludes that governance should remain institutionally mediated: responsibility and liability should not be transferred to artificial systemsmerely on the basis of fluency or performance, but should instead be calibrated through explicit accountability gateways, evidential thresholds, and agency-preserving design constraints.
Degree
thesis:*- Grantor dc:publisher
- University of Crete (UOC)
- Year dc:date
- 2026
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Drikakis, Anastasia
- Δρικάκη, Αναστασία
Subjects
dc:subject × 18- Τεχνητή συνείδηση
- Συνείδηση πρόσβασης
- Τεχνητή ικανότητα για δράση
- Ηθική ευθύνη
- Αναστοχαστική ισορροπία
- Μεγάλα γλωσσικά μοντέλα
- Artificial consciousness
- Access consciousness
- Artificial agency
- Moral responsibility
- Reflective equilibrium
- Large language models
- Φιλοσοφία, Ηθική και Θρησκεία
- Ανθρωπιστικές Επιστήμες και Τέχνες
- Φιλοσοφία
- Philosophy, Ethics and Religion
- Humanities and the Arts
- Philosophy
Rights
- Language dc:language
- eng
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/62544