Deepfakes at Face Value: Image and Authority

Based on the paper "Deepfakes at Face Value: Image and Authority" by James Ravi Kirkpatrick, here is a detailed technical summary covering the problem, methodology, key contributions, results, and significance.

1. Problem Statement

The paper addresses a critical gap in the ethical literature regarding deepfakes (synthetic media generated via deep learning). Existing ethical frameworks primarily rely on harm-based accounts, arguing that deepfakes are wrong because they cause:

• Moral harms: Psychological trauma, reputational damage, blackmail, and financial loss.
• Epistemic harms: Erosion of trust in media, distortion of public discourse, and the undermining of "epistemic backstops" (recordings as proof).

The Core Issue: These harm-based accounts fail to explain the moral wrongness of deepfakes in "bare wronging" scenarios. These are cases where a deepfake is created but:

1. Never distributed (remains private).
2. Causes no tangible harm to the victim (no reputational damage, no psychological trauma).
3. Is created even when the source material was publicly available.

The author argues that if deepfakes are only wrong when they cause harm, then creating a private, non-harmful deepfake of a public figure would be morally permissible. However, the author posits that most people intuitively feel such an act is still a moral violation. The paper seeks to identify the specific normative interest violated in these "harmless" cases.

2. Methodology

The paper employs normative ethical analysis and philosophical argumentation structured as follows:

• Critical Review: The author systematically critiques existing accounts (Harm-Based, Gender Oppression, Privacy Rights, and Social Identity Control) to demonstrate their inability to explain "bare wrongings" (e.g., private deepfakes or veridical deepfakes of consenting actors).
• Thought Experiments: The author utilizes specific case studies to test hypotheses:
  • Pornographic Deepfake: A private deepfake created from public photos where the victim never knows and no distribution occurs.
  • Pornstar's Deepfake: A deepfake of a porn actor that is veridical (accurate to a real scene they consented to) but created without their specific consent for the new synthetic generation.
  • Forged Cheque Analogy: Used to illustrate that the wrong lies in the usurpation of authority (forging a signature) rather than the content (the debt was real).
• Theoretical Synthesis: The author integrates Owens' theory of "bare wrongings" and normative interests (interests in controlling the normative landscape, such as rights and obligations) to construct a new ethical framework.
• Distinction Analysis: The paper draws sharp distinctions between depiction (artistic interpretation), simulation (algorithmic generation), and private mental acts to define the scope of the proposed theory.

3. Key Contributions

The paper introduces the Authority Interest View and the concept of Algorithmic Conscription.

A. The Authority Interest View

The central thesis is that deepfakes are morally wrong because they violate a subject's authority interest: the legitimate interest in having authority over the governance of one's biometric identity and the provenance of one's agency.

• Core Claim: Individuals have a right to be the sole origin (author) of actions and performances attributed to their identity.
• The Violation: Creating a deepfake treats the subject's biometric data (face, voice) as a "generative resource" to be conscripted, bypassing their authority to determine how their identity is used.

B. The Right Against Algorithmic Conscription

The paper defines a specific right against the algorithmic conscription of identity.

• Unlike traditional "Right of Publicity" (which focuses on commercial use), this right applies to non-commercial contexts and covers the process of generation, not just the distribution.
• It distinguishes between Depiction (an artist interpreting a subject through their own labor/skill, e.g., a painting) and Simulation (an algorithm using the subject's biometric data as raw material to generate a synthetic performance). Only the latter constitutes a wrongful usurpation of agency.

C. Reframing Harm

The paper reclassifies the harms caused by deepfakes (reputation loss, psychological distress) as secondary wrongs. The primary wrong is the constitutive violation of authority (the bare wronging). The harms are downstream consequences of this initial violation of autonomy.

4. Results and Findings

The application of the Authority Interest View successfully resolves the theoretical problems that stumped previous accounts:

1. Explanation of Private Deepfakes: The theory explains why creating a private deepfake is wrong even without distribution. The wrong is the act of conscription itself, which denies the subject's autonomy and treats them as a tool, regardless of whether anyone else sees it.
2. Explanation of Veridical Deepfakes: In the "Pornstar's Deepfake" case, even if the content is accurate (the actor did perform the act), the deepfake is wrong because it usurps the actor's authority to be the source of the new digital performance. The content is accurate, but the provenance is forged.
3. Scope Limitation (The "Overreach" Defense): The theory successfully distinguishes wrongful deepfakes from permissible uses:
   • Artistic Depiction: Paintings or caricatures are permissible because they are mediated by the artist's agency, not the subject's biometric data.
   • Private Imagination: Using a photo for private sexual fantasy is permissible because it is a mental act, not an externalized manipulation of the subject's digital presence.
   • Consented Uses: Deepfakes are permissible when the subject grants authority (e.g., reviving a dead actor with estate permission, or voice restoration for ALS patients).

5. Significance

• Theoretical Advancement: The paper moves the discourse on AI ethics beyond consequentialism (harm-based) to deontological principles (rights and authority). It provides a robust philosophical foundation for why deepfakes are inherently problematic, even in the absence of detectable damage.
• Legal and Policy Implications: The findings suggest that legislation should not focus solely on preventing harm or regulating distribution. Instead, laws should prioritize protecting the right of authorship and control over biometric identity.
  • It argues for a shift from "Right of Publicity" (commercial focus) to a broader "Right to Digital Self-Representation."
  • It implies that disclaimers or watermarks do not cure the moral wrong, as the violation occurs at the moment of unauthorized generation (conscription), not at the moment of deception.
• Broader Application: The framework extends beyond AI deepfakes to other image violations like non-consensual intimate images (NCIIs/revenge porn), framing them as violations of the authority to control the provenance and audience of one's intimacy.

In summary, Kirkpatrick argues that the fundamental wrong of deepfakes is not that they lie or cause pain, but that they usurp the human capacity for self-representation, turning a person's identity into a raw material for algorithmic generation without their consent.