Susser, D., Roessler, B. and Nissenbaum, H. (2019). Technology, autonomy, and manipulation. Internet Policy Review 8(2). DOI 10.14763/2019.2.1410.
Key takeaways
From the reading, NOTE-745.
- The definition (p. 4). “manipulating someone means intentionally and covertly influencing their decision-making, by targeting and exploiting their decision-making vulnerabilities.” Covert means “in a way they aren’t consciously aware of, and in a way they couldn’t easily become aware of”.
- Openness is the line. Persuasion and coercion are “forthright”. Deception is one species of manipulation; playing on desires, emotions and cognitive biases are others. Overt temptation and guilting are failed attempts: “it is the attempted covertness that is central” (p. 5). Nudges manipulate when hidden and exploiting a bias, not when informational.
- Why technology suits it (pp. 6–7). Surveillance exposes general and “person-specific” vulnerabilities; platforms are “adaptive choice architectures”; habituated technology becomes “invisible”. Not every dark pattern or gig-app prompt manipulates, since some are overt.
- The harm is to autonomy (pp. 9–10). Manipulation leads people to act “toward ends they haven’t chosen” and “for reasons not authentically their own”. The economic harm is derivative.
- Beneficent manipulation still harms (pp. 10–11). The harm is to “the process of decision-making, not its outcome”: the target is made “opaque to himself”. But the paper claims only that manipulation “always carries a harm”, not that it is always wrong.
- Scope. Individual decisions only; manipulation of institutions and systems is set aside (p. 3).
Standing in the record
Filed on 2026-10-12 at the owner’s request with QUESTION-070 and QUESTION-071, and read in NOTE-745. It is held as the standard statement of the covert-influence account, which Noggle’s survey (LIT-1007) quotes and objects to. Its harm argument is filed as THEORY-235, of covert influence rather than of manipulation as such.
For QUESTION-070 it supplies the definition with the intent requirement that the question tests. It never examines that requirement, and it treats firms as the intending manipulators even where the influence is chosen by an adaptive system. For QUESTION-071 its criterion is historical and epistemic: whether the agent was, and could easily have become, aware of the influence. It places the harm in authenticity, the family of the set-aside CLAIM-016. It does not address the global manipulation of the four-case argument.
No instruction for machine-learning practice; nothing here belongs in the anthology.