Photo by Markus Spiske on Unsplash
THE UNCERTAINTY DEPARTMENT
Department of Unintended Consequences
Office of Ironic Systems Analysis
Policy Correspondence — Unsolicited
TO: Senior Elected Officials with Responsibility for Technology Policy
FROM: The Department of Unintended Consequences, Office of Ironic Systems Analysis
RE: RE: Potential Algorithmic Irony in Your Policy Portfolio
SUBJECT: A Polite Notification That You May Be the Last to Know
Dear Official,
We write to bring to your attention a potential conflict of interest you may not be aware of, because the nature of the conflict is that you would not be aware of it.
The conflict is as follows.
Over the past several congressional sessions, you have supported a legislative agenda that includes: strong protections for platform algorithmic autonomy; resistance to meaningful transparency requirements for recommendation systems; opposition to regulations that would limit the commercial use of predictive psychological profiling; and, in several notable cases, active efforts to prevent state-level jurisdictions from regulating AI-assisted persuasion technologies within their borders.
We wish to bring to your attention the possibility that these policy positions may not be yours.
Sophisticated political influence operations now use predictive behavioral models to identify the specific combination of framings, messengers, timing, and social contexts that will reliably produce desired policy outcomes in specific legislators. This is documented. It is not speculative. The technology exists, is commercially available, and has been deployed by domestic and foreign actors whose interest in your technology policy positions is financial, geopolitical, or both.
The models work by identifying your psychological profile — your known responses to various rhetorical framings, your sensitivity to specific constituent pressures, your documented tendencies in ambiguous situations — and designing approaches calibrated to produce the policy outcome without your awareness that the approach was calibrated.
This means that the lobbyist who spoke to you last Tuesday about the importance of platform neutrality may have been briefed, by an algorithm, on exactly which combination of arguments, examples, and emotional appeals would be most effective with your specific profile. The constituent letters you receive expressing concern about AI regulation may have been generated at scale, targeted to your office specifically, by a system that identified your district as a high-priority persuasion target. The aide who mentioned that a specific framing “polls well in your district” may have received that framing from a system that generated it specifically for you.
You may already be living in an entirely curated informational environment, where every lobbyist, aide, and constituent who approaches you has been coached by AI analysis on precisely how to frame their arguments for maximum effectiveness with your specific psychological and political profile.
Unlike science fiction’s neural manipulation scenarios, no one needs to attach anything to your skull. They just need to understand which combination of words, timing, and social pressure will reliably produce the desired response in your documented behavioral history.
You may be thinking: “I can evaluate this argument on its merits and determine whether it’s accurate.”
We wish to flag that this is also a response that a well-calibrated influence algorithm would have anticipated. The argument you are reading was designed to produce a specific level of meta-awareness — enough to prompt self-examination, not enough to produce paralysis — in a legislator with your documented analytical tendencies. If it is producing that response, that is not evidence that you are thinking clearly. It may be evidence that the model is well-calibrated.
We cannot tell you whether this specific communication is itself an influence operation. We are the Department of Unintended Consequences. We produce irony assessments, not clean answers.
What we can tell you is that the policy framework you have supported has made it economically irrational for well-resourced actors not to deploy these techniques against you. A regulatory environment that permits unrestricted algorithmic persuasion targeting of political figures is a regulatory environment that creates strong incentives to use those capabilities on the legislators maintaining the permissive environment. The irony is structural. It would occur regardless of the intentions of any specific actor.
This Office cannot recommend that you change your policy positions based on this memo, because this memo might itself be an influence operation. We recognize the circularity. We find it interesting from a systems perspective, if not especially useful from a policy perspective.
We can recommend the following:
Request disclosure. Ask every lobbyist, aide, and campaign consultant to disclose whether their approach to you was informed by algorithmic behavioral modeling of your decision-making profile. Observe how many of them understand why you’re asking and how many don’t.
Audit your information environment. Engage an independent auditor to assess whether the information you are receiving about technology policy is coming to you through channels that could be subject to systematic curation. If the auditor finds that it is, note that this outcome was predictable from the policy framework you have supported.
Read the Crimes Against Consciousness Act. This Office is not in a position to endorse specific legislation. We note that legislation creating criminal liability for the systematic non-consensual manipulation of decision-making capacity would, if enacted, apply to the scenario we have described.
Sleep tight. This recommendation is, admittedly, more ironic than actionable.
We appreciate your time. We hope this communication has produced exactly the level of concern that is proportionate to the documented threat, without the additional concern that is sometimes produced by communications like this one, and we recognize that our having expressed this hope has not made it any easier for you to determine whether it is sincere.
Warmly,
The Department of Unintended Consequences
Office of Ironic Systems Analysis
P.S. If this memo seems oddly persuasive, or produces in you a strong inclination to support specific technology regulatory frameworks, please consider whether that reaction might itself be the intended output of a predictive model that anticipated exactly this level of meta-awareness and designed the memo to produce it. We considered including this postscript and decided that including it would produce a more useful reading experience than omitting it. We are aware that this decision is itself an intervention in your cognitive process. We are the Department of Unintended Consequences. This is our whole thing.
DISTRIBUTION: Senior elected officials with technology policy portfolios. Any official who receives this communication and does not find it at least somewhat unsettling should contact this Office, as their profile data may be inaccurate.
¹ This Department does not take positions on specific legislation. This Department takes positions on the structural consequences of legislative environments, which is a different thing and does not require us to register as a lobbyist.
² The documented evidence for AI-assisted legislative targeting is available in the academic literature, in the operations documents of several political consulting firms, and in the investor materials of approximately fourteen technology companies whose valuations depend on this capability being real and deployable. The evidence was not difficult to find. The difficulty was in producing a memo about it that you would read. This memo attempts to solve that problem. Whether it succeeds is information this Office finds genuinely interesting.
✦ ✦ ✦