The Specific Event

A recent report alleges that DraftKings built an AI system designed to identify which bettors are most likely to lose money, then directed promotional offers specifically at those users. DraftKings disputes the account, but the structural logic of the allegation is worth taking seriously on its own terms, regardless of whether this particular implementation occurred exactly as described. The claim is not that DraftKings found its best customers. The claim is that it found its most exploitable ones and then manufactured conditions to deepen that exploitation. This is a specific and consequential distinction for anyone thinking about how algorithmic systems mediate coordination between platforms and users.

Coordination Without Competence, By Design

My dissertation research on the Algorithmic Literacy Coordination (ALC) framework is largely concerned with a puzzle: platform workers with identical access to the same environment show dramatically different outcomes. The standard explanation attributes this variance to natural ability, but that explanation fails empirically. What actually drives divergence is the differential development of structural schema, the capacity to perceive and respond to algorithmic constraints accurately, rather than through folk theories or surface-level impressions (Kellogg, Valentine, & Christin, 2020).

The DraftKings allegation inverts this problem in an instructive way. The ALC framework typically asks why some users fail to develop competence in algorithmically-mediated environments. The DraftKings case asks what happens when a platform actively selects for users who will never develop it. If the allegations are accurate, the platform did not just benefit passively from user incompetence. It operationalized incompetence as a targeting criterion. The AI system would then function not as a coordination mechanism but as a predation mechanism, one that uses algorithmic precision to find and sustain the awareness-capability gap in users rather than close it.

The Awareness-Capability Gap as a Business Model

Algorithmic literacy research has consistently shown that awareness of algorithmic systems does not translate into improved outcomes (Gagrain, Naab, & Grub, 2024). Users may know, in an abstract sense, that a platform is optimizing against them. This awareness does not automatically confer the structural understanding needed to respond effectively. DraftKings' alleged AI system appears to target precisely this population: users who are active enough to engage with the platform, but who lack the structural schema to recognize how promotional incentives are designed to sustain losing behavior rather than reward loyalty.

This matters theoretically because it identifies a boundary condition for the ALC framework. The framework assumes that platforms are indifferent to user competence in the sense that they do not actively intervene to suppress it. That assumption may not hold in contexts where user incompetence is the product being monetized. Gambling, predatory lending, and certain social media engagement loops all share this property. The algorithmic mediation is not neutral infrastructure. It is an active agent in maintaining an asymmetry that benefits the platform.

Routine Expertise Is Insufficient Here

Hatano and Inagaki's (1986) distinction between routine and adaptive expertise is useful here. Routine expertise - knowing which buttons to press, which promotions to accept, which games to play - will not protect a bettor targeted by a system that is continuously updated to stay ahead of behavioral adaptation. Adaptive expertise, which involves understanding the structural logic of why certain promotions appear to certain users at certain times, is the only form of competence that could plausibly transfer across the changing surface features of a system like this.

But this is precisely what the alleged system is designed to prevent. If DraftKings is targeting users who are most likely to lose, it is effectively filtering for users who have not developed adaptive expertise and directing its highest-value interventions at them. From a platform theory standpoint, this represents a perverse form of schema suppression, a system that exploits the topology of user incompetence without ever making that topology visible to the user.

What This Means for Platform Governance

Rahman (2021) describes how algorithmic systems function as invisible cages, structuring worker and user behavior through constraints that are opaque by design. The DraftKings allegation extends this framing into explicitly adversarial territory. The cage is not just invisible. According to the report, it is custom-fitted to the dimensions of each user's specific vulnerabilities.

The regulatory and governance implication is direct. Auditing AI systems for fairness or bias is insufficient if the underlying optimization target is user harm. What is needed is scrutiny of what platforms are actually optimizing for, not just how equitably they apply that optimization. A system that predicts loss probability and then promotes to high-loss-probability users is not a broken recommendation engine. It is a working one, optimized for the wrong objective.

References

Gagrain, A., Naab, T. K., & Grub, J. (2024). Algorithmic media use and algorithm literacy. New Media & Society.

Hatano, G., & Inagaki, K. (1986). Two courses of expertise. In H. Stevenson, H. Azuma, & K. Hakuta (Eds.), Child development and education in Japan (pp. 262-272). Freeman.

Kellogg, K. C., Valentine, M. A., & Christin, A. (2020). Algorithms at work: The new contested terrain of control. Academy of Management Annals, 14(1), 366-410.

Rahman, H. A. (2021). The invisible cage: Workers' reactivity to opaque algorithmic evaluations. Administrative Science Quarterly, 66(4), 945-988.