Social media algorithms aren't ignoring what you like by mistake—it’s actually a math requirement for how they filter your feed.
March 26, 2026
Original Paper
Digital Twins as Behavioral Kalman Filters Understanding How AI Systems Learn About Us-And What It Means for the Next Generation
SSRN · 6346319
The Takeaway
The paper applies engineering 'Kalman filters' to AI and finds that when systems are optimized for engagement, they are mathematically guaranteed to produce biased models of the user. For the next generation, this creates a 'structural necessity' of personality distortion rather than a mere technical glitch.
From the abstract
AI systems increasingly construct probabilistic models of human preference what we term behavioral digital twins by continuously updating estimates of internal states from observable behavioral signals. We argue that Kalman filtering, the mathematical framework underlying optimal state estimation in engineering systems, provides a theoretically precise and practically illuminating lens for analyzing how these systems work, how they fail, and why their failures follow predictable patterns that ex