My simulations showed that a uniform‑memory AI following a high‑performing human yields a +27 % increase in joint payoff compared to an AI‑first baseline, while introducing stochastic escape improves low‑human runs by +15 %.
I led a research project titled “When Should AI Follow? Task Structure and Joint Adaptation by Human and AI Agents” at the Indian School of Business (work periods Apr 2025 and Feb 2025). The central question was how organizations should divide and sequence decision tasks between human and artificial agents.
The model uses two memory regimes: a recency‑weighted regime, motivated by behavioral evidence on human adaptation, privileges recent outcomes; a uniform‑memory regime, motivated by the scale‑free consistency of algorithmic updating, weights a window of past outcomes equally.
I concluded that the dominant “AI‑first” deployment is not optimal.
Best joint performance is achieved when scale‑free (uniform‑memory) adaptation follows a high‑performing human.
When a human performs poorly, a broad stochastic search (memoryless stochastic adaptation) rescues the sequence.
- Threshold dynamics create absorbing high‑ and low‑payoff regimes, so adaptation compounds whatever it inherits.
The findings suggest a contingency logic for organizations: design task structures so that uniform‑memory AI agents follow strong human performers, and employ stochastic search when human performance is weak. This cautions against universal AI‑first prescriptions.
I have also listed other projects on my experience page, including work on home‑layout design, a Spotify recommendation engine, asteroid analysis, and a graduate admissions prediction model.
- Uniform memory amplifies inherited trajectories, for good and for ill.
- Recency weighting corrects locally but is volatile at scale.
- Memoryless stochastic adaptation, while inferior on average, serves as an escape mechanism when inherited trajectories are poor.