The Augmentation Trap: AI Productivity and the Cost of Cognitive Offloading

Abstract

Experimental evidence confirms that AI tools raise worker productivity, but also that sustained use can erode the expertise on which those gains depend. We develop a dynamic model in which a decision-maker chooses AI usage intensity for a worker over time, trading immediate productivity against the erosion of worker skill. We decompose the tool's productivity effect into two channels, one independent of worker expertise and one that scales with it. The model produces three main results. First, even a decision-maker who fully anticipates skill erosion rationally adopts AI when front-loaded productivity gains outweigh long-run skill costs, producing steady-state loss: the worker ends up less productive than before adoption. Second, when managers are short-termist or worker skill has external value, the decision-maker's optimal policy turns steady-state loss into the augmentation trap, leaving the worker worse off than if AI had never been adopted. Third, when AI productivity depends less on worker expertise, workers can permanently diverge in skill: experienced workers realize their full potential while less experienced workers deskill to zero. Small differences in managerial incentives can determine which path a worker takes. The productivity decomposition classifies deployments into five regimes that separate beneficial adoption from harmful adoption and identifies which deployments are vulnerable to the trap.

Figures (8)

• Figure 1: Total discounted value vs. steady-state value ($\beta=1$, $\gamma=1$, $\kappa=0.3$, $\delta=0.1$, $\bar{S}=1$). The green curve plots the change in total discounted value $V(\bar{S})-\bar{S}/\delta$, which is positive throughout the adoption region, meaning adoption is always privately rational. The blue curve plots the change in steady-state value $V(\hat{S})-\bar{S}/\delta$, which dips below zero in the steady-state loss region ($\alpha_{0}<\alpha\le\alpha_{1}$).
• Figure 2: The discount-rate gap ($\alpha=0.5$, $\beta=1.3$, $\gamma=1$, $\kappa=0.3$, $\bar{S}=1$). The change in total discounted value (green) is positive for all discount rates above the adoption threshold, meaning adoption is always privately rational. The change in steady-state value (blue) is negative throughout, meaning the long-run position is worse than no AI. The more impatient the decision-maker is, the worse the loss becomes. Cf. Figure \ref{['fig:bzero']}, which plots the same decomposition over $\alpha$ at fixed $\delta$.
• Figure 3: Optimal AI usage as a function of skill for the three complementarity regimes. When $\beta>1$, higher-skill workers use AI more; when $\beta<1$, lower-skill workers use AI more; when $\beta=1$, usage is flat. Dots mark steady states. Parameters: $\alpha=1.2$, $\gamma=1.0$, $\kappa=0.3$, $\delta=0.1$, $\bar{S}=1$. Curves shown for $\beta=1.5$ (complementary), $\beta=1.0$ (neutral), and $\beta=0.5$ (leveling).
• Figure 4: Five regions of the $(\alpha,\beta)$ parameter space ($\gamma=0.15$, $\kappa=0.3$, $\delta=0.1$, $\bar{S}=1$). Solid line $C_0$: adoption onset; dotted line $C_1$: automation onset; dashed line $B$: long-run break-even; dash-dot line $D$: $\alpha - \gamma = \bar{S}$, above which full automation yields higher output than the no-AI baseline. Steady-state loss (Region II, pink) is the wedge between $C_0$ and $B$ where adoption is rational but the steady state is worse than no AI. Figure \ref{['fig:bzero']} is the $\beta=1$ cross-section.
• Figure 5: Persistent skill stratification under $\beta<1$ with $(1-\beta+2\kappa a\bar{S})\bar{S} > 2\gamma$. Parameters: $\beta=0.3$, $\bar{S} = 5$, $\kappa = 0.1$, $\gamma = 1$, $\alpha=2.5$, $\delta=0.1$. Workers with initial skill below the unstable equilibrium $S_{\mathrm{eq}} \approx 1.67$ adopt AI heavily and converge to $\hat{S} = 0$. Workers above $S_{\mathrm{eq}}$ use little or no AI and converge to $\bar{S}$. The same technology widens the skill distribution permanently.

Theorems & Definitions (27)

• Lemma 1: Quadratic value and linear usage policy
• Proposition 1: Steady-state loss in the skill-neutral case
• Proposition 2: Steady-state loss under full automation
• Proposition 3: Usage is increasing in skill when $\beta>1$
• Proposition 4: Usage is decreasing in skill when $\beta<1$
• Proposition 5: Overuse under short-termism
• proof : Proof sketch
• Proposition 6: Worker skill externality reduces AI usage
• Definition 1: Augmentation trap
• Proposition 7: Conditions for the augmentation trap