Today's Work, Tomorrow's Workforce: Incentivizing Learning in AI-Assisted Work

Moritz Diener; Daniel Hendriks; Philipp Spitzer; Gerhard Satzger; Joshua Holstein; Wei Sun

cs.HCNov 4, 2026

Today's Work, Tomorrow's Workforce: Incentivizing Learning in AI-Assisted Work

Authors: Moritz Diener, Daniel Hendriks, Philipp Spitzer, Gerhard Satzger, Joshua Holstein, Wei Sun

Organizations: Karlsruhe Institute of Technology (KIT) Karlsruhe, Germany · IBM Research USA

Abstract

At work, people develop skills through experience: novices acquire skills by completing tasks, while experts retain theirs through continued practice. AI assistance changes this process by offering workers a choice between solving tasks independently and accepting the AI's answers. While relying on AI saves immediate effort, it can also displace the practice needed to acquire and retain skills, including those required to oversee the AI. We extend an existing accept-or-solve model to a sequence of tasks in which workers' skills evolve through independent practice and reliance on AI. We compare myopic workers, who maximize immediate utility, with strategic workers, who also value their future skills. We show that novices can become trapped at low skill levels, while experts may lose skill, even as AI-assisted outcome accuracy remains high. Strategic workers, by contrast, may choose to solve independently even when AI is more accurate, recognizing the future value of acquiring and retaining their skills. We then study incentives that reward correct independent work and characterize the payments needed to encourage skill acquisition and retention. We show that independent work can become self-sustaining once workers acquire sufficient skill, eliminating the need for continued incentives. Counterintuitively, supporting a higher skill target can sometimes reduce long-run incentive costs, either by enabling self-sustaining independent work or by lowering the payments needed to maintain proficiency. Our findings thereby shift the focus from whether workers should rely on AI on individual tasks to how organizations can sustain human expertise over time, and which skills are worth investing in.

Figures & tables

Choice	Correct with prob.	correct	incorrect	Expected utility	Skill for the next task
		Utility if final answer is
Accept AI	$P_{AI}$	$\gamma$	$0$	$u_{A}=\gamma P_{AI}$	falls to $p^{-}$ with probability $\phi_{d}$
Solve independently	$p$	$\gamma-\lambda$	$-\lambda$	$u_{S}(p)=\gamma p-\lambda$	rises to $p^{+}$ with probability $\phi_{u}$

Table 1: One task in the dynamic accept-or-solve model. Utilities by choice and correctness of the final answer follow Bansal et al. [8, Table 1] ; the last column is this paper’s extension. Skill otherwise stays at

p

, with

p^{-}=p

at the lowest and

p^{+}=p

at the highest level.

Figure 1: Rational choices for a strategic worker at AI accuracy

P_{AI}

and expected horizon

H

. Blue shading indicates solving independently, and red shading indicates accepting the AI’s answer. The braces show the appropriate-reliance benchmark, which favors solving when

p_{0}\geq P_{AI}

. Hatching marks where the strategic choice differs from that benchmark. Points on the dashed line show the running examples at

H=100

. (A) The novice (

p_{0}=.50

) accepts the competent AI at

H=100

despite the opportunity to build skill. (B) The expert (

p_{0}=.80

) accepts the near-expert AI at every horizon. Appendix A.7 derives the strategic boundaries.

Figure 2: Expected worker skill over completed tasks for a myopic worker with no incentive (grey), the effort incentive of Holstein et al. [34] (red), or the learning incentive in Eq. ( 11 ) (blue). The dotted horizontal line marks AI accuracy. (A) With competent AI, only the learning incentive builds novice skill. (B) With near-expert AI, both incentives retain expert skill at

.80

, so their curves overlap. (C) With above-expert AI, only the learning incentive retains expert skill. In panels A and C, the effort and no-incentive curves overlap.

Appendix figures & tables9 assets

Supplementary material from the paper’s appendix.

Appendix

Symbol	Meaning	Running example
$\mathcal{P}=\{a_{1}<\dots<a_{N}\}$	skill levels (unaided accuracy)	$\{.40,.50,.60,.70,.80\}$ , $N=5$
$p,\ p^{+},\ p^{-}$	current level and its upper and lower neighbor	e.g. $p=.60$ , $p^{+}=.70$ , $p^{-}=.50$
$P_{AI}$	accuracy of the AI	competent $.62$ , near-expert $.77$ , above-expert $.82$
$\gamma$	reward for a correct final answer	$1$
$\lambda$	effort cost of solving independently	$.05$
$\phi_{u}$	probability of a learning step after solving	$.02$

Appendix

Table 2: Notation and running-example values.

Main text	Content	Appendix
Eqs. ( 1 ), ( 2 )	skill levels as a Markov chain, expected next-task accuracy	A.1
Table 1 , Eqs. ( 3 ), ( 4 )	utility of one task, immediate payoff gap	A.2
Eq. ( 5 )	myopic decision rule	A.3
Eq. ( 6 )	lifetime value, meaning of $\delta$	A.4
Eqs. ( 7 ), ( 8 )	Bellman equation, strategic decision rule	A.5
—	how the model is solved	A.6

Appendix

Table 3: Where each equation of the main text is derived.

Panel of Figure 3 — Figure 3: Skill dynamics as a Markov chain. (A) The controlled chain. If the worker solves, skill rises by one level with probability $\phi_{u}$ ; if the worker accepts, it falls by one level with probability $\phi_{d}$ ; otherwise it stays. At the boundaries, a step that would leave the grid is replaced by staying, so the corresponding self-loop has probability one. (B–D) Chains under fixed policies on the running-example grid. Fill marks the action at each level; self-loops are omitted. (B) A strategic worker with the competent AI accepts at $.40$ and $.50$ and solves above. No transition connects $.50$ and $.60$ , so a worker starting at $.50$ can only drift to $.40$ . (C) With the near-expert AI, every level accepts, and skill drifts to $.40$ . (D) The target incentive for $p^{\star}=.70$ makes the worker solve up to the target, while the worker accepts at $.80$ . Skill eventually alternates between $.70$ and $.80$ ; the share of tasks at $.70$ sets the long-run cost of the target (Appendix A.9 ). Parameters as in Section 3.2 .

Figure 4: Strategic independent work pays off despite an initial cost. Starting at

p_{0}=.60

with the competent AI (

P_{AI}=.62

), the curve shows expected cumulative undiscounted utility under the strategic policy minus that under the myopic policy. The largest shortfall is

1.30\gamma

after 38 tasks, and the gain first becomes nonnegative after 80 completed tasks. The discounted lifetime gain is

4.06\gamma

. Parameters as in Section 3.2 , with

\delta=.99

Parameter	Values	Count
Effort cost $\lambda/\gamma$	$.01$ , $.025$ , $\mathbf{.05}$ , $.10$ , $.20$	5
Learning probability $\phi_{u}$ (per step of $.10$ )	$.005$ , $.01$ , $\mathbf{.02}$ , $.05$ , $.10$ , $.25$	6
Decay probability $\phi_{d}$ (per step of $.10$ )	$.0025$ , $.005$ , $\mathbf{.01}$ , $.02$ , $.05$ , $.10$	6
Horizon $H=1/(1-\delta)$	$2$ , $5$ , $10$ , $20$ , $50$ , $\mathbf{100}$ , $200$ , $500$ , $1000$	9
AI accuracy $P_{AI}$	$.30$ , $.31$ , …, $.96$	67
Skill grid	5 levels on $[.40,.80]$ ; 3, 9, or 17 equally spaced levels on $[.40,.80]$ ; 7 levels on $[.30,.90]$ ; 8 levels on $[.60,.95]$ ; diminishing steps $\{.40,.50,.59,.67,.74,.80\}$ ; increasing steps $\{.40,.44,.50,.58,.68,.80\}$	8

Appendix

Table 4: Parameter values in the full factorial design. Running-example values in bold.

	Result	Checked	Failed
1	Strategic policy is a threshold in skill (solve at $a_{n}\Rightarrow$ solve at all $a_{m}>a_{n}$ )	868,320	0
2	Strategic policy equals the closed form, Eq. ( 17 ) a	868,320	0
3	Lifetime values equal the closed form, Eq. ( 18 )	868,320	0
4	$V$ is nondecreasing in skill; the strategic worker solves wherever the myopic one does	868,320	0
5	Low-skill trap: if the worker accepts at $a_{1}$ and $a_{2}$ , then $V(a_{1})=V(a_{2})$	520,524	0
6	Expert deskilling: all levels accept and $V$ is constant iff $a_{N}<P_{AI}+\lambda/\gamma$	868,320	0

Appendix

Table 5: Qualitative results checked in the full factorial design. Each row counts the configurations in which the condition applies and those in which it fails. Rows 1–7 concern workers without incentives, rows 8–14 the incentives of Section 5 . Rows 11 and 13 count every target in every configuration; rows 8, 12, and 14 do not depend on the horizon; row 7 counts combinations of the other parameters, each compared across the six decay probabilities.

Figure 5: The low-skill trap and expert deskilling across parameters. (A) Strategic choice of a novice at

p_{0}=.50

on the running-example grid with

\lambda/\gamma=.05

, as a function of the expected number of learning steps over the horizon,

\kappa=\phi_{u}(H-1)

, and the AI accuracy

P_{AI}

. Below the curve, the novice solves. Between the curve and

P_{AI}=a_{N}-\lambda/\gamma

, solving would pay at the top level, but the novice accepts (low-skill trap). Above that line, everyone accepts (expert deskilling). The curve is the closed form of Eq. ( 17 ); markers are boundaries computed with the exact solver for three learning probabilities and several horizons, which fall on the same curve. The running example has

\kappa=.02\times 99=1.98

. Decay

\phi_{d}

does not affect the regions. (B) Width between the theoretical trap boundaries,

a_{N}-\hat{a}(.50)

, for six skill grids. The horizontal axis uses the baseline learning probability

\phi_{u}^{(.10)}

, with each grid’s probability scaled as described in Appendix B.1 . The plotted width is measured before restricting the trap interval to feasible AI accuracies

P_{AI}\in[0,1]

. If both boundaries lie within

[0,1]

, this equals the feasible trap width and is independent of

\lambda/\gamma

. If a boundary lies outside

[0,1]

, only the intersection with

[0,1]

is feasible; its width can be smaller and depend on

\lambda/\gamma

. Neither width depends on the decay probability

\phi_{d}

Figure 6: Which retention target costs less in the long run, for a worker at

a_{N}=.80

on the running-example grid. The higher target

a_{N}

is cheaper below the boundary

\phi_{u}/\phi_{d}=\gamma s/\Delta u(a_{N})

, the lower target

a_{N-1}

above it. The solid boundary uses

\lambda/\gamma=.05

, the others

.025

and

.10

. Dots are exact long-run payments for

\lambda/\gamma=.05

, colored by the cheaper target. The two marked points are the running example with the near-expert AI (

P_{AI}=.77

) and the above-expert AI (

P_{AI}=.82

), both with

\phi_{u}/\phi_{d}=2

Variation	Novice	Expert	Trap width	Anna trapped	Sam deskills	Target	Cost	$a_{N}$	$a_{N-1}$
	Boundaries in $P_{AI}$					Bridging target		Retention
Running example	$.590$	$.75$	$.160$	yes	yes	$.60$	$25.50$	$.020$	$.040$
$\lambda/\gamma=.025$	$.615$	$.775$	$.160$	yes	no	$.60$	$21.75$	—
$\lambda/\gamma=.10$	$.540$	$.70$	$.160$	yes	yes	$.70$	$34.00$	$.070$	$.057$
$\phi_{u}=.01$	$.537$	$.75$	$.213$	yes	yes	$.60$	$51.00$	$.020$	$.060$
$\phi_{u}=.05$	$.660$	$.75$	$.090$	no	yes	$.60$	$10.20$	$.020$	$.020$

Appendix

Table 6: Running-example outcomes when one parameter is varied at a time. Novice boundary : largest

P_{AI}

at which the strategic worker solves at

.50

. Expert boundary :

a_{N}-\lambda/\gamma

. Trap width : their difference. Anna trapped : a strategic Anna at

.50

accepts with the competent AI. Sam deskills : a strategic Sam at

.80

accepts with the near-expert AI. Bridging target : cheapest target that ends payments for a novice starting at

a_{1}

with the competent AI, and its expected total cost. Retention : long-run payment per task to retain

a_{N}

a_{N-1}

with the near-expert AI; — if Sam solves without an incentive. Strategic workers use

H=100

unless stated otherwise. On the grid with 9 levels,

\phi_{u}=.04

and

\phi_{d}=.02

. The grids with diminishing and increasing steps keep the running-example values

\phi_{u}=.02

and

\phi_{d}=.01

AI and Labor

AI and labor studies how artificial intelligence changes jobs, wages, skills, hiring, and the organization and quality of work.

See all 99 papers

Path Dependence under Adaptive AI Delegation
Mar 3, 2026Lingxiao Huang, Nisheeth K. VishnoiHuman-AI Collaboration
Nanjing University · Yale University
Managing the Human Fallback: Skill Investment Under Improving AI and Worker Mobility
Jun 27, 2026Simrita Singh, Naireet Ghosh, Tinglong DaiHuman-AI Collaboration AI and Labor
Leavey School of Business, Santa Clara University, Santa Clara, California, 95053 · Naveen Jindal School of Management, University of Texas at Dallas, Richardson, Texas, 75080 · Carey Business School, Johns Hopkins University, Baltimore, Maryland 21202
How AI Assistance Affects Human Skill Development: A Study of Learning with Logic Puzzles
Aug 24, 2026Shang Wu, Catarina G Belem, Shuyuan Fu +2Human-AI Collaboration
University of California, Irvine USA

Mar 3, 2026J/K move · Enter open · S save

Path Dependence under Adaptive AI Delegation

Lingxiao Huang, Nisheeth K. Vishnoi

Nanjing University · Yale University

Human-AI Collaboration

Open paper AskPDF arXiv

Repeated AI assistance can improve immediate task performance while reducing the skill available for future independent work. We develop a mathematical framework for this long-run tradeoff. The model tracks two state variables: a latent human skill level governing expected independent performance, and a delegation level representing the learner's evolving tendency to rely on AI. Skill changes through error-driven learning under practice and decay under delegation; delegation responds to observed performance, increasing when AI-assisted work appears to outperform independent work. We analyze the resulting dynamics and contrast them with fixed delegation. With fixed delegation, skill follows a one-dimensional learning-decay process with a single stable equilibrium. With adaptive delegation, the coupled system has two attracting equilibria separated by the stable manifold of an interior saddle. The existence and geometry of this separatrix require a global phase-plane analysis of the coupled dynamics. The system is path-dependent: small differences in initial skill or reliance can lead to different long-run outcomes. We use this characterization to show that AI assistance can improve short-run performance while producing worse long-run performance than a no-AI baseline. Increasing AI capability can enlarge the basin of attraction of the low-skill equilibrium, making delegation appear beneficial for longer while increasing the risk of eventual skill loss. The qualitative picture is observed to persist across alternative specifications. Together, these results show that the risk is not AI assistance itself, but the coupling between performance-driven reliance and use-dependent skill change.

Abstract

AI and Labor

Explore similar work

Path Dependence under Adaptive AI Delegation

Managing the Human Fallback: Skill Investment Under Improving AI and Worker Mobility

How AI Assistance Affects Human Skill Development: A Study of Learning with Logic Puzzles