Algorithm-Augmented Work and Domain Experience: The Countervailing Forces of Ability and Aversion

Card Grid View — Paper by Allen & Choudhury (2021)

1. Inverted U-Shape Finding

Core empirical finding
Relationship between domain experience and algorithm-augmented performance is an inverted U-shape
Moderate-experience workers benefit most from algorithmic tools
Low-experience workers lack ability to use algorithm effectively
High-experience workers suffer from algorithmic aversion
Key result
Only moderate-experience workers performed significantly better WITH algorithm vs manually

2. Two Countervailing Forces

Ability
Domain experience improves ability to evaluate and apply algorithmic advice
Experts can spot errors and adapt recommendations
But ability has diminishing returns
Aversion
High-experience workers reject correct algorithmic advice
Egocentric advice discounting: trust own judgment more
Algorithmic aversion increases with experience
Interaction
Ability rises then plateaus; aversion keeps increasing
Result: inverted U-shape

3. Why Low-Experience Workers Struggle

Primary reasons
Lack domain knowledge to evaluate algorithm's output
Cannot distinguish good vs bad recommendations
May apply algorithm blindly or not at all
Underlying mechanism
Low ability to integrate algorithmic insights into work
Insufficient mental models to validate predictions
Need more training and exposure to build algorithmic literacy

4. Why High-Experience Workers Reject AI

Algorithmic Aversion
Egocentric advice discounting: overvalue own judgment
Belief that their expertise surpasses the algorithm
Distrust of "black box" recommendations
Additional mechanism
High-experience workers have more to lose by delegating
Professional identity tied to expertise
Comfortable with own methods, resistant to change

5. Study Context & Methodology

Setting
TECHCO: large tech company with NLP-based ticket resolution tool
Tool used natural language processing to standardize and suggest fixes
Data
Time-spent data on ticket resolution
Comparison: manual vs algorithm-assisted resolution
Domain experience measured by tenure
Qualitative findings
Interviews revealed egocentric advice discounting as key mechanism

6. Solutions & Managerial Recommendations

For low-experience workers
Provide training on how to use algorithmic tools
Build algorithmic literacy gradually
Expose them to successful AI-assisted outcomes
For high-experience workers
Demonstrate algorithm's value through examples
Frame AI as complement, not replacement
Incentivize adoption, address professional identity concerns
General
Match algorithm deployment to experience levels
Don't assume experts will automatically benefit

7. Implications for AI & Knowledge Work

AI cannot fully replace humans
Many knowledge-worker tasks require tacit knowledge hard to codify
Domain experience matters for effective AI use
Key takeaways
Work experience influences algorithm-augmented performance
One-size-fits-all AI deployment fails
Organizations need tailored strategies for different experience levels
Future AI capabilities
Algorithms complement but do not replace human expertise
Augmentation > Automation for knowledge work