Research Q & A

Q1: Why “Education Vintage” instead of simply “age”?

A: Age captures general experience and generational identity, but Education Vintage specifically measures the technological paradigm under which a director’s cognitive framework was formed. Two directors of the same age could have vastly different vintages if one pursued a late-career STEM degree. Vintage isolates the knowledge-relevance dimension from the broader age construct.

Q2: Why is ChatGPT the right exogenous shock?

A: ChatGPT’s release (November 30, 2022) satisfies three conditions for a valid DiD shock: (1) it was discrete and identifiable, (2) it was largely unanticipated in timing and magnitude, and (3) it created a clear demarcation between pre-GenAI and post-GenAI strategic environments. Unlike gradual AI adoption, ChatGPT forced an immediate strategic response from boards.

Q3: How do you distinguish GenAI adoption from general AI adoption?

A: We use a validated keyword dictionary that separates GenAI-specific terms (“Large Language Model,” “Transformer,” “GPT,” “Foundation Model”) from generic technology terms (“Big Data,” “Cloud,” “Machine Learning”). This ensures we capture the specific competence-destroying discontinuity of GenAI rather than incremental AI investment.

Q4: What about reverse causality – don’t innovative firms just hire younger directors?

A: This is our primary endogeneity concern. We address it through: (1) firm fixed effects controlling for time-invariant firm characteristics, (2) instrumental variable strategy using director birth year, and (3) the DiD design itself, which measures within-firm changes around an exogenous shock rather than cross-sectional correlations.

Q5: Can the Reskilling Index truly capture lifelong learning?

A: The Reskilling Index is an imperfect proxy. It captures formal certifications and degrees from BoardEx but misses informal learning (online courses, self-study, conference attendance). We acknowledge this limitation and suggest future research incorporating richer reskilling data.

Q6: Why pre-2012 as the vintage cutoff for “modern AI”?

A: 2012 marks the AlexNet breakthrough in deep learning (ImageNet competition), which fundamentally shifted the AI paradigm from rule-based/symbolic approaches to neural network/deep learning methods. This represents a natural technological discontinuity point in AI education curricula. We conduct robustness checks with alternative cutoffs (2010, 2015).

Q7: How does this differ from studies on board digital literacy?

A: Existing studies measure digital literacy as a binary (has/doesn’t have tech experience). Our contribution is adding the temporal dimension – not just whether directors have technical knowledge, but when that knowledge was acquired and whether it remains relevant to the current technological regime.