1 cumulative citations
View corpus contextSignals that deep learning had greater application potential prompted firms to sharply ramp cumulative investments in prior research; rather than hoarding know‑how, firms disclosed development paths to recruit external innovators and learn from their contributions.
How to grow new applications out of old research? Evidence from firm cumulative investments in deep learning
openalex
quasi_experimental
medium evidence
7/10 relevance
Summary only
summary available; pdf_status=paywall
DOI
Source
PDF
Firms substantially increased cumulative investment in their prior deep-learning research after receiving signals of elevated application potential, and they publicly disclosed development trajectories to attract and learn from external innovators, facilitating co-evolution of applications.
Citation observations
Cumulative provider counts captured on specific dates; providers are never combined.
1 cumulative citations
View corpus contextAbstract Research Summary Firm technological research has the potential to spawn multiple applications. Despite recognizing such potential, past literature disagrees on the process through which firms discover and grow new applications out of their past technological research. I examine this question in the context of deep learning, taking a question‐driven approach. Difference‐in‐difference analysis suggests that firms radically increased cumulative investments in past deep learning research upon signals indicating elevated application potential of deep learning. Furthermore, rather than investing in proprietary efforts, firms disclosed their cumulative development trajectories to engage external innovation efforts from which they learn and build. Grounded in these findings, I propose that the discovery and growth of new applications of past research entails unfolding innovation interdependence which motivates firms to co‐evolve with external innovators. Managerial Summary Firm technological research has the potential to spawn multiple applications. This article examines how firms cumulatively invest in their past technological research to grow new applications in the context of deep learning. Employing a difference‐in‐difference approach, analysis suggests that firms radically increased cumulative investments in deep learning after a shock elevating the application potential of their past deep‐learning research. Furthermore, firms publicly disclosed their cumulative development trajectories to attract innovation efforts from application sectors while actively learning from the attracted efforts to innovate further. These findings suggest that firms engaged, leveraged and co‐evolved with external innovation efforts to discover and grow new applications of their past research.
Summary
Main Finding
Firms responded to a signal that raised the application potential of past deep‑learning research by sharply increasing cumulative investments in that past research. Rather than pursuing those investments solely behind closed doors, firms publicly disclosed their cumulative development trajectories to attract external innovators, then learned from and co‑evolved with those external efforts. The paper frames this behavior as driven by "unfolding innovation interdependence"—the idea that discovering and growing new applications from past research requires interaction with outside innovators.
Key Points
- A shock that signaled elevated application potential of deep learning triggered large increases in firms' cumulative investment in their prior deep‑learning research.
- Firms favored disclosure of their development trajectories (rather than pure proprietary secrecy) to mobilize external innovation activity.
- External innovators' activity generated knowledge that firms absorbed, enabling further internal innovation—i.e., firms and external innovators co‑evolved.
- The mechanism proposed is "unfolding innovation interdependence": application discovery/growth depends on interactions across organizational boundaries, so firms strategically engage external innovators.
- Results highlight a strategic complementarity between internal past research and external innovation efforts in the growth of new applications.
Data & Methods
- Context: deep‑learning research and its applications (paper takes a question‑driven approach within this technological domain).
- Identification: difference‑in‑difference (DID) analysis comparing firms differentially affected by a shock that raised the application potential of deep learning.
- Outcomes studied (as described in the abstract): cumulative investments in past deep‑learning research, firms' public disclosure behavior (cumulative development trajectories), and evidence of attracted external innovation and subsequent learning.
- Empirical strategy leverages temporal variation around the shock to infer causal effects of elevated application potential on firm investment and disclosure behavior.
- (Paper likely contains robustness checks and operational details of measures and sample, not specified in the abstract.)
Implications for AI Economics
- Open strategy as an investment multiplier: strategic disclosure can be an efficient mechanism to mobilize external R&D, increasing returns to prior research investments.
- Externalities and complementarities: deep‑learning research generates positive externalities that firms can internalize indirectly via engagement with outside innovators, implying coordination failures or gains that markets/policy can influence.
- Rethinking firm boundaries: when applications are discovered through external interactions, the optimal boundary of firm R&D becomes more porous—models of firm behavior should account for co‑evolution with ecosystems.
- Policy relevance: intellectual‑property rules, standards, and incentives for open sharing affect how quickly and widely applications develop from foundational AI research.
- Measurement and modeling: empirical and theoretical models of AI-driven growth should incorporate dynamic, networked innovation interdependence (feedback between prior research, disclosure, external contributors, and cumulative application development).
Assessment
Paper Typequasi_experimental
Evidence Strengthmedium — The paper exploits a quasi-experimental DD design to estimate a causal effect of an external signal on firm investment behavior, which is stronger than simple correlations; however, without randomization the estimate depends on parallel trends and the plausibility that the signal is exogenous to other firm-level shocks, and the abstract does not report robustness checks, falsification tests, or alternative identification strategies.
Methods Rigormedium — Use of difference-in-differences and tracing of disclosure and external collaboration dynamics is appropriate for the question and suggests careful panel analysis, but the abstract lacks detail on control variables, robustness, instrumenting, treatment definition, sample selection, and checks for anticipatory effects or differential trends.
SampleFirm-level panel of organizations with prior deep-learning research activity, measuring cumulative investments in deep learning, public disclosures of development trajectories, and engagement from external innovators; treatment defined by an event or signal that elevated the perceived application potential of deep learning (time frame, industry coverage, and geographic scope not specified in the abstract).
Themesinnovation adoption org_design
IdentificationDifference-in-differences comparing firms that received an exogenous signal of elevated application potential for their past deep-learning research to otherwise similar firms over time; identification relies on pre-trend parity and the timing of a shock that plausibly changes expected application value.
GeneralizabilityFocused on deep learning; results may not generalize to other technologies or earlier stages of tech, Likely limited to firms already active in deep-learning research (selection bias), Unclear geographic/sector coverage—may not apply to small firms, non-tech industries, or different institutional environments, Dependent on the nature and observability of the 'signal'—other shocks may behave differently, Findings about public disclosure and external co-evolution may vary with IP regimes and industry norms
Claims (4)
| Claim | Direction | Outcome | Confidence & Evidence | Details |
|---|---|---|---|---|
| Difference-in-difference analysis suggests that firms radically increased cumulative investments in past deep learning research upon signals indicating elevated application potential of deep learning. Research Productivity | positive | cumulative investments in past deep learning research |
Reading fidelity
high
Study strength
medium
|
not reported
|
| Rather than investing in proprietary efforts, firms disclosed their cumulative development trajectories to engage external innovation efforts from which they learn and build. Innovation Output | positive | public disclosure of development trajectories and attraction of external innovation efforts |
Reading fidelity
high
Study strength
medium
|
not reported
|
| Firms publicly disclosed their cumulative development trajectories to attract innovation efforts from application sectors while actively learning from the attracted efforts to innovate further. Innovation Output | positive | learning from external innovation efforts and subsequent firm innovation activity |
Reading fidelity
high
Study strength
medium
|
not reported
|
| The discovery and growth of new applications of past research entails unfolding innovation interdependence which motivates firms to co-evolve with external innovators. Innovation Output | positive | co-evolution with external innovators / innovation interdependence |
Reading fidelity
high
Study strength
speculative
|
not reported
|