My notes via GPT. How innovation happens
1) Incentives shape problem-choice. When funding, governance, or comp compacts tolerate early failure and reward long-horizon success, researchers and founders take bolder bets and produce more novel work. That’s now shown both theoretically and empirically (e.g., Manso’s model; HHMI vs NIH natural experiment by Azoulay et al.). Massachusetts Institute of Technology+1
2) Team size and novelty. Small teams and lone inventors are disproportionately “disruptive” (they shift fields), while large teams excel at refinement and scaling. Also: the overall rate of disruptive work in papers and patents has fallen in recent decades—whatever your theory, that’s the fact pattern you’re designing against. Nature+1
3) Breakthroughs = recombination + weird edges. High-impact ideas often come from atypical combinations of knowledge and from recombining components in unusual ways. That’s been shown in patents and papers (Fleming’s “recombinant uncertainty”; Uzzi et al. on atypical combinations). Absorptive capacity—the ability to recognize and use external knowledge—is the enabler, and it’s built by prior related work and internal R&D. ScienceDirect+2MIT Press+2
4) Organization matters. Incumbents fail not just at radical tech but at architectural changes (same components, new linkages). Exploration vs. exploitation is a real trade-off; winners cultivate “dynamic capabilities” and protect complementary assets (manufacturing, distribution, data, standards), or they lose the rents to faster imitators. NTNU IoT+2sjbae.pbworks.com+2
5) Place still matters. Knowledge spillovers decay with distance. Dense clusters raise the odds of collisions, skilled labor pooling, supplier depth, and faster diffusion. Silicon Valley vs Route 128 is the classic contrast; Porter’s cluster logic generalizes beyond tech. ResearchGate+2unica.it+2
6) Social returns to R&D exceed private returns. Society under-invests in research without policy help. That’s the core growth-economics result (Schumpeterian models; direct estimates of R&D social returns). Public/mission-oriented “ARPA-style” programs are one proven correction when designed with autonomy and empowered program directors. individual.utoronto.ca+2Stanford University+2
7) Diffusion ≠ invention. Adoption runs on networks, norms, and incentives (Rogers). Weak ties and boundary spanners often matter more than your best friends for spreading new ideas. teddykw2.files.wordpress.com+1
8) Scaling drives cost-curves. Learning/experience curves are real but not magic: costs fall with cumulative deployment and with directed R&D. Treat Wright’s law and “generalized Moore’s laws” as forecasting tools with error bars, not as destiny. Of (im)possible interest+2François Lafond+2
Where the evidence is strong vs. shaky
Robust:
Exploration-friendly incentives increase novelty (HHMI vs NIH; Manso). NBER+1
Local spillovers/clusters raise innovation outputs. unica.it
Small vs large teams specialize in disruption vs development. Nature
Social returns to R&D >> private returns; policy leverage is large. Stanford University
Absorptive capacity (built via prior related R&D) predicts who benefits from external knowledge. josephmahoney.web.illinois.edu
Nuanced/contested:
“Decline in disruption.” The fact is solid; the causes (metric artifacts, topic maturity, incentives, publication norms) are debated. Nature
Patents/IP and innovation. Appropriability matters, but effects vary by field; patent thickets can hinder cumulative innovation (mixed empirical results; read Levin et al. on appropriability). Brookings
Learning-by-doing vs R&D. Both contribute to cost decline; weights differ by sector (energy is unusually R&D-sensitive). wikis.mit.edu
References (not checked):
Schumpeter (1942) – Creative destruction frames the whole debate.
Nelson & Winter (1982), An Evolutionary Theory of Economic Change – Firms learn; routines and selection drive industry dynamics. inctpped.ie.ufrj.br
Dosi (1982) – Technological paradigms/trajectories explain path-dependence. sjbae.pbworks.com
Pavitt (1984) – Sectoral patterns: industries innovate differently. wuecampus.uni-wuerzburg.de
Abernathy & Clark (1985) – “Transilience” map for how innovations disrupt markets/capabilities. sjbae.pbworks.com
Teece (1986), “Profiting from Technological Innovation” – Complementary assets & appropriability explain who captures value. politicipublice.ro
Cohen & Levinthal (1990) – Absorptive capacity: prior related knowledge as the gate to external ideas. josephmahoney.web.illinois.edu
Henderson & Clark (1990) – Architectural innovation blinds incumbents. NTNU IoT
March (1991) – Exploration vs exploitation; why organizations under-explore. sjbae.pbworks.com
Rogers (2003), Diffusion of Innovations (5e) – Adoption curves, opinion leaders, network channels. teddykw2.files.wordpress.com
Granovetter (1973) – Weak ties spread ideas across communities. CMU School of Computer Science
Porter (1998) – Clusters raise productivity, innovation, and new firm formation. biblioteca.fundacionicbc.edu.ar
Saxenian (1994) – Why Silicon Valley beat Route 128: open networks > closed hierarchies. G-City
Bresnahan & Trajtenberg (1995) – General-purpose technologies as growth engines. Chicago Journals
Mansfield (1995) – How much industry innovation actually builds on academia (surveys & lags). EconPapers
Kortum & Lerner (2000) – Venture capital causally boosts patenting in VC-intensive sectors. Massachusetts Institute of Technology
Manso (2011) – Optimal innovation incentives tolerate early failure, reward long-term success. Wiley Online Library
Azoulay, Graff Zivin & Manso (2011) – HHMI’s tolerant funding produces riskier, more novel science than NIH. faculty.haas.berkeley.edu
Wu, Wang & Evans (2019) – Big teams develop; small teams disrupt. Nature
Park, Leahey & Funk (2023) – Measured disruptiveness in science/patents has declined since the ’80s. Nature
Bloom, Jones, Van Reenen & Webb (2020) – “Ideas getting harder to find”: rising research inputs per unit progress. Stanford University
Fleming (2001) – Recombination logic: novelty comes with higher variance—most fail, rare hits pay. ScienceDirect
Uzzi et al. (2013) – Atypical knowledge combinations correlate with high impact. MIT Press
Farmer & Lafond (2016) – Many tech costs fall roughly exponentially; you can forecast with uncertainty bands. François Lafond
Wright (1936) – The original learning curve (experience → lower cost). Of (im)possible interest
Azoulay et al. (2019) – The ARPA/DARPA funding model: autonomy + empowered PMs + active program management. NBER
(For broad overviews, the Oxford Handbook of Innovation and Handbook of the Economics of Innovation are solid gateways. rudyct.com+1)