The economics of discovery, with Ben Reinhardt

Ben Reinhardt explains why the legally-encoded three-bucket system (basic, applied, development) fails to capture how breakthrough science actually happens. Using semiconductors and transistors as examples, he demonstrates how real research involves tangled loops between fundamental physics discoveries and product development, with teams simultaneously working on quantum mechanics and practical amplifiers.

• •The basic/applied/development taxonomy is encoded into US law but doesn't reflect reality of research
• •Semiconductor development required simultaneous work on quantum mechanics theory and practical product design
• •Most useful research projects involve continuous loops between fundamental science and application
• •Process knowledge (tacit expertise not captured in papers) is critical but ignored by this framework
• •This categorization creates funding mismatches for projects that don't fit neatly into buckets

Overview of how ~$900 billion in US research funding is allocated: roughly $700B goes to development, with the remainder split 2:1 between applied and basic research. While businesses contribute ~$600B, the vast majority goes to development work, and much of what's coded as 'R&D' includes routine software feature development due to tax incentives.

• •$900B total US research spending: ~$700B development, ~$130B applied, ~$70B basic
• •Businesses contribute ~$600B but almost entirely in development category
• •Software companies can code new features as 'development expenses' for R&D tax credits
• •This creates distorted statistics about actual research vs routine product work
• •Federal government remains largest funder of pre-commercial basic research

Ben argues that university tech transfer offices should be 'burned to the ground' - only 16% are profitable, they generate single-digit billions annually despite handling all university spinouts (including RNA vaccines and Google), and they create massive friction for researchers trying to commercialize their own inventions through onerous licensing deals and bureaucratic delays.

• •Only 16% of tech transfer offices are profitable; most actively lose money
• •Total revenue across all US tech transfer offices is single-digit billions despite handling major innovations
• •Researchers must license their own inventions from universities, facing delays and unreasonable equity demands
• •University of Waterloo's model (letting inventors own inventions) works better with voluntary donations
• •Tech transfer offices consume 40% of professors' time on bureaucracy instead of research

The PI (Principal Investigator) system requires individual professors to handle fundraising, research execution, team leadership, and administration for very specific project scopes. This is equivalent to requiring startups to have only one executive doing all roles, with funding tied to specific features rather than general company operations.

• •PIs must simultaneously fundraise, execute research, lead teams, and handle admin
• •Grant system organized around specific projects with rigid scopes, not flexible organizational funding
• •Equivalent to banning CTOs and requiring one executive to do everything in a startup
• •Scientists good at research but poor at admin/management get frozen out of funding
• •No mechanism exists for funding at organizational level above individual PI labs

Grant funding specifies exactly how money must be spent, with most earmarked for grad students rather than equipment or professional staff. This creates incentives against automation and professional expertise, forcing labs to use five grad students for repetitive tasks instead of buying robots, directly impacting research productivity and quality.

• •Grant money is pre-allocated: consumables, personnel (grad students/postdocs), minimal equipment
• •Cannot hire specialized professionals - must use grad student trainees instead
• •Easy to get funding for grad students, hard to get funding for automation equipment
• •PIs explicitly choose grad students over robots because funding structure incentivizes it
• •This contributes to stagnant science productivity compared to other economic sectors

Patrick shares a case study of a COVID drug trial blocked by a 200-question intake form (only 4 questions were required) that the team couldn't modify because only one grad student knew PHP, and she wasn't available. The solution required calling in external favors rather than using proper funding mechanisms to hire professional help.

• •Drug trial had 200-question form blocking patient enrollment when only 4 questions were required
• •Team couldn't modify it because only grad student knew PHP and wasn't available
• •No funding mechanism existed to quickly hire professional developer
• •Had to rely on personal network favors to solve urgent research blocker
• •Illustrates how grad-student-only model creates critical bottlenecks in time-sensitive research

US legislation views grad students as serving dual purposes (workforce training + research production), leading to overproduction of PhDs and lower research quality. Using trainees for all research is like having interns build every software feature, with additional problems from lack of continuity when students graduate and institutional knowledge disappears.

• •Federal law treats grad students as both workforce training and research production
• •Creates overproduction of PhDs when 'more science' automatically means 'more grad students'
• •Research quality suffers from using trainees instead of experienced professionals
• •Continuity problems when grad students graduate and take institutional knowledge with them
• •As of 2013, many research labs still didn't use basic tools like Git for source control

Despite systemic problems, a growing community of 'misfits' is experimenting with new institutional models. Focused research organizations, ARPA-style programs (like UK's ARIA), and organizations like Speculative Technologies are inverting traditional constraints by funding professional teams working on useful problems without bureaucracy or artificial basic/applied distinctions.

• •Focused research organizations structure research more like startups with dedicated teams
• •ARPA-model programs emerging globally (ARIA in UK, various DARPA-inspired initiatives)
• •New orgs funding professional teams without bureaucratic overhead or basic/applied bucketing
• •Multiple simultaneous experiments testing different institutional approaches
• •US research system still world-leading despite flaws, with amazing people doing great work