You have 4 summaries left

Smart Friends

Turning Discoveries Into Companies with Sam Arbesman of Lux Capital

Tue Apr 25 2023
ScienceTechnologyInnovationAIStartupsKnowledgeScientific ResearchIntellectual PropertyEducationLeveragePhysicsComplexity ScienceGaming


This episode explores the intersection of science, technology, and innovation. It delves into the role of AI in startups, the evolution of knowledge, the process of scientific research, the importance of intellectual property, the future of education, and the power of leverage. The episode also touches on topics like alternative research organizations, directed vs undirected research, and the connection between physics, complexity science, and gaming.


AI has both underperformed and drastically overperformed in a short amount of time.

The field of AI has experienced both disappointments and breakthroughs within a relatively short period.

Scientific knowledge undergoes a process of refinement rather than complete reversal.

Scientific facts are constantly being refined and updated as new discoveries are made.

Universities should find a balance between IP ownership and scientific progress.

While universities aim to capture value from their intellectual property, it is important to ensure that this does not hinder scientific progress.

Education will undergo significant changes with the integration of AI technology.

AI has the potential to revolutionize education by enabling personalized and interactive learning experiences.

The intersection of physics, complexity science, and gaming is a fascinating area of exploration.

The combination of these fields offers unique insights into the nature of systems and the possibilities of artificial life.


  1. AI and Innovation
  2. Knowledge and Science
  3. Scientific Research and AI
  4. The Process of Science
  5. Innovation and Research Organizations
  6. Directed vs Undirected Research
  7. Intellectual Property and Scientific Progress
  8. The Future of Education
  9. Exploring Science and Technology
  10. Physics, Complexity Science, and Gaming
  11. The Power of Leverage

AI and Innovation

00:00 - 07:05

  • AI has underperformed and drastically performed in a short amount of time.
  • Sam Arbusman is the scientist in residence at Lux Capital.
  • They discuss startup strategies for technical startups like Boston Dynamics and DeepMind.
  • They explore how to accelerate innovation across industries.
  • Science fiction can be a leading indicator of technological advancements.
  • Scientific facts change over time.
  • Sam shares his perspective on technologies that will become massive companies in the future.
  • Sam's grandfather influenced his interest in science, technology, and science fiction.
  • Science fiction explores the societal implications of technological advancements.
  • Harry Selden from Isaac Asimov's Foundation trilogy is an interesting early influence on Sam.

Knowledge and Science

06:47 - 12:54

  • The Foundation Trilogy by Isaac Asimov has influenced many people in the quantitative social sciences field, including economist Paul Krugman.
  • The idea of understanding the statistical properties of societies as a whole, despite the unpredictability of individual decisions, is intriguing.
  • Network science and complex science are fields that aim to quantify how different systems work, whether biological, technological, or social.
  • The Half-Life of Facts explores how knowledge changes over time and how patterns emerge when looking at large masses as systems.
  • Similar to radioactive materials, individual facts may be unpredictable, but when grouped together, they become more predictable.
  • There are regularities in how knowledge grows and changes, such as the doubling time for papers in a field or the half-life of certain fields.
  • Quantitative social understanding includes fields like meta-science or meta-research that study the science of science.
  • Ongoing tracking of verifiability and changes in scientific conclusions would be valuable for ensuring accurate information is used in laws and policies.
  • Science is a human endeavor with its own biases and idiosyncrasies.

Scientific Research and AI

12:35 - 18:35

  • Scientific research is a human endeavor with biases and limitations.
  • AI techniques can be used to analyze scientific literature and determine the nature of citations.
  • Undiscovered public knowledge refers to information that is known but not connected in the literature.
  • Keyword searches and medical databases were used in the past to discover connections between papers.
  • AI can help stitch together different domains of knowledge.
  • Jargon barriers still exist despite the availability of information on the internet.
  • Commonly accepted scientific facts can be overturned or refined over time.
  • The Lindy effect suggests that things that have been around longer are more likely to stay, but there are exceptions.
  • Scientific knowledge undergoes a process of refinement rather than complete reversal.

The Process of Science

18:13 - 24:29

  • Isaac Aswell's quote highlights the idea that our understanding of truth is constantly evolving and refining.
  • Science is a process of refinement, where ideas are constantly being worked on at the frontier of knowledge.
  • There is a divide between the perception and reality of science, with a body of knowledge constantly being referred to while new discoveries are made at the frontier.
  • Science is more about the method of inquiry and constant learning than just a body of knowledge.
  • Scientists embrace the overturning of ideas as it signifies progress and more to learn.
  • The motto of the Royal Society reflects this mindset: 'Nulis in verba' meaning 'Take nobody's word for it.'
  • Academic rigor in science is similar to market rigor, where beliefs are constantly checked against physical or economic realities.
  • Examples like Boston Dynamics and DeepMind show how companies can be built around valuable scientific ideas before they become commercializable products.

Innovation and Research Organizations

24:06 - 30:51

  • Building something valuable in the field of reinforcement learning could make a company an appealing acquisition target.
  • Expanding the types of structures for organizations and companies is exciting and allows for more collaboration between business, research, and science.
  • Changing incentives from academic to commercial can accelerate the timeline for technology hitting the market.
  • Determining when an idea should transition from research to startup is challenging but important for impact.
  • Not all research products should become companies; finding a use case that people actually want is crucial.
  • There are still limited situations where an idea that feels like research should be a traditional startup.
  • Alternative research organizations explore new institutional forms beyond deep tech startups, university labs, and corporate labs.
  • The high-dimensional space of institutions offers opportunities for combining startups and research organizations or creating entirely new forms.
  • The Cambrian explosion of different organizational forms will hopefully yield templates for others to use.

Directed vs Undirected Research

30:24 - 36:43

  • There are times when a clear coordinating function is needed to move things forward, but there are also instances where undirected research is beneficial.
  • Undirected research allows for curiosity-driven exploration and the discovery of novel ideas.
  • Combining different discoveries can lead to innovative solutions.
  • Both directed and undirected approaches are necessary for most innovation.
  • Progress in undirected research requires sample size and time through the law of large numbers, while directed research relies on precise understanding and strategic decision-making.
  • The outcome of these processes gives a sense of agency and optimism in pushing progress forward.
  • Alternative research organizations like Think and Switch operate using a studio model, bringing together teams of experts for specific projects.
  • This modular approach allows them to take advantage of top-notch talent who may be available for shorter periods of time.
  • Tech transfer offices play a role in commercializing discoveries from academic institutions, but there are challenges in transferring intellectual property (IP) effectively.
  • Alternative research organizations aim to sidestep some of these challenges by creating their own IP that can be licensed and create value.
  • There is still ongoing exploration on how to balance IP capture with the public good nature of science.

Intellectual Property and Scientific Progress

36:14 - 42:52

  • There is a need to find the right balance between intellectual property (IP) ownership and capturing value.
  • Universities often strive to capture more value from their IP but may hinder scientific progress as a result.
  • Tech transfer offices put onerous terms on cash payments or royalties, which may not align with long-term success.
  • Universities can afford to play a longer game than other institutions, but career risks and short-term incentives may hinder progress.
  • There is an opportunity to examine and modify the terms and conditions of tech transfer for better outcomes.
  • Best practices in education may need to be reevaluated in light of emerging technologies like AI.
  • AI has the potential to disrupt education by enabling personalized and self-paced learning.
  • Rather than preventing students from using AI tools, schools should embrace them and enhance educational opportunities.
  • Universities serve multiple functions including research, education, and socialization.

The Future of Education

42:35 - 49:18

  • The university experience serves as a socialization function for young people to learn how to be adults.
  • Different roles and functions of universities should be examined and modified if necessary.
  • Lower tier universities may be more willing to experiment and change rapidly compared to top-tier universities.
  • Arizona State University has engaged in interesting experimentation at the university level, such as connecting scientists with science fiction writers.
  • Education will undergo significant changes, including personalized and interactive educational content with AI technology.
  • Technology should be used in partnership with humans in education, similar to calculators being used alongside basic math skills.
  • AI tools like chat GPT can be used for fun, generating ideas, and probing the underlying cognitive map within these systems.
  • Language models have their own story worlds that are different from the physical world, influenced by narrative structures and tropes.

Exploring Science and Technology

49:01 - 55:32

  • Language models like Luigi from Super Mario can easily switch into a weird agent of chaos, embodying narrative and story worlds.
  • The scientist in residence at Lux Capital surveys the landscape of science and technology to find areas, communities, and individuals to be involved with.
  • The role involves finding interesting companies or individuals for investment, connecting ideas or individuals to the portfolio of companies, and exploring areas that will eventually be relevant to the investment species.
  • Lux Capital focuses on emerging deep tech and frontier tech, including weird science fictiony things.
  • Clean tech has had a resurgence recently, with renewable cost curves coming down, but it hasn't had the returns venture capital wanted.
  • The distinction between real ideas that can become outside hits, real ideas not ready for venture yet, and not real ideas is an interesting taxonomy.
  • Augmented reality and virtual reality have underperformed in terms of personal interest.
  • AI has both underperformed and drastically overperformed in a short amount of time.

Physics, Complexity Science, and Gaming

55:03 - 1:01:33

  • The intersection of physics, complexity science, artificial life, and gaming is deeply connected and interesting.
  • Recommendations for sci-fi books include the Three-Body Problem trilogy, Neil Stevenson's works (especially the Baroque cycle), Ian Banks' Culture novels, and the novel Babel.
  • The Three-Body Problem trilogy starts at our current state and takes readers to unimaginable places.
  • Neil Stevenson's historical works are fascinating, such as the Baroque cycle which explores the scientific revolution and the invention of the modern monetary system.
  • Ian Banks' Culture series is set in a post-scarcity society where humans can do whatever they want with super intelligent minds running everything.
  • Babel is a fantasy novel that explores translation as a technology with implications for industrial evolution, empire, colonialism, and language.

The Power of Leverage

1:01:12 - 1:07:29

  • Our brains aren't evolved to comprehend how much leverage is possible in modern society.
  • There's a revolution going on, man. Pay attention to it and get exposed to it.
  • You're going to make money along the way and have fun.
  • This is the new form of leverage.
  • Take a few quiet moments for yourself. Breathe deep and be well.