In this episode of the 20 minutes VC, host Harry Stebbings interviews Aaron VanDevender, the Scientist in Residence at Founders Fund, exploring the integration of scientific expertise into venture capital. VanDevender discusses his transition from quantum computing and academia to the VC world, his role in assessing new technologies, and the impact of scientific diligence in investment decisions. The conversation also delves into the potential overvaluation of AI and data sets, the societal implications of self-driving trucks, and the concept of universal basic income. Additionally, VanDevender highlights his excitement for an investment in IR Labs, a company working on silicon photonics for optical computing. Insights from Kevin Hartz and Brian Singerman of Founders Fund are acknowledged for facilitating the interview.
"So in the past few years we've seen some of the most forward thinking funds integrate scientists in residence or chief scientists into their fund." "Now, Aaron is the scientist in residence at Founders Fund, one of the world's leading funds with investments in the likes of Facebook, Airbnb, SpaceX, Spotify and many more incredible companies."
The quotes highlight the trend of venture capital funds incorporating scientific expertise into their teams and introduce Aaron VanDevender's significant scientific background and current role at Founders Fund.
"And so they said, we are looking at doing more of these types of high technology, high science risk kinds of companies, and we really want to have your expertise on board within our team to help evaluate them and cultivate some of those opportunities." "One is providing a scientific context for the investments that we make...The second part is, once we get interested in something, into a company doing the scientific diligence...And there's a very thin line between genius and crazy, right."
Aaron describes the multifaceted role of a chief scientist in a VC firm, which includes evaluating high-risk, high-tech investment opportunities, providing scientific context, and distinguishing between feasible and infeasible scientific ventures.
"So part of it is thinking it through from first principles...One is just how thorough is the knowledge and the understanding of the people that are proposing it...If it turns out that the way of thinking about it is very shallow, then that typically reveals itself pretty quickly."
This quote emphasizes the importance of first principles thinking and depth of knowledge in evaluating scientific ideas for investment, distinguishing between those that are just crazy enough to work and those that are simply unfeasible.
"Yeah, so I think that most of the things that call themselves AI out there are really just a rebranding...And so they sort of got together and decided that this is a terrible place to be in terms of the status relative to the value that they thought they were providing. And so they rebranded, and it's gone through several iterations."
Aaron's quote suggests a critical view of the AI industry, arguing that its current prominence is due more to strategic rebranding from less glamorous origins in actuarial science rather than substantial advancements or innovations.
"But that sort of petered out. And so then they got rebranded as data science, and then as big data, and then as machine learning, and then." "Now most recently as artificial intelligence." "But at the end of the day, what most of the things are actually doing there is just actuarial science with." "Somewhat better computers, like actuarial science that has AWS kind of computing ability applied to it."
The quotes highlight the progression and repeated rebranding of the field, suggesting skepticism about its true innovation. The comparison with actuarial science implies that the core function has not significantly changed, only the computational power has improved.
"Anytime that you have something that's continually rebranding, that continuously has to come back and call itself something new, that's a good indication that it's a scam, that it didn't live up to the promise the last time. And so now it has to find." "A new way to make promises."
These quotes express a critical view of the constant rebranding of data science and AI, suggesting that it could be a tactic to mask underperformance or to create new hype after previous iterations did not meet expectations.
"One way of describing artificial intelligence or a particular deep learning, since at the core of it is matrix multiplication operations." "You're sort of doing linear algebra at industrial scale." "We haven't gotten to the point for." "Most of these things or most of." "The things that call themselves AI, where." "You're really doing something that is fundamentally." "Different, you're just making it bigger or just having larger scale doesn't actually change." "The nature of the calculation." "And I would even sort of push back on when people tout their big data credentials that most of the things." "That are sort of big data are actually not that big. You'd like to think of these petabytes and exabytes floating around there and coming up with really amazing insights that could never have been seen. But a lot of times it's really in the sort of megabytes and gigabytes kinds of range. And it's always beneficial to say that you have this big data edge when really it's just more marketing and branding."
The quotes explain that the core operations of AI are not fundamentally different from traditional calculations; they are just performed on a larger scale. Moreover, the actual size of big data is often exaggerated, and the term is used for its marketing appeal rather than its technical accuracy.
"So the place where I look for the real breakthroughs in artificial intelligence and where I would like to see the." "Field progress more, is in the way that it deals with ambiguity." "I think that the biggest distinction between." "Human intelligence, which is what artificial intelligence." "Is meant to model after or meant." "To work towards, is that humans have." "A much higher tolerance for ambiguity. In computers, where you have things are." "Explicitly programmed, you have logic. Whenever there is a sort of corner case that wasn't envisioned by the programmer." "Or a kind of boundary condition that." "Is not explicitly handled, then things break down, and so it becomes very fragile. And that's exactly the kind of situation that our modern homo sapien brains were." "Evolved to process through, and that gives." "Us a huge evolutionary competitive advantage." "So being able to, when you have things you haven't seen before, to be able to think about it in a higher level of abstraction and deal with." "The ambiguity in a reasonable way and still be able to move forward. And most of the AI applications that we are using out there, deep learning, pattern matching kinds of things, can't actually do that, can't use abstract thinking to deal with ambiguity. They can only sort of say, well, this is just the closest thing that I've seen to it before, and so we're just going to go with that."
These quotes emphasize the importance of developing AI that can handle ambiguity, much like human intelligence. The ability to process unforeseen situations and think abstractly is highlighted as a key area for AI breakthroughs, as current AI systems are limited to pattern matching and lack the capacity for abstract reasoning.
"Well, I think there are sort of layers to that ambiguity." "Right." "There's different levels of abstraction that the ambiguity can present itself, and at some point we will be able to deal with a full layer, which includes things." "Like moral ambiguity, and be able to make rational, reasonable decisions about that." "And so I think when you get to the point of human level artificial general intelligence, then that's the kind of thing you're talking about. But there are still lots of problems that I think we can be able." "To solve where things are just sort of visual ambiguity, mechanical ambiguity, these types." "Of things that hopefully we'll be able." "To make serious progress on the way to the human level general intelligence, and."
These quotes discuss the concept of general artificial intelligence and the importance of handling different layers of ambiguity. The speakers suggest that the ability to deal with various types of ambiguities, including moral ones, is crucial for reaching the level of human intelligence in AI.
"No, not really. Most of the data sets out there, or the defensibility of data sets out there, I think, are oversold. It's a very compelling, very tantalizing notion that, oh, I just have so much data and nobody else has it, and therefore it must be special. But most of that data out there tends to be noise, and it's just a burden to process. Most of your algorithms are designed just to filter through it and not actually." "Learn anything from it." "And so that tends to not be." "Nearly as valuable as people think it is." "The other thing that, where people get trapped up about the specialness of their data sets, is that you would like to believe when a data set truly is valuable, it has this property that each new incremental piece of data is marginally more valuable to the whole thing than the last one that you put in. So you get this kind of accelerating effect, rather than a diminishing returns kind of effect. Most of the data sets out there." "Have this diminishing returns effect. A good example is like genomic databases, where people try and say, okay, well, we sequenced 10,000 people instead of 1000 people or 100,000 people, implying that their." "Data sets are therefore ten times more valuable because they have ten times more sequences. But if you don't actually see any." "New genes or any new alleles in." "Those ten x more sequences, you're just." "Seeing repeated copies of the earlier ones." "Then you get this sort of diminishing." "Return effects for your effort." "A lot of the asymptotic ceilings of those data sets happen pretty early. And so thinking, well, I spent all this effort, now I have this really valuable data set. Well, you probably could have done 90%." "As well with one that was 10% as big, and competitor comes along, and." "Maybe they not have as much resources." "But they can still do a decent job, then it doesn't hold up as a competitive advantage."
The quotes express skepticism about the actual value and uniqueness of large data sets, suggesting that the volume of data is not necessarily a competitive advantage. They point out that most data sets exhibit diminishing returns rather than an accelerating value with the addition of new data, and that competitors can often achieve similar results with smaller, more manageable data sets.
"I think a lot of folks in." "The self driving car space maybe just assumed that since Google had already done the work of getting all of the Street View data, that they had this insurmountable advantage, that no one would sort of be able to build self driving." "Cars because they didn't have the same street level data. But it turns out that most of our streets are not really that different. And so, even if you have a much, much smaller data set than Google." "You can still do a really good job of developing a self driving car." "And in fact, it almost forces you." "Into a better position because if you have complete knowledge of the entire street." "Grid, you can succumb to the temptation." "Of programming in individual use cases where." "Your agent would come to something that's ambiguous and you can just hard code in the right answer, whereas if you only have access to a small amount." "You are forced to make your algorithms more robust. And so you end up with a." "Better result because they have to deal." "With that ambiguity directly."
These quotes challenge the notion that having a massive data set, like Google's Street View data, is essential for developing self-driving cars. They argue that smaller data sets can lead to the development of more robust algorithms as they force companies to handle ambiguity directly, rather than relying on hard-coded solutions.
"Yeah, I think the societal implications are really huge. I think it's almost impossible to understate." "How big of a deal it is."
The quote acknowledges the magnitude of the societal impact that self-driving trucks could have, particularly in terms of employment for truck drivers. The speaker believes the implications are substantial and cannot be overstated.
"And so if you think that truck drivers, which are a thousand times more causing the same type of redundancy and displacement as coal miners experienced as natural gas, came online, then is a big potential for societal upheaval."
This quote emphasizes the scale of potential job displacement and societal disruption that could result from the automation of truck driving, comparing it to the significant impact the natural gas boom had on coal miners.
"The whole trucking system serves two purposes at the same time. One, it's a goods distribution system... But it's also a wealth distribution system."
This quote explains the dual role of the trucking industry, not only in distributing goods but also in spreading wealth across various points in the network, particularly in service-oriented towns that rely on truckers.
"So the best case scenario is we use that goods distribution to our advantage."
This quote suggests that society can benefit from the improved goods distribution system brought about by automation if it can harness it to create new economic opportunities and jobs.
"It's really going to be sort of the jobs that kind of don't exist yet, that can't exist because goods distribution is so expensive. But if the goods distribution network gets so much better, then all those indirect jobs can be created."
This quote highlights the potential for new job creation that may arise from the efficiencies gained in goods distribution through automation, although these jobs are not yet defined or existing.
"My biggest concern... is we do have trouble decoupling wealth and virtue in our political system."
This quote reflects the speaker's concern that the political system's difficulty in separating concepts of wealth and virtue may complicate the implementation of UBI.
"It's like, well, so we changed the goods distribution system. We changed the wealth distribution system that has been a consequence of the trucking system. And so now we're just trying to sort of force a replacement wealth distribution system that we can manage."
This quote critiques the idea of UBI as a forced replacement for the existing wealth distribution system without considering the broader potential of the improved goods distribution network.
"One of the myths that we have told ourselves about technology is that we can use it to sort of create a situation where place doesn't matter, where proximity doesn't matter."
This quote challenges the common belief that technology diminishes the significance of physical location, pointing out that proximity continues to be important in various aspects of life.
"These are all things that were born on the Internet, came out of that digital technology, but they actually make proximity more valuable."
This quote underscores the irony that digital innovations, which were once thought to make physical location irrelevant, have instead made proximity even more valuable through services that connect people and businesses in the physical world.
"The technology doesn't separate you from the people; it actually makes being near people more valuable. It amplifies the upside of proximity."
This quote explains that technology, contrary to separating people, has made physical proximity even more valuable by amplifying the benefits of being close to others.
"The accelerating rising cost of housing, especially in high productivity cities like San Francisco, where in order to really get access to a lot of those kinds of services that give you employment opportunities and relationship opportunities and educational opportunities and all the great things in life."
This quote highlights the challenge of the high cost of living in cities where there are abundant opportunities for personal and professional growth.
"My favorite book is tales of a 1001 arabian nights. The complexity and the layering of it really appeals to me."
Aaron expresses his appreciation for the complexity and narrative structure of "Tales of a 1001 Arabian Nights," which has multiple interwoven stories.
"Quantum computing in 60 seconds, why so exciting? And where are the opportunities super exciting because of the computational potential, is dramatically more than classical or von Neumann computing."
This quote summarizes the excitement around quantum computing, emphasizing its superior computational capabilities over traditional computing systems.
"The hardest element is the context switching. One of the things I liked most about being a research quantum physicist is that you can think about one problem all the time."
Aaron discusses the difficulty of having to frequently switch between different topics, which is a contrast to his past focused research work.
"Well, I guess I'm a little bit old school in that I still love to read slashdot.org."
Aaron shares his preference for Slashdot, highlighting its historical importance and continued value despite the rise of newer platforms.
"I'm really excited about IR Labs or smaller investments from FF Science and what they are doing is silicon photonics."
Aaron expresses enthusiasm for IR Labs' work in silicon photonics, indicating a transformative potential in the field of computing.