Episode #66: Three Breakthroughs: Digitizing Smell Using AI

Frequently Asked Questions

FAQ

• Where is the transcript for this webinar?

  Samantha Herrick:
  Welcome to The Tech Optimist. This is a podcast brought to you by Alumni Ventures, a show where we tell you the stories of tomorrow.

  Mike Collins:
  As these technologies—which are so exciting and so profound in areas of robotics and AI—these have societal implications.

  Samantha Herrick:
  That is Mike Collins, co-founder and CEO of Alumni Ventures.

  Naren Ramaswamy:
  The applications that this can have are numerous. It’s giving a machine a new sense.

  Samantha Herrick:
  That is Naren Ramaswamy, Senior Principal at Alumni Ventures. And that’s me. My name is Sam, and I am the tech note writer, editor, and overall lover of the show.

  Hello everyone. Welcome back to a Three Breakthroughs episode of The Tech Optimist. You met our guests, you met me, you know the speakers of this episode. Now we’re going to do something a bit different today. I don’t want to tell you the topics today. I want you to be surprised.

  But I can definitely give you some hints about the breakthroughs that Naren and Mike are going to get into today: we’re going to get into robots, we’re going to get into bugs, and we’re going to get into boats. Those are your hints. I’ll stop talking and we’ll hop right into the episode. Please enjoy the show.

  As a reminder, The Tech Optimist Podcast is for informational purposes only. It is not personalized advice and it is not an offer to buy or sell securities. For additional important details, please see the text description accompanying this episode.

  Mike Collins:
  Welcome to The Tech Optimist Podcast. This is our Three Breakthroughs show. I’m here with Naren Ramaswamy. Welcome, Naren.

  Naren Ramaswamy:
  Hi, Mike. Excited to do this.

  Mike Collins:
  Yeah, so this is a show where we try to pull together three things that caught our eye—things that we think have profound implications for technology, venture capital, entrepreneurship, and innovation.

  The first one sounds a little odd: the mapping of a fruit fly brain. Probably can’t say that three times in a row—lucky to say it once correctly. But this was a big project that involved about 75 labs and 300 researchers.

  Samantha Herrick:
  Okay, this breakthrough is so fascinating to me and to the entire AV team. I actually found an article from Smithsonian Magazine titled Scientists Unveil the First-Ever Complete Map of an Adult Fruit Fly’s Brain Captured in Stunning Detail.

  We’re going to display a visual or video of this colorful, digital map of the fruit fly’s brain while I read a few parts of the article. This article was written by Margharita Bassi, a daily correspondent for Smithsonian Magazine, and it came out on October 4th of this year.

  I’ll read a few sections because they give more context to what’s going on. It’s also a really good segue into Naren’s breakthrough on AI and machines that can smell.

  The researchers used artificial intelligence to reconstruct the fly’s brain cells based on image data. But the model wasn’t perfect and made some mistakes. So, the researchers created the Flywire Consortium to recruit hundreds of volunteers to proofread and annotate the 3D brain—called a connectome—by hand.

  At the end of this monumental collective effort, the team produced the most complete brain map of any organism to date. It included nearly 140,000 neurons, 8,453 different types of neurons, and more than 54.5 million synapses.

  The findings were published in a series of nine papers in Nature. Unraveled end-to-end, the fruit fly’s brain neural wiring could stretch out to more than 490 feet—longer than four blue whales aligned nose to tail. This was reported by Carl Zimmer of The New York Times.

  How can this help us in the future? The connectome and its data reveal patterns of healthy brain function, which could be consequential for treating brain diseases. Of the 8,453 neuron types, scientists identified 4,581 that were newly discovered.

  The brain map also provides insights into neurotransmitters such as dopamine and serotonin secreted by various types of neurons. Anita Devineni, a fruit fly expert at Emory University (not involved in the study), told The New York Times that she relies on this publicly available data to plan new experiments.

  Scientists hope to eventually map the brain of a male fruit fly to complement this female brain and then move on to more ambitious projects, like creating a connectome for a mouse brain—which has over 1,000 times as many neurons as a fruit fly. That’s pretty bonkers.

  Mike Collins:
  Obviously, fruit flies are a basis for a lot of research. Scientists use them for many reasons. The work being done to map their brains in such detail really points the way toward mapping the human brain.

  When we can do human brains, there’s exciting potential for disease treatment, brain-computer interfaces, and even longevity applications where uploading or downloading your brain might be possible—maybe even in our lifetime.

  These advances will first help people with diseases or brain injuries where the cost-benefit is profound. But again, I think this is a huge breakthrough. It’s noteworthy and not talked about enough. Understanding the human brain is an area where we’re making great progress.

  Naren Ramaswamy:
  Absolutely.

  Mike Collins:
  I’m not diving into exactly what this all means—people can read more about it in the show notes—but as a breakthrough, I think it’s big. Any thoughts, Naren?

  Naren Ramaswamy:
  It’s amazing. When you think about biology as the study of the human body, the tools available now to biologists at the cutting edge are more advanced and accessible than ever before. It’s an exciting time to be a scientist.

  And this is just biology—we’ll talk about a couple of other breakthroughs too. As we’ve discussed in past episodes, the pace of innovation is increasing.

  The multidisciplinary implications are fascinating because neural networks that govern AI today were originally architected based on what we understood of the brain. Now, thanks to AI, we have an even better understanding of the brain. This creates a virtuous cycle. It’s fascinating.

  Mike Collins:
  And then, seeing again, some Nobel Prize announcements—it’s Nobel Prize season—and looking at where these prizes are going now. We’ll probably talk more about this in future episodes.

  A couple of the winners were working in the area of biochemistry with AlphaFold, understanding proteins and protein folding. Again, this isn’t my area of expertise, but it’s clear that AI tools are deeply improving our understanding of how the human body works. These tools create feedback loops and the ability to design small molecules with systems like AlphaFold and its spinouts and startups.

  These are some of the most exciting investment areas in life sciences. They’re going to fundamentally change medical discovery—from bespoke, trial-and-error approaches where we stumble upon discoveries (like some reptile spit producing unpredictable results in a lab) to a more engineered, targeted approach.

  In the future, we’ll be able to ask: What do we want to solve? What result do we want? What kind of molecule or protein do we need to make that happen?

  When it becomes an engineering problem, results improve geometrically—better speed, efficacy, and fewer side effects. These are very related breakthroughs that point to a revolution in discovery.

  Naren Ramaswamy:
  Yeah, that’s a great point about the discovery process becoming more engineering-centric rather than just serendipity. It’s fascinating and it’s a good segue into the breakthrough I wanted to talk about.

  Samantha Herrick:
  Don’t go anywhere. We’ve got a commercial break, and then we’ll be right back.

  Speaker 5:
  Exceptional value creation comes from solving hard problems. Alumni Ventures’ Deep Tech Fund is a portfolio of 20 to 30 ventures run by exceptional teams tackling huge opportunities in AI, space, energy, transportation, cybersecurity, and more.

  These game-changing ventures have strong lead venture investors and practical approaches to creating shareholder value. If you’re interested in investing in the future of deep tech, visit av.vc/deeptech to learn more.

  Naren Ramaswamy:
  We’ve talked about AI that can read the internet, see images, and provide insights. It can hear, you can talk to it, and it can respond. But one sense it hasn’t touched yet is smell and aromas. Scientists are working on that as well.

  Researchers are now able to predict the characteristics of certain smells and even create new scents. A pioneering company in this space is called Osmo. We’re not an investor—at least not yet—but it’s really cool tech, so I thought it was worth highlighting.

  Samantha Herrick:
  In regard to the breakthrough Naren shared, Toyota researchers put out a really awesome video about a year ago showing the advances they’ve made with their robots. The video is titled Teaching Robots New Behaviors.

  These behaviors are fascinating—mainly new dexterous actions learned from human demonstrations. Just watching this video and hearing from the researchers—and then pairing that with Naren’s breakthrough—it’s incredible to think about what scientists, engineers, and contractors can build.

  If AI continues to advance and we truly start replicating biology and the human brain—especially with the fruit fly brain mapping we discussed earlier—and if the mechanical side catches up with the AI side, we’re in for some amazing breakthroughs.

  Here’s that video from Toyota. Afterward, we’ll get right into the interview. Enjoy.

  Russ Tedrake:
  Here at TRI, we believe robots should amplify people at home and at work. To achieve this, we conduct cutting-edge research to make robots more flexible, robust, and general-purpose.

  We’ve had a breakthrough. Today, TRI is announcing a new method to teach robots dexterous skills quickly and easily. Our approach is built on a powerful generative AI technique called diffusion policy.

  This allows us to teach robots much faster and with significantly fewer demonstrations. It holds great promise for creating what we call large behavior models.

  Just like large language models have revolutionized chatbots, these behavior models will allow robots to perform useful work in ways never possible before.

  Ben Burchfiel:
  Up until now, most robotic manipulation has focused on pick-and-place tasks, where a robot is limited to rearranging simple sets of objects.

  Our new approach goes far beyond that. It lets us explore much closer to the hardware’s limits. TRI’s robots are now capable of using tools, pouring liquids, peeling vegetables—it’s exciting to see them engaging with their environments in rich, multifaceted ways.

  This is all achieved without changing any code or explicitly programming new skills. Using this technique, we’ve taught over 60 diverse behaviors to our fleet of robots.

  The process starts with a teacher demonstrating a small set of skills through teleoperation. Then, our AI-based diffusion policy learns in the background over a matter of hours.

  It’s common for us to teach a robot in the afternoon, let it learn overnight, and come in the next morning to find it performing a new behavior successfully.

  To make this level of dexterity possible, every part of the robot platform must be solid—from the hardware through the entire software stack.

  One key enabler is giving human teachers a sense of touch through a haptic teleoperation device. Just like people, robots learn better when they have a sense of touch.

  A perfect example is flipping a pancake: the robot must make contact with a surface it can’t see. Without a sense of touch, it struggles. With it, it successfully learns the skill.

  Russ Tedrake:
  This is only the beginning. Our team is focused on achieving the large behavior models I mentioned earlier.

  We anticipate the next breakthrough will come when we’ve trained robots with enough dexterous skills that they can generalize—performing new skills they’ve never been explicitly taught.

  To achieve this, we’re building a diverse curriculum for robot learning—a “kindergarten for robots” to teach foundational skills useful for working alongside people.

  We’re on pace to teach hundreds of new behaviors by the end of the year and over 1,000 by the end of 2024. We’re leveraging simulation to augment real-world teaching and developing tools for fleet learning so that when one robot learns, they all learn.

  I’ve been working in robotics research for a long time. The tasks we’re now seeing these robots perform are simply amazing. Even a year ago, I wouldn’t have predicted we were this close to such dexterity. The speed at which we’re teaching new skills is astounding. Nearly every day, I wake up to messages showing a robot doing something it couldn’t do the day before.

  This is an incredible time to be a roboticist.

  Naren Ramaswamy:
  The applications this can have are numerous. It’s giving a machine a new sense.

  For example, we’ve long talked about how dogs can detect cancer because their sense of smell is 10,000 times better than ours. Imagine having AI around you that could monitor your health that way.

  We also use insect repellents, deodorants, and sprays without fully understanding their effects on the human body. With this technology, what if we could engineer healthier, sustainable scents?

  This has far-reaching implications across disciplines. It’s especially pivotal because certain scents have similar molecular structures yet smell completely different.

  Understanding this deeply and building a foundation model to predict what a scent “looks like” and what its characteristics are is fascinating.

  Mike Collins:
  And it’s just interesting too—human sight is highly evolved, and we get a lot of data from it. If we think of all these senses as sensors feeding into our biological computer, the amount of data we get via sight is immense. Our eyes are essentially sensors hardwired directly into the brain.

  But our sense of smell is underdeveloped compared to other animals, like dogs. We don’t even know what we’re missing as human beings. These senses evolved over time, but there’s as much data with smell as there is with sight—it’s just that our system hasn’t developed to capture and process it fully.

  Sound, touch, smell, vision—all of these can be sensed and then processed by the brain. Seeing innovation across all these senses and having a computational silicon brain to process data still requires sensors and inputs.

  I think mapping and modeling these senses is fascinating. Smell, in particular, is such an interesting area. We only see glimpses of its potential, like dogs detecting cancer, but there’s a whole spectrum of information happening there that we don’t understand at all because our sensory system is so underdeveloped compared to others.

  Naren Ramaswamy:
  And it didn’t need to develop. If you look at human history, vision became the primary sense we relied on. The development of other senses was left to chance.

  Now, it’s becoming an engineering problem—which has some very interesting implications.

  Samantha Herrick:
  Okay, we’ve got a second for a sponsor spotlight, and then we’ll be right back. Hang tight.

  Speaker 5:
  Do you have a venture capital portfolio of cutting-edge startups? Without one, you could be missing out on enormous value creation and a more diversified personal portfolio. Alumni Ventures, ranked a top-twenty VC firm by CB Insights, is the leading VC firm for individual investors. Believe in investing in innovation? Visit av.vc/foundation to get started.

  Mike Collins:
  All right, and then the third one we have today is about what’s been going on with a couple of unions—specifically the Hollywood strike and the longshoremen strike.

  Samantha Herrick:
  The longshoremen strike is a significant labor action that began on October 1st of this year, involving dock workers at major ports along the East and Gulf coasts of the United States.

  Here’s an overview: Approximately 45,000 to 50,000 dock workers represented by the International Longshoremen’s Association (ILA) walked off the job. This is the first major East Coast dock strike since 1977.

  The strike affected 36 ports from Maine to Texas, which collectively handle about half of U.S. imports and exports.

  You’re probably asking: “Why did they strike?” The main issue was compensation. The union sought substantial wage increases in a new six-year contract. Technology and job security were also concerns—automation threatened jobs, and there were worries about foreign interests in imports.

  How did this impact business and the economy? It disrupted supply chains and could lead to shortages and price increases. Industries like manufacturing, construction, retail, and eCommerce were majorly impacted. There were concerns about inflation, holiday season product availability, and pricing.

  As of October 3rd, the strike was suspended. A tentative agreement was reached, which includes a 62% pay raise over six years, according to ABC News. From there, we don’t know much else—we assume it’s suspended indefinitely, but sometimes these things take time to fully resolve.

  We wanted to provide this context so that when we get into the rest of Mike and Naren’s conversation, you’ll better understand their discussion.

  Mike Collins:
  What I found interesting—and why this is on the show this week—is that in both cases, part of the negotiations involved salaries and the usual stuff, but also explicitly addressed automation and technology.

  Unions are now identifying technology as a negotiation point. I’m not getting into the specifics of the negotiations, but I’m glad to see that workers are happy and that this choke point in the economy is, at least temporarily, resolved.

  For our show, the point is this: we all need to anticipate that as these technologies—which are exciting and profound in areas like robotics and AI—become more common, they’ll have societal implications. There will be second- and third-order effects.

  The longshoremen are concerned about robots and AI taking their jobs. It’s a difficult but predictable situation.

  If we look back through history, every big breakthrough has shadow trends—secondary effects—that inevitably follow.

  Mike Collins:
  You and I are working on a piece right now about a shadow trend we call the bespoke economy. This theory suggests that as technologies automate work and life, human beings will crave bespoke, high-touch, story-laden experiences, products, and services. We’ll touch more on that later.

  Naren Ramaswamy:
  Personalized stuff.

  Mike Collins:
  Yeah, personalized stuff that really has a human dimension to it. The shadow trend of having any food customized for me and delivered instantly is that I’ll also want to go to a farmer’s market. That’s the counter or shadow trend we’re going to see.

  But I think what we’re seeing with the Longshoremen strike and the Hollywood strike is a third thing: the backlash trend. When change happens, there’s a predictable backlash.

  We saw this with the release of ChatGPT. The immediate reaction from some schools was to outlaw it because they didn’t know how to handle it. If you look back through history, from the Luddites onward, every major technology has sparked backlash.

  We’ll see that with AI and robotics. It’s very difficult to hold back the dam for long, but people should anticipate it. They’ll need to strategically position themselves around these changes.

  There’s a great opportunity in shadow trends. AV invests every week in companies aligned with the main trends in AI, deep tech, space, and more. But we’re also really interested in the shadow and secondary effects.

  Another example: when the automobile was invented, there was huge backlash initially and massive secondary effects. The rise of suburbia, automobile vacations, highway systems—all were driven by that foundational innovation. But there were also enormous deaths on the streets of New York City. The term “jaywalking” emerged because so many people were being run over.

  With today’s innovations—bigger, stronger, more powerful than ever—the secondary effects, shadow trends, and backlashes will also be bigger and stronger than ever before.

  Naren Ramaswamy:
  All of them come together. All the good things of technology bring along their backlashes. And to your point about history—it may not repeat itself, but it does rhyme.

  If you look at the last 200 years, technology has won out every time. Automation emerges, initially making workers’ jobs easier—they work less, and are more productive. But as it improves, eventually the jobs themselves are threatened. Wages stagnate, workers strike—understandably, because their livelihoods are impacted.

  But once that tumultuous period passes, new opportunities and new markets are created that are even bigger in value. Technology has actually lifted societies as a whole. Extreme poverty has significantly diminished over the last 200 years because of technological progress.

  We’re The Tech Optimist show, so it’s important to highlight that.

  Mike Collins:
  And the data supports it. For all the negative unintended consequences, overall these technologies are net positive. Very few of us would want to teleport back 200 years.

  Today, a middle-class person’s quality of life is significantly better than the most resourced person on the planet 200 years ago. Longevity, healthspan, opportunity, literacy, equality—far from perfect and very unevenly distributed globally, but overall, net positive.

  What I’ve observed is that these changes often take a generation. When there’s an economic disruption, it’s hard for humans to change lanes quickly, especially depending on where they are in their life journey. Often, it’s the next generation that leaves the struggling small town built on a disrupted industry.

  The issue now is speed. These changes are happening so quickly. Is society—and especially the younger generation—equipped to flex into a totally different future? We may not have an entire generation to transition this time.

  Naren Ramaswamy:
  It’s going to be quicker.

  Mike Collins:
  Yes, quicker and potentially more abrupt. We need to anticipate that future carefully and think about how to mitigate the challenges. These disruptions shouldn’t be taken lightly. Let’s see them coming, anticipate them, and work through them.

  Naren Ramaswamy:
  Yeah, great point.

  Mike Collins:
  Great. Excellent. Thank you, Naren. We’ll do it again.

  Naren Ramaswamy:
  Thanks, Mike. All right, see you later. Bye.

  Samantha Herrick:
  Thanks again for tuning into The Tech Optimist. If you enjoyed this episode, we’d really appreciate it if you’d give us a rating on whichever podcast app you’re using, and remember to subscribe to keep up with each episode.

  The Tech Optimist welcomes any questions, comments, or segment suggestions, so please email us at [email protected] with any of those and be sure to visit our website at av.vc. As always, keep building.