September 12, 2023
If you’re looking for help thinking about the implications of exponential change in all areas of technology, one of the best people you can turn to is Azeem Azhar. He's a writer, entrepreneur, and investor who publishes the incredibly popular and influential Substack newsletter Exponential View, which takes deep dives into AI and other subjects with world experts. In 2021 Azeem published a whole book along the same lines called The Exponential Age: How Accelerating Technology is Transforming Business, Politics, and Society, and he joined Harry on the show in early 2022 to talk about that. This summer, the book came out in paperback—and just this month, Azeem worked with Bloomberg Originals to launch a limited-run TV show and podcast called Exponentially with Azeem Azhar. So it seemed like a great time to revisit Harry's 2022 interview, which resonates with current events even more now than it did when we first aired it.
56 min 35 sec
8.29.23
Professionals in drug discovery, drug development, and healthcare may not grasp the scale of the change that’s coming to their business thanks to generative AI models like GPT-4. They need to get up to speed fast if they want to stay competitive and incorporate generative AI into their work in a way that’s effective and safe. Fortunately there are plenty of people in the life sciences industry thinking about how to help with that. And one of them is Harry's guest this week, Lana Feng. She’s the CEO and co-founder of Huma.ai, and under her leadership the company has been working with OpenAI to find ways to adapt large language models for use inside biotech and pharmaceutical companies. GPT-4 and competing models are extremely powerful. But for a bunch of reasons that Lana explains in this episode, it wouldn’t be smart to apply them directly to the kinds of data gathering and data analysis that go on in the biopharma world. Huma.ai is working on that problem. They’re building on top of GPT-4 to make the model more private, more secure, more reliable, and more transparent, so that companies in drug development can really trust it with their data and not get tripped up by issues like the hallucination problem.
50 min 13 sec
8.15.23
Harry's guest this week is Joe DeVivo, the new CEO of Butterfly Network. The company's goal is to make it radically easier for doctors or medical technicians to perform an ultrasound exam on any part of the body, and radically cheaper for a patient to get one. The companyt makes an FDA-cleared, handheld ultrasound scanner called the Butterfly iQ. The first big thing that’s different about the iQ is that it uses silicon-based microelectromechanical sensors, instead of a traditional piezoelectric crystal element, to generate and receive the ultrasound waves. That means the device is fully digital, rather than analog. The second big thing that’s different is that the iQ transmits the ultrasound data to a standard iPhone or iPad instead of a big, expensive ultrasound cart. The doctor or technician can see the live ultrasound image right on a handheld device, and use the image to aim the sensor correctly to get the best possible picture to make a diagnosis. All of that is bringing down the cost of equipping a clinic with ultrasound technology dramatically, and over time it should also bring down the cost of administering an ultrasound exam. It also opens up the possibility of adding AI assistance to the software, so that doctors or technicians can get usable images with less training. The net result is that Butterfly is making it economically feasible to use ultrasound for diagnostic imaging in a lot more places, including clinics in developing countries where ultrasound was out of reach before due to the high cost of the technology and a shortage of trained ultrasonographers.
54 min 20 sec
8.1.23
This week Harry's guest is....Harry! We're flipping the script and giving Harry a chance to wax eloquent about AI in healthcare and drug research, the growing role of personal health monitoring devices, the unique features of the Boston life science ecosystem, the meaning of the recent downturn in biotech investment, the most common mistakes made by new entrepreneurs, and much more. This week's guest interviewer is Wade Roush, who hosts the tech-and-culture podcast Soonish and has been the behind-the-scenes producer of The Harry Glorikian Show ever since Harry started the show in 2018.
65 min 34 sec
7.18.23
Harry's guest this week is Dr. Isaac Kohane, chair of the Department of Biomedical Informatics at Harvard Medical School and co-author of the new book The AI Revolution in Medicine: GPT-4 and Beyond. Large language models such as GPT-4 are obviously starting to change industries like search, advertising, and customer service—but Dr. Kohane says they're also quickly becoming indispensable reference tools and office helpmates for doctors. It's easy to see why, since GPT-4 and its ilk can offer high-quality medical insights, and can also quickly auto-generate text such as prior authorization, lowering doctors' daily paperwork burden. But it's all a little scary, since there are no real guidelines yet for how large language models should be deployed in medical settings, how to guard against the new kinds of errors that AI can introduce, or how to use the technology without compromising patient privacy. How to manage those challenges, and how to use the latest generation of AI tools to make healthcare delivery more efficient without endangering patients along the way, are among the topis covered in Dr. Kohane's book, which was co-written with Microsoft vice president Peter Lee and journalist Carey Goldberg.
58 min 38 sec
7.5.23
In the same way that written English is built around an alphabet of just 26 letters, all life on Earth is built around a standard set of just 20 amino acids, which are the building blocks of all proteins. And just as we've invented special characters like emoji to go beyond our standard letters, it turns out that biologists can expand their repertoire of powers using non-standard amino acids—those that either occur rarely in nature, or that can only be made in the lab. GRO Biosciences, a spinout from the laboratory of the renowned synthetic biology pioneer George Church at Harvard Medical School, is one of the companies working to explore the exciting applications of non-standard amino acids (NSAAs), and Harry's guest this weeks is GRO's co-founder and CEO, Dan Mandell. He says NSAAs could help overcome some of the limitations that keep today’s gene and protein therapies from being used more widely, while also expanding the kinds of jobs that protein-based therapies can do.
60 min 38 sec
6.20.23
Owning a dog can be a joy, but one sad downside is that dogs are highly prone to cancer—six million of them are diagnosed with the disease in the U.S. each year. Harry's guest this week, Christina Lopes, is co-founder and CEO of a company called One Health that's working to improve cancer outcomes for our canine friends. The company offers a precision cancer diagnosis and treatment service called FidoCure that takes what we’ve learned about genomic testing of tumors in humans and uses it in veterinary clinics. Vets can submit a dog’s tumor sample for DNA sequencing, and FidoCure's report will show whether the animal has specific mutations that could help determine which cancer drug will be most effective. Harry and Christina talk about how that process works, why dogs are more vulnerable to cancer in the first place, where she got the idea for the company, and how One Health's work could benefit dogs and humans alike.
62 min 1 sec
6.6.23
Unlike cancer, brain diseases like epilepsy, Alzheimer’s disease, or depression don't tend to have easily measured biomarkers that could help doctors tailor treatments, or that could help researchers develop more effective drugs. So in neurology and psychiatry, the precision medicine revolution hasn't really arrived yet. But Beacon Biosignals, where Harry's guest Jacob Donoghue is the co-founder and CEO, is trying to change that. Beacon is focused on making electroencephalography into a more reliable and useful data source for diagnosing and treating neurological disease. EEG has been a common medical tool for almost 100 years, but interpreting an EEG readout is slow and expensive—all of which makes it the perfect candidate for machine learning analysis. By using computation to peer deeper into EEG data, Donoghue thinks it should be possible to identify subtypes of problems like epilepsy or Alzheimer’s, and help neurologists understand which patients will respond best to which therapies. On top of that, better EEG measurements could also give drug developers and regulators more clinical endpoints to measure when they’re trying to evaluate the safety and efficacy of new drugs for CNS diseases. If Beacon’s vision comes true, the precision medicine revolution might finally start to reach the brain.
42 min 54 sec
5.23.23
Large language models are already changing the business of search. But now they’re about to change the practice of medicine. Harry's guests, Vivek Natarajan and Shek Azizi, are both researchers on the Health AI team at Google, where they're pushing the boundaries of what large language models can achieve in specialized domains like health. This spring their team announced it would start rolling out a new large language model called Med-PaLM 2 that’s designed to answer medical questions with high accuracy. (The model got an 85 percent score on the U.S. Medical License Exam, the test all doctors have to take before they’re allowed to practice.) It's been clear for a while that consulting with an AI would eventually become an indispensable part of every medical journey—whether you’re a patient searching for information about your symptoms, or a doctor looking for an expert second opinion. And now that such a future is almost here, the work Vivek and Shek are doing at Google feels both exciting and a little bit scary.
60 min 54 sec
5.9.23
Harry's guests this week are Sri Kosaraju, the CEO of Inscripta, and Richard Fox, a former Inscripta scientist who just rejoined the company. In reabsorbing Infinome—the Inscripta spinout Fox described to Harry in a spring 2021 episode of the show—Inscripta is placing a big bet on biomanufacturing, the creation and fermentation of genetically customized microbes that can pump out medical, agricultural, and nutraceutical products, and more. Inscripta had previously focused on a benchtop "bio-foundry" machine called Onyx that that makes programmed edits to bacterial or yeast cells at thousands of different points in their genome in parallel. Now it's pivoting away from selling the machine and instead focusing on becoming a "power user" of its own technology—with the ultimate plan of marketing multiple biomanufactured products.
55 min 38 sec
4.25.23
Harry's guest Jen Nwankwo is the founder and CEO of the drug discovery company 1910 Genetics. The company focuses on finding the most promising new drug candidates for stubborn health problems—and it takes a refreshingly agnostic approach to everything else. 1910 doesn’t hunt for just small-molecule drugs or just protein therapies. It explores both. It doesn’t utilize just one form of neural networking or machine learning. It uses whatever model produces the best science for a given problem. It doesn’t hunt for drugs using just wet lab data or just computational simulations. It does both. It isn’t just assembling its own pipeline of drugs or just partnering with larger pharma companies. It’s working on both. At a time when AI and machine learning focused drug discovery companies are sprouting up faster than dandelions—each one touting some specific reason why its model is better than all the others—1910 Genetics is has a more inclusive approach to solving classic problems in pharmacology, and it’s one that should spread to other parts of the life science business.
49 min 42 sec
4.11.23
Tears are a signal of more than just our emotions. The liquid in tears comes from blood plasma, and contains a lot of the same proteins and other biomolecules that circulate in the bloodstream—including those released as a byproduct of the inflammation around tumors. Harry's guests Anna Daily and Omid Moghadam are from a startup called Namida Lab that’s the first company to market a lab test using tears to predict cancer risk. Specifically, Namida’s test assesses the short-term risk that a patient might have breast cancer, as a way of helping them decide how soon to go in for a mammogram. And beyond breast cancer, the company aims to build a whole business around risk assessment and diagnostics, using just the biomarkers in tears. Eventually it could be possible to collect a sample of your tears on a small strip of absorbent paper, send it in to Namida Lab, and find out whether you have colon cancer, pancreatic cancer, prostate cancer, or ovarian cancer. Namida’s big vision, as Moghadam and Daily tell it, is to use tear testing to make precision medicine and diagnostics more accessible and affordable, including to patients who might live far away from tertiary care centers.
42 min 34 sec
3.28.23
It may feel like generative AI technology suddenly burst onto the scene over the last year or two, with the appearance of text-to-image models like Dall-E and Stable Diffusion, or chatbots like ChatGPT that can churn out astonishingly convincing text thanks to the power of large language models. But in fact, the real work on generative AI has been happening in the background, in small increments, for many years. One demonstration of that comes from Insilico Medicine, where co-CEO Alex Zhavoronkov has been writing and talking since 2016 about the power of generative AI algorithms called GANs to help design new drugs. This February, in a milestone moment for the company, the FDA granted orphan drug designation to a small-molecule drug for idiopathic pulmonary fibrosis that Insilico discovered using its own GANs. Zhavoronkov joins Harry to talk about how Insilico got to this point, why he thinks the company will survive the shakeout happening in the biotech industry right now, and how its suite of generative algorithms and other technologies such as robotic wet labs could change the way the pharmaceutical industry operates.
89 min 37 sec
3.14.23
If we knew how to design small-molecule drugs to attach to binding pockets on any given RNA molecule to interrupt or modulate its functions, it could open up a whole new realm of medical treatments. The problem is, if all you know about an RNA molecule is its nucleotide sequence, it’s very hard to predict where those binding pockets might be and what kind of drug might fit into them. As a PhD student at Stanford, Raphael Townshend designed a deep learning model to tackle that problem. Called ARES, the model started with a proposed structure for an RNA molecule with a known nucleotide sequence, and predicted whether that structure would turn out to be correct compared to real-world data. It turned out to be stunningly accurate—and unlike the algorithms behind generative AI models like ChatGPT or DALL-E, it built up its skills based on a tiny data set consisting of just 18 examples of known RNA structures. Now Atomic AI is building on Townshend's original model to create an engine for discovering new small-molecule drugs that could potentially interrupt any disease where RNA is a player.
42 min 42 sec
2.28.23
The medical news publication STAT calls Will Flanary “the Internet’s funniest doctor.” The guests we bring on the show usually talk about how technology is changing healthcare, but Will and his wife Kristin are changing healthcare in a very different way—through comedy. A former standup comic who trained as an ophthalmologist and runs a successful ophthalmology practice in Oregon City, Oregon, Will is better known by his alter ego “Dr. Glaucomflecken.” His short videos have millions of views on YouTube and TikTok, and feature a cast of quirky characters, all played by Will himself, who lightly satirize medical culture and the idiosyncracies of the US healthcare system. And now Will and Kristin have a hybrid comedy and interview podcast called “Knock, Knock, Hi” where they bring on guests who share their own weird and hilarious medical stories.
49 min 48 sec
2.14.23
Within the last couple of years it’s become possible to sequence the entire genome of a newborn baby—all six billion base pairs of DNA—and diagnose potential genetic disorders in about 7 hours. That’s already happening in a handful of hospitals, with a focus on babies who are showing symptoms of rare genetic disorders. But within five years, says Harry's guest Dr. Stephen Kingsmore, it should be possible to extend this rapid whole-genome sequencing to every baby in every hospital, whether they’re showing symptoms or not. Kingsmore is president and CEO of the Institute for Genomic Medicine at Rady Children’s Hospital in San Diego, where he’s been leading an aggressive push to prove that rapid whole-genome sequencing and diagnosis can not only save the lives of newborns, but save the healthcare system a lot of money by making hospital stays shorter and therapies more directed.
43 min 13 sec
1.17.23
Harry's guest this week is Ryan Field, chief technology officer at a Los Angeles startup called Kernel. The company is developing a bicycle-helmet-shaped device that measures neural activity in the brain in real time. The first version, called Kernel Flow, contains more than 50 low-power lasers that beam light through the scalp and skull into the outermost layers of the brain. Hundreds of detectors built in the helmet collect the light that’s scattered back to measure oxygen levels in the brain’s blood supply, which is an indirect measure of neural activity. Field says the company doesn't have specific applications for the technology in mind, but he's betting that researchers and developers will come up with multiple ways to use Kernel Flow to help consumers gauge their state of mind or visualize how their brains are responding to different activities and therapies.
58 min 05 sec
1.3.23
Out of all the dozens of types of cancer that occur in humans, we habitually screen for only five: breast, cervical, colon, prostate, and lung. But what if there were a single test that could detect 50 types of cancer, based on a simple blood draw? That's exactly what's possible today, thanks to the Galleri test, introduced by Grail in 2021. The $949 test, which won breakthrough designation from the FDA in 2019, uses machine learning to assess the patterns of methyl groups—molecules that attach to chromosomes and control gene activity—in free-floating DNA shed by tumors. This week Harry interviews Grail's president, Dr. Josh Ofman, who says that if multi-cancer early detection tests like Galleri are eventually approved for population-level screening, it could help avert 100,000 deaths per year.
63 min 47 sec
12.20.22
Harry's guest this week, Carlos Ciller, started a company called RetinAI whose mission is to help eye doctors, eye surgeons, and scientists studying the eye manage and analyze the data from new kinds of eye imaging, including optical coherence tomography (OCT), fundus photography, and fluorescent angiography. At one level, RetinAI is just doing its part to cure a huge headache the show has revisited many times: the lack of standards and interoperability in the healthcare IT world. They want to make it possible to store and analyze digital images of the eye no matter what technology or device was used to capture it. But once that data is stored in a structured way, it’s possible to use machine learning and other forms of artificial intelligence to sort through image data and identify pathologies or double-check the judgments of human physicians. So RetinAI is developing algorithms that could make it easier to diagnose and treat common conditions like age-related macular degeneration—a form of damage to the retina that causes vision loss in almost 200 million people around the world. Ciller told me he started out his career as a telecom engineer and never thought he’d wind up running a 40-person company that works to help people with vision problems. But at a time when there’s so much new data available to diagnose disease rand identify the best treatments, journey’s like his—from the computer lab to the clinic—are becoming more and more common.
60 min 27 sec
12.6.22
This week January.ai co-founder and CEO Noosheen Hashemi returns to the show after her debut interview in July 2021. January makes a smartphone app that uses machine learning algorithm to learn how a diabetes or pre-diabetic patient's blood glucose levels respond to different foods. After collecting data from a wearable continuous glucose monitor, or CGM, for a few days, the app can start making predictions about a user's future blood glucose levels, even after they stop wearing a CGM. And that can help them make smarter decisions about what, when, or how much to eat, or how much they need to exercise after eating. Harry interviews Hashemi about the company's work to update the app, as well as a second company Hashemi co-founded, Eden's, to make supplements that promote gut microbiome health and—as a results—steadier blood glucose levels.
47 min 13 sec
11.22.22
About half a million babies are born every year through IVF. That number would probably be a lot higher if the procedure were cheaper and more accessible—but making that happen would mean transforming IVF from an artisanal craft into something more like a modern automated factory, with AI helping doctors and technicians make faster and better decisions at every step. And that’s exactly what Harry's guest Mylene Yao, the co-founder of Univfy, is doing. Univfy helps patients with two aspects of the IVF process. The first is using machine learning to provide patients with a more accurate assessment of the odds of success, before they decide whether to invest in one or more IVF cycles, which can cost up to $30,000 per cycle. The second is financing. Univfy works with a bank called Lightstream to provide up to $100,000 in financing for up to three rounds of IVF, with a large refund as part of the deal if the treatments don’t result in a baby. Harry talks with Dr. Yao about the prospects for far broader access to IVF, now that the field is finally adopting more ideas from the worlds of technology and finance.
57 min 19 sec
