Human curiosity as AI naturalist guide
Perplexing observations and productive holes in science
Last week I learned marijuana use has been consistently associated with increased lung capacity, a surprising phenomenon for which we have no explanations. After all, you’d never imagine smoke exposure to give you bigger lungs. [This doesn’t mean it’s beneficial for you - Please read the article]
It’s a fascinating puzzle that’s been known for decades, and observed in almost every population in which it’s been measured. But when I tried to ask an AI about the riddle of marijuana use and vital capacity, I got nothing…
It looked right past the unexpected narrative violation, and offered me the respiratory symptoms commonly resulting from marijuana smoking.
We’re becoming collectively enthusiastic that AI will accelerate scientific discovery and deliver prodigious, world-changing insights. But so far its contributions feel like the results of a tireless laborer. As Benedict Evans puts it, it’s doing the work of “one intern with infinite patience.”
A more apt analogy is the painstaking work of the naturalists of the Enlightenment and Victorian eras, who made tens of thousands of meticulous observations of details of plant and animal life.
AI has already identified phenomena we’ve overlooked or could not see. For example, after studying 84,743 scans, a new machine learning model can tell male from female retinas. Previously, no one knew there was a difference.
As Raymond Joseph Teller, of the magician duo Penn & Teller, explains, "Sometimes magic is just someone spending more time on something than anyone else might reasonably expect."
I’m sure there are and will be many more such cases, in drug discovery and beyond.
But the marijuana story reminds me that humans, too, have powerful abilities to recognize patterns. And we’re even more uniquely tuned to spot anomalies, and become inquisitive in moments when our expectations break. Don’t bet against us realizing when there’s something to learn by exploring the question, “What is going on here?” We’re especially good at noticing differences over longer periods of time in how things unfold compared to how we expected.
In respiratory alone, we’ve followed our observations to some remarkably productive theories. Evidence that people in Bavaria who, as children, lived in proximity to cows had lower rates of asthma gave us the hygiene hypothesis and opened new lines of work in immunology. Another theory about the origins of asthma arose after we realized that young migrants to the US were surprisingly protected from the disease. More generally, physician-epidemiologist David Barker noticed that cohorts of individuals who’d lived through collective tragedies (eg, famines, pandemics) decades earlier, developed higher rates of disease later in life.
We already ask remarkable questions, and have for a long time. For many years, the Edge Foundation ran an annual project where leading thinkers responded to provocative questions, or shared the most important puzzles in their fields. Wikipedia has a page full of unanswered questions/unsolved problems across scientific domains. So many big conundrums, and more by the day.
Just this week, I’ve run into reviews that highlight areas of curious uncertainty and significant potential impact in respiratory disease:
A recent issue of the European Respiratory Journal reviewed the questions at the center of our early, emerging understanding of the airway epithelium, its role in respiratory health and disease, and its potential as a target of prevention and treatment. For example http://dx.doi.org/10.1183/13993003.01397-2023
Somewhere between 70-90 percent of those with COPD have never been diagnosed, resulting in missed opportunities to start lung-preserving treatments. A new paper summarizes the challenging tradeoffs in screening and case finding what is already the world’s third biggest killer
An American Thoracic Society expert panel review distressingly reported that after all the work that has gone into developing and testing interventions to reduce indoor air pollution in low- and middle-income countries, "it remains unclear which household energy interventions reduce exposure, improve health, can be scaled, and are sustainable."
According to data from the U.S. Centers for Disease Control and Prevention, American Indian and Alaska Native people continue to suffer a higher prevalence of asthma than any other specific racial or ethnic group in the United States. The Indian Health Service has announced a Strategic Initiative - Asthma Control in Tribal communities (ACT). We need better ideas for how to control asthma and reduce the burden of asthma-related illness and death among American Indian and Alaska Native people.
In each case, we’ve identified fundamental challenges or perplexing observations, synthesized from years of curated science, and need help comprehending the situation we find ourselves in.
Ahead lie startling insights, new avenues for further research, and provocative breakthroughs. The trick is in figuring out how to use the exhaustive naturalists we now have at hand to explore new unseen angles on these challenges.
More, recent writing
I wrote a post on LinkedIn about the current market’s benign neglect of ambitious medical device startups.
And another one on LinkedIn about the startups working on diagnostic and therapeutics tool for earlier detection and better treatment of lung cancer.
Other stuff
💥Incredible hiring opportunity → One of my favorite former Propeller colleagues — our head of dev ops, a technical monster, and an ace leader and pillar of our team — is becoming available and looking for his next role. Please contact me for details.
I’m helping a founder with an ophthalmic drug delivery business, about which I know very little. I’d appreciate any suggestions/introductions to people with relevant expertise I might speak with to learn more
I’ll be in London the week of May 20th and would love recommendations about companies I should meet, best places to eat, hang out, etc.
