Jan 11, 2018 · 23m · a16z
When Biology Moves to Engineering
0:00 / 0:00
▶ Watch on YouTube →
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
Vijay Pande of Andreessen Horowitz explains how biology is transitioning from an empirical, trial-and-error science into a predictable engineering discipline. By applying artificial intelligence, software design principles, and systematic engineering across cellular, behavioral, systemic, and longevity scales, modern medicine can effectively resolve biological technical debt and dramatically improve human health.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. How this is scored →
The host as informed peer 0.0 Guest teaching 6.7 Guest disagreement 1.2 The host pushing back 0.0
speaking balance: gold is the host, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 21:25 Rejecting the 'Blood Boy' distribution model
Vijay forcefully rejects the idea of scaling young blood transfers as 'a very bad idea', advocating instead for machine learning analysis to isolate active molecules.
Hardest push from the host ▶ 0:11 No host pushback (Monologue talk)Because this episode consists of a monologue keynote presentation, host Steph Smith does not speak or offer pushback at any point.
Biggest teaching moment ▶ 14:15 Explaining Cello software and EDA circuit toolsVijay educates the audience on how bioengineers at MIT adapted electronic design automation and Verilog code to reliably engineer cellular circuits.
The host holds their own ▶ 0:11 No host interventionThe host does not speak during the recorded presentation, resulting in zero instances of host expertise or pushback.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | The host as informed peer | Guest teaching | Guest disagreement | The host pushing back | Why |
|---|---|---|---|---|---|---|
| Evolution's Features and Bugs | 0 | 6 | 1 | 0 | Vijay delivers a solo keynote presentation framing evolutionary biology through software concepts like technical debt and Y2K. Because this is a monologue, the host is absent and host-side metrics remain at zero. | |
| Biological Circuit Complexity and Human Limits | 0 | 7 | 1 | 0 | Vijay explains biological circuit complexity, arguing it exceeds human comprehension and necessitates AI systems like AlphaGo Zero to identify patterns independently. Host metrics remain at zero in this continuous monologue. | |
| Prediction Accuracy: AI vs. Traditional Diagnostics | 0 | 7 | 2 | 0 | Vijay contrasts traditional 50% diagnostic accuracy with 90%+ AI-driven accuracy from companies like Freenome and Cardiogram. He explicitly rejects the 'doctors vs. computers' premise as a false dichotomy. | |
| Three Confluent Trends Driving the AI Biology Shift | 0 | 6 | 2 | 0 | Vijay uses a humorous bridge-versus-drug metaphor to contrast traditional empirical discovery with true engineering. The host remains unengaged in this presentation segment. | |
| Engineering Cellular Circuits with Software | 0 | 8 | 1 | 0 | Vijay educates listeners on MIT's Cello software, showing how Verilog programming can be applied to design biochemical circuits with high predictive accuracy. Host scores are zero due to the presentation format. | |
| Engineering Behavioral Therapies for Chronic Disease | 0 | 7 | 1 | 0 | Vijay details Omada's behavioral therapeutic platform, comparing weekly digital iterations to search engine A/B testing. Host activity remains non-existent. | |
| Engineering Healthcare Systems to Reduce Waste | 0 | 6 | 1 | 0 | Vijay discusses how Patient Ping coordinates healthcare logistics to reduce expensive emergency room waste. The talk continues uninterrupted without host participation. | |
| Engineering Longevity and Aging Science | 0 | 8 | 2 | 0 | Vijay explains young blood plasma research in aging and strongly dismisses the literal 'Blood Boy' distribution model as a bad idea compared to AI biomarker discovery. | |
| Summary: Engineering Biology Across All Scales | 0 | 5 | 0 | 0 | Vijay concludes his presentation by summarizing the shift from discovery science risk to engineering solutions in healthcare. |
Statements from this episode (24)
Pande: Biology can now be engineered like traditional disciplines
“Opportunities where we now finally can engineer biology the way we engineer other areas.”
Pande: Evolution acts as a software engineer, creating features and bugs
“Evolution has created a ton of features. It's an amazing software engineer of sorts. But, you know, with these features come bugs.”
Pande: Many simple organisms possess cellular mechanisms that prevent cancer
“A lot of simple organisms just don't get cancer. They have mechanisms to fight that.”
Pande: Biology operates like software development burdened by technical debt
“And if you think about it, biology is all about technical debt. Biology is trying to get that MVP out. You're trying to, like, survive the T-Rex coming after you. You know, you're not gonna be able to make things perfect.”
Pande: Bioengineering requires deciphering complex code similar to resolving Y2K
“I think you might be seeing where I'm going with this, is that that's a lot like what we have to do in biology. You know, we have to get into the code which is a mess and understand it, and just understanding the code of biology is something extremely difficul…”
Pande: Biological complexity exceeds the limits of unassisted human understanding
“And when you talk to a biologist, and I think what we're starting to realize is that there's a fundamental conclusion that comes from this, is that biology is so complicated that it's probably beyond what the human being can understand.”
Pande: Identifying cancer with AI is functionally similar to facial recognition
“Well, it turns out that identification is very similar to identifying someone in a picture.”
Pande: AI genomic analysis will outperform humans in cancer detection
“So the hypothesis here and the opportunity is that this is something that we could do considerably with AI considerably more accurate than people.”
Pande: Traditional diagnostic tools like PSA tests are only 50% accurate
“PSA test has 50% accuracy. Identifying colorectal cancer from blood, also about 50% accurate.”
Pande: AI cancer detection startup Freenome achieves over 90% accuracy
“Companies like Freenome that look for the signals for cancer from blood can do considerably better. You know, 90% and plus.”
Pande: Cardiogram predicts atrial fibrillation via wearables with 97% accuracy
“So a cardiogram can take wearable information and predict atrial fibrillation with 97% accuracy.”
Pande: Wearable diagnostics require AI software and datasets, not new hardware
“They've shown that the missing piece is not more hardware, but is software. The missing piece is artificial intelligence and gold standard data sets that can take what the Apple Watch can do, And make high quality predictions.”
Pande: Computers outperform doctors in dermatology and ophthalmology diagnostic predictions
“It's clear that in many areas, we've seen in dermatology, ophthalmology, many different areas, that computers can be more accurate than doctors in predicting.”
Pande: AI will augment rather than replace human physicians
“It's never going to be computers versus doctors. It's a false dichotomy. It's a false comparison. It's always going to be doctors and computers. And naturally, they would be better than computers or doctors alone. And I think this is the future of healthcare.”
Pande: Billion-dollar drug candidates fail clinical trials almost every week
“Compare that to drugs, like a billion dollar drug fails clinical trials, I mean, that's like every week. That, I think, really hits home the difference between engineering and discovery science risk.”
Pande: Lab robotics accelerate trial-and-error but do not constitute true engineering
“And the big, big advance was robots, and so robots really is just not engineering, it's just faster people. You can get to your thousand or 10,000 chances just faster.”
Pande: MIT's Cello designs biochemical circuits in Verilog with 90% accuracy
“Bioengineers from MIT have written a software called Cello, which allows them to do exactly the same thing for biochemical circuits. You first code the circuit, in this case, in the very same language, in Verilog, the same language you'd use to code electronic…”
Pande: Seven of the top 10 selling drugs are cellular protein drugs
“Seven out of the top 10 drugs right now that are sold are protein drugs made by cells.”
Pande: Omada's digital therapy demonstrates greater diabetes efficacy than metformin
“AMADA has a behavioral therapy, which is in a sense similar to areas like cognitive behavioral therapy, or CBT, derived from the CDC's Diabetes Prevention Program. And AMADA can demonstrate efficacy that actually exceeds that of metformin.”
Pande: Digital therapeutics optimize faster than drugs via weekly A/B testing
“But for a digital therapeutic like this, you can iterate week after week doing A-B testing, which in the medical world is really a randomized clinical trial that you get to run every week. You make it better and better and better and engineer it and improve it…”
Pande: Modern healthcare is reactively treating illness rather than maintaining health
“Healthcare is not healthcare. Healthcare is sick care.”
Pande: Leading Alzheimer's therapeutics only delay progression by six to nine months
“For something like Alzheimer's disease, the leading drug delays Alzheimer's disease by like half a year or three-quarter of a year.”
Pande: BioAge uses machine learning on young blood to develop anti-aging therapeutics
“And so if you think about the game plan that Freenome runs for taking blood and understanding where there's cancer and what's the cancerous agents in there, what BioAge is doing is doing essentially the same thing, looking at the blood of young, using machine …”
Pande: Machine learning will convert biological science risks into predictable engineering problems
“I think what we expect to see is the shift from many areas that used to be areas of science risk through engineering mechanisms and machine learning now become really engineering problems. With that in mind, we can finally understand and tackle the technical d…”