Jan 10, 2018, 7:32 PM ETConference Presentation, Keynote

When Biology Moves to Engineering

Evolution is characterized as a "software engineer" that prioritized rapid survival (MVP) over perfection, resulting in inherent "bugs" such as cancer, Alzheimer's, PTSD, and Type 2 diabetes.
These biological imperfections mirror "technical debt" in software engineering, where compromises made to meet deadlines lead to systemic fragility and future maintenance costs.
The current challenge in healthcare is not just treating symptoms but engineering fixes for this accumulated biological debt, similar to resolving the Y2K crisis.
The Y2K analogy illustrates that while the problem (spaghetti code/aging biology) seemed insurmountable, systematic engineering approaches can resolve complex legacy systems without catastrophic failure.

Traditional human analysis is insufficient for biology's complexity; AI is required to interpret data beyond human cognitive limits.
Unlike traditional machine learning, AI (e.g., AlphaGo Zero) can independently learn features and concepts without human labeling or interaction.
AI diagnostics function by processing raw data (e.g., DNA bases) through deep learning layers to identify complex patterns, analogous to identifying facial features in images.
Statistical evidence shows a shift from 50% accuracy in traditional diagnostics (e.g., PSA tests, colorectal cancer blood tests) to 90%+ accuracy with AI-driven solutions (e.g., Freenome).
Cardiomagram demonstrated that software algorithms can achieve 97% accuracy in predicting atrial fibrillation using Apple Watch data, proving that "missing hardware" can be compensated for by superior AI.
The future of medical practice is defined as "Augmented Intelligence," where AI and doctors collaborate rather than AI replacing human physicians.

Ubiquity of AI: The technology has become accessible and easy to deploy, creating a unique historical window for application.
Data Availability: The scale of biological and clinical data has reached a threshold necessary for AI algorithms to function effectively.
Targeted Impact: Current efforts focus on high-mortality issues (cancer, heart disease, diabetes), moving beyond academic problems to solve critical healthcare failures.

Definition of Engineering: The speaker contrasts "discovery" (trial-and-error, high failure rates) with "engineering" (predictable design, high success rates).
Risk Disparity: While a bridge collapsing is a scandal, drug failures in clinical trials are routine; the goal is to treat biology with the predictability of civil or electronic engineering.
Current Limitations: Most academic labs still rely on "pre-industrial" manual methods (pipetting), merely accelerating discovery rather than enabling engineering.
True Engineering Goal: The objective is to design systems that work on the first or second attempt (1 in 10,000 success rate) rather than testing 10,000 variations to find one success.

Cellular Engineering (Cello):
- MIT engineers created "Cello," a tool allowing the design of biochemical circuits using Verilog, the same language used for electronic microprocessors.
- This tool predicts circuit functionality with 90% accuracy, allowing a single design iteration to work 99% of the time.
- Application includes optimizing protein production (current top 10 drugs are protein-based) and advancing CAR-T therapies from primitive "hand-coded" methods to automated design.
Behavioral Engineering (Omada):
- Behavioral diseases like depression and diabetes are complex systems unsuitable for single-target drug cures (unlike antibiotics).
- Omada uses digital therapeutics that can be iterated weekly via A/B testing, a speed impossible for traditional drug development.
- This approach achieves efficacy exceeding metformin while maintaining zero toxicity.
Healthcare System Engineering (Patient Ping):
- The current healthcare system evolved organically with significant "technical debt," leading to high waste and poor coordination.
- Patient Ping provides transparent messaging to coordinate care between payers and providers, discouraging expensive ER visits in favor of cost-effective treatments like physical therapy.
- The generated data creates a network effect capable of supporting new applications for care coordination and waste reduction.

Reframing Healthcare: The speaker argues "healthcare" is a misnomer for "sick care," as current systems only react to illness rather than preventing it.
Aging as an Engineering Problem: Recent science indicates that young blood contains regeneration factors that can reverse age-related phenotypes in older organisms.
The Therapeutic Pathway: Instead of transfusing young blood directly, the focus is on identifying the specific active agents in the blood using machine learning (BioAge).
Impact Potential: Slowing aging by 50% could delay Alzheimer's by 40 years; slowing it by 100% could delay it by 80 years.
Strategic Shift: The overarching trend is moving from "science risk" (uncertain discovery) to "engineering problems" (solvable design challenges) across the entire spectrum of biology.