✨⚕️ Must AI cure cancer?
Miracle drugs are important. So is reversing the long-term decline in drug discovery productivity
My fellow pro-growth/progress/abundance Up Wingers in the USA and around the world:
Another week, another dystopian scenario revolving around advances in artificial intelligence.
As I have written several times recently,12 I have deep skepticism about job-pocalypse scenarios that ignore a) new job creation, b) the difference between AI capabilities in the lab versus AI adoption in the workplace, and c) all the numerous political economy constraints potentially preventing unprecedented employment disruption.
Alas, my skepticism doesn’t make for a particularly compelling sci-fi story that appeals to a society that’s been marinating in techno-catastrophism for more than a half century.
But what about the AI upsides? Don’t they make a compelling story? What about a world made cancer-free due to AI accelerating drug discovery and enabling near-miraculous breakthroughs?
Indeed, that’s an Up Wing scenario frequently mentioned by AI enthusiasts, including AI company CEOs, to paint a picture of a world that justifies nearly any and all downsides—not to mention continued massive AI infrastructure investment.
The techno-war on cancer
It’s hardly a pie in the sky notion. The optimist case rests on how AI affects speed, scale, and precision. Cancer research generates oceans of genomic, clinical, and imaging data that no human team can fully integrate. New AI systems, however, can rapidly detect hidden patterns, predict risk, and match patients to highly personalized therapies.
And, yes, they could also compress drug discovery timelines by screening billions of molecules and identifying promising targets far earlier. This reduces the costly trial-and-error that defines cancer research today. The dream: cancer as a preventable and totally treatable condition.
That said, curing cancer tout de suite is quite the metric for judging AI success or failure, or whether AI is worth various aspects of inevitable societal disruption.
Reversing Eroom’s Law
Let’s think more broadly about AI and medicine. For decades, the pharmaceutical industry has suffered the opposite of Moore’s Law. Even as computing grew cheaper and faster, drug discovery grew slower and more expensive. The cost of developing a new medicine has roughly doubled every nine years, while the number of drugs produced per research dollar has steadily declined, a productivity trend nicknamed “Eroom’s Law.”
This is where AI is already beginning to matter. While conjuring miracle cures overnight would be awesome, improving the economics of the discovery process—reducing errors, accelerating decisions, and cutting wasteful spending—is no small feat. Such seemingly incremental improvements strike directly at the forces that have been driving Eroom’s Law.
A recent Goldman Sachs report, “Byte-ology: Quantifying AI’s value creation in drug development,” illustrates the mechanism at play. Here’s your trouble: Drug development functions like an extraordinarily expensive funnel. It typically costs more than $2 billion and takes roughly 10 to 15 years, with fewer than 10 percent of candidates that enter preclinical testing ultimately reaching approval.
Much of the industry’s declining productivity reflects how much money is spent testing molecules that ultimately go nowhere.
AI appears to be finally changing that equation. Goldman’s analysis of roughly 100 drug candidates developed3 using AI finds an overall success rate of about 10 percent, compared with roughly six percent historically—an increase of nearly 60 percent. (In drug development, that “success rate” refers to the share of candidates that survive each stage of testing without failing.) The biggest gains show up early in the pipeline, where most drugs collapse. AI-developed candidates perform better in Phase 1 and Phase 2 trials because machine-learning systems help scientists design and identify molecules more likely to behave safely and work as intended before expensive human testing begins.
In effect, AI acts as a better front-end filter that helps scientists avoid costly dead-ends. And with hundreds of experimental drugs entering development each year, even modest improvements in the odds that any one candidate survives testing can add up quickly across the entire pipeline.
A few other interesting take-aways:
AI could reduce time to market by roughly 20 to 25 percent, shortening the process by about three years overall. Much of this acceleration comes from compressing the earliest discovery phase by roughly 1.5 years and speeding clinical testing through faster patient recruitment and data analysis.
By improving candidate selection and streamlining trial operations, AI could lower total development costs by roughly 25 to 30 percent, including estimated reductions of roughly 28 percent in capital costs required to advance drugs through clinical stages.
The combined effects of higher success rates, faster timelines, and lower costs could create between roughly $80 billion and $400 billion in industry value over the next decade.
From the report:
AI-driven acceleration of drug discoveries — both via a higher success rate and a swifter process — has potential to unlock significant value for the sector where there are increased debates on returns from R&D. We also see higher/faster drug discoveries playing a prominent role in debates around the broader corporate and social value of AI relative to costs.
About that last bit: The report acknowledges considerable investor interest in the value of “AI solutions” given both the tremendous hyperscaler investment and concerns about AI power demand. (“We see the growth in power consumption from data centers by the end of the decade will represent the equivalent of adding another top-10 power consuming country.”)
More:
While there is recognition that AI can provide personal and corporate efficiency solutions, as spending continues to rise we have received increasing questions on larger societal-impactive solutions. We have highlighted five (not meant as a complete list of solutions):
1. Healthcare — Accelerating discovery and more efficient care
2. Agriculture — Improving yields and reducing waste
3. Energy — Optimization and efficiency in power generation/physical assets and consumption efficiency
4. Human Capital — Reskilling and Upskilling
5. Education — Interactivity and Personalization
Among these, drug discovery/process potential comes up the most in investor discussions.
Yes, it’s tempting to see all this talk of AI curing cancer—especially from Silicon Valley—as a response to the technology’s mounting political backlash. Polls show rising public unease, data-center projects increasingly provoke local protests, and social media teems with essays warning of looming catastrophe. A sudden medical miracle would no doubt quiet much of the noise. But so should a steady reversal of the industry’s long slide in productivity.
If AI can help restore the lost pattern of falling costs and rising research efficiency in the drug sector—counteracting the forces behind Eroom’s Law—it would mark a profound shift worth celebrating.
AI played a central role in discovering, designing, or selecting the drug candidate before clinical testing. “AI mode of discovery, which we broadly characterized as: (1) repurposing (i.e., using AI to link an existing drug to a new indication); (2) design and optimization (e.g., using AI to design better chemistry against well-known biologic targets); and (3) target discovery (AI to identify novel or relatively unknown/unaddressed targets).”
On sale everywhere The Conservative Futurist: How To Create the Sci-Fi World We Were Promised
