Imagine curing every disease — not in the next century, but in the next seven years. Cancer, neuromuscular diseases, maybe even male pattern baldness. That includes aging, to the extent reversing, stopping, or slowing aging is physically possible. It can be done.
Drug discovery and disease treatment generally are in the dark ages. Discovery usually begins with molecular screening, which is done by computers. Computers, well that’s advanced right? No, not when you consider that most of the molecules chosen end up being expensive dead ends. Which means that presumably a lot of the molecules discarded might have been promising.
Then you go to pre-clinical, the classic Petri dish or cell cultures vessels. Then various cheap surrogates like worms or fruit flies, then Mickey Mouse or Rizzo the Rat, and finally human beings usually divided into three phases of clinical trials.
Ultimately, bringing a single drug to market averages about $2.6 billion, spent over a period of 10 to 15 years. About 75% of total R&D costs come from failures, as only around 11% of drug candidates progress from discovery to clinical trials, and only about 1% ultimately reach the market. These failures ultimately contribute to scientific knowledge but meanwhile are lost expenditures.
There’s got to be a better way.
There will be. Call it “The Nuclear Option.”
In 1992, the U.S. ceased all actual nuclear weapons testing, instead shifting to computer simulations. And… it didn’t work out so well at first. Despite occupying whole buildings, super computers back then just didn’t have the oomph. An iPhone 14 Pro Max, much less the forthcoming 17 models you may be drooling over, has more raw on-board processing power than 1995’s top 500 supercomputers in the world combined. But as super computers have become more super, nuke testing has become more reliable.
When we have sufficient power, we need to do the same with the human body. Insert a human genome into supercomputers then insert the “insult” (cancer, an infection, elements causing old age), then insert treatments or preventions.
Initially, it might be a generic genome. Humans appear to be about 99.9% genetically identical. Problem is, that still allows for about 6 million genetic differences between two individuals considering the entire diploid genome of about 6 billion bases. So ideally, we would eventually want treatments of individual genomes. You provide a DNA sample and the computer spits out a treatment.
In any case, we will need an absolutely massive increase in computational power (along with concomitant increases in memory and other aspects.) But the good news is we’re getting it.
We are no longer slaves to “Moore’s Law,” actually just an observation that twice as many transistors were being packed onto the same space every 18 months. That was later changed to every 24. That roughly converted to advances in computing power. Conversely we are now seeing AI supercomputer performance double every 9 months, fueled both by increasing the number of AI chips and improving chip efficiency. At that pace, the fastest AI supercomputers would be about 645 times faster in seven years than today. Add just nine months and it’s almost 1,400 times faster. Add nine months and … right.
We can’t even begin to imagine such power. Thus, the science community has gawked over the speed increase of the latest top supercomputer, El Capitan, over its predecessor, Frontier. El Capitan came online in January of this year with a peak performance of 2.79 quintillion calculations per second compared to 2.05 for Frontier in 2022. That’s merely a 45% increase in 30 months. Not much to write home about.
“We can’t contemplate 645 times faster! as Jack Nicholson’s Colonel Jessup might have shouted in A Few Good Men.
But wait, there’s more!
Today’s supercomputers don’t take advantages of two developments very much in the works. One is photonic chips. Those transmit signals at the speed of light, 1,000 times faster than today’s electronic chips. They’re also vastly more resource efficient. Nobody doubts photonics will go mainstream soon, though bottlenecks will initially favor use of hybrid electronic-photonic chips. Still, whatever speed we get within seven years will multiply the speed of our computers by that much.
Then there’s the favorite of dreamers everywhere, quantum computers. Just a few years ago you could read that they were untenable because of their error-making propensity. But tremendous progress has been made towards reducing that in many ways. They’ve now gone from “maybe” to “when” will they be fully commercialized.
Classical binary computers process information using bits that can only be in one of two states, 0 or 1. In contrast, quantum computers use qubits that can represent both states at the same time, allowing them to be exponentially faster for certain complex problems.
The global quantum computing market size was estimated at 1.42 billion in 2024 and is projected to reach 4.24 billion by 2030, which is peanuts except that it means they’re already at work. As with photonic chips, the earlier versions will be hybrids.
In pure form, how fast can they calculate? There have been claims of quantum chips a quadrillion times faster than whole supercomputers. Problem is, inevitably those making the claims have used equations favoring quantum computing with no real-world application.
Until we have quantum computers in pure form, we won’t know exactly what their capabilities are. That said, drug discovery “stands as an ideal candidate” to benefit from these new computers, as Scientific Reports recently stated. That’s in part because, as with decryption, there’s no substitute for the brute force of being able to do so many calculations in so little time.
None of this is to say that sheer computing power is all that’s needed. Thus, we must learn much more about protein folding. But there again, processing power has proved vital such that companies led by Google’s AlphaFold have made incredible strides.
Other areas? Waaaay back in 2014, the Beagle supercomputer at Argonne National Laboratory could analyze 240 full human genomes in about 50 hours by parallelizing the workflow of genome sequencing data processing, including cleaning, alignment, and variant calling. The focus was on whole human genome sequencing.
But now concentrating on what really counts, genetic variants, the laboratory has announced a 100X speedup of genetic analysis analyzing over 2,000 traits in a large, diverse cohort from over 2069 participants.
Currently the leader of the pack in generating entire synthetic genomes and chromosomes, including bacterial genomes, mitochondrial genomes, and yeast chromosomes, is EVO 2. It’s a massive open-source AI model from Arc Institute, Stanford, and NVIDIA, trained on trillions of genomic data points to understand, predict, and design biological sequences.
It’s quite impressive, but again limited by today’s computing power. (It taps into the cloud so the power varies.)
This won’t be a sudden quantum leap situation of nothing now, everything in seven years. Already AI is improving the process of choosing the best molecules for testing. AI-developed drug candidates have shown higher clinical trial success rates (80-90% in Phase I clinical trials) compared to traditional methods (around 40%), with an increasing number of AI-designed drugs entering clinical stages annually. It’s just been announced that the first fully AI-generated drug – meaning AI created both the target and the molecular design – is entering clinical trials in the U.S.
And if the seven years left of my original (unpublished) 10-year prediction is off, it won’t be like claims of massive melting of the Arctic ice sheet that stem back to at least 1968 and keep getting renewed or essentially every prediction Paul Ehrich ever made including what time lunch would be served. It will be just a bit off.
Further, while so far as I know I’m the first to propose specifically eliminating disease in this manner, recently Sir Demis Hassabis, co-founder of Google DeepMind said all disease could be cured by AI within a decade. Essentially the same time frame. If I’m a kook, then so is he. Except he’s rich, so he would be called “eccentric.”
Far from something your grandkids or kids will experience, this is for you.
