6 months ago
3
In November 2020, arsenic the satellite battled the COVID-19 pandemic, a antithetic benignant of breakthrough captured planetary attention. Google DeepMind announced that its AlphaFold exemplary had solved the protein-folding problem, 1 of biology’s astir stubborn puzzles. The announcement was hailed arsenic the technological equivalent of a satellite landing. Newsrooms called it a gyration that could bring caller medicines to marketplace faster than ever before.
But fractional a decennary later, the flood of caller cures has not materialised. Despite billions of dollars being invested successful artificial quality (AI), cause find remains a dilatory and costly process. This paradox lies astatine the bosom of what analysts Jack Scannell, Alex Blanckley, Helen Boldon, and Brian Warrington called Eroom’s Law successful a 2012 paper.
Quantity-quality mismatch
When Gordon Moore predicted successful 1965 that computing powerfulness would treble each 2 years portion costs halved, helium captured the astonishing gait of advancement successful electronics — a regularisation that came to beryllium called Moore’s Law. But successful medicine, the other has happened. Eroom’s Law (‘Moore’ spelt backwards) observes that the fig of caller drugs discovered per cardinal dollars spent has been falling steadily for decades.
Today, it costs respective times much to bring a cause to marketplace than it did successful the 1970s, contempt the availability of vastly superior computers, labs, and algorithms. In short, the chips person raced up but the pills person slowed down.
In cause discovery, each caller attraction begins with a hypothesis, an educated thought oregon conjecture astir however a molecule mightiness power disease. For decades, the existent constraint has ne'er been the quantity of hypotheses but the quality. Even earlier the advent of AI, researchers generated millions of plausible ideas, astir of which led nowhere. With today’s AI systems, that fig has grown to billions, yet the prime of hypotheses has not improved. Algorithms tin exponentially summation the quantity of hypotheses but cannot heighten the prime by infusing it with intuition oregon imagination. The leap from quantity to prime remains a distinctly quality privilege.
Creativity and chaos
AI utilizing heavy learning techniques, specified arsenic AlphaFold, thrives connected patterns wherever clear, well-defined relationships are hidden wrong data. The protein-folding occupation suited this perfectly. By 2015, scientists had already mapped implicit 1.5 lakh macromolecule structures done 5 decades of quality effort utilizing X-ray crystallography, fluorescence spectroscopy, and macromolecule atomic magnetic resonance spectroscopy.
There was a known question, a immense dataset, and an thought of what a close reply — each conceptualised by humans — should look like.
AlphaFold’s occurrence was frankincense akin to a superb pupil topping a nationalist entranceway exam, specified arsenic the NEET oregon UPSC. The questions were hard but predictable; the syllabus was immense but good known; and years of quality groundwork had built the coaching material. With capable computational practice, the pupil could execute apical ranks.
Drug discovery, however, is not an examination; it is an enactment of exploration. It resembles a cricket endowment scout trying to spot a aboriginal Virat Kohli successful a dusty colony crushed for his IPL squad oregon a governmental expert attempting to foretell who mightiness go India’s adjacent premier minister. There is nary fixed pattern, nary acceptable syllabus, and nary reliable coaching manual. On the different hand, randomness dominates successful the wilderness successful which cause find operates.
Accidents v. AlphaFold
Penicillin was discovered due to the fact that Alexander Fleming forgot to screen a petri dish. Insulin was discovered done a bid of messy experiments connected dogs, conducted by Frederick Banting and Charles Best, who were simply trying to isolate pancreatic extracts. Paracetamol originated from a 19th-century misidentification successful a laboratory notebook and metformin was studied for the attraction of influenza earlier its relation successful diabetes was understood.
Today’s satellite is besides acold much ethical and careful, rightly so. Every molecule is required to walk done stringent preclinical tests and multi-phase objective trials earlier reaching patients. This caution portion indispensable has besides slowed the travel of discovery. Earlier scientists could trial chaotic ideas with comparative freedom; today’s researchers navigate mountains of paperwork and hazard assessments. So adjacent erstwhile AI proposes a promising molecule, the way to a medicine vessel remains a agelong and arduous marathon.
AlphaFold could win successful cracking a computational situation due to the fact that it was solving a bounded problem: 1 wherever rules existed and quality scientists had already mapped the territory. To beryllium sure, AI volition proceed to reshape assorted aspects of medicine, including screening, objective proceedings design, and cause repurposing. But expecting it to make oregon make caller cures single-handedly is folly. AI excels erstwhile guided by questions that humans already cognize however to inquire and verify, frankincense ensuring its answers are close and reliable. More broadly, AI tin reproduce cognition astatine a faster gait but not ideate oregon make it. So portion it volition proceed to reshape assorted aspects of medicine, including screening, objective proceedings design, and cause repurposing, expecting it to make oregon make caller cures single-handedly would beryllium folly.
As past shows, each large leap successful medicine, from insulin to paracetamol, began with a quality caput consenting to rotation beyond the data.
(Note: AI’s capabilities described present are arsenic of November 2025.)
Dr. C. Aravinda is an world and nationalist wellness physician. The views expressed are personal.
