What Does DeepMind’s AlphaFold Mean for Human Scientists?White Coats at Dawn by Dr Jenny Rohn
In the science fiction series The Expanse, about people living and working in the larger Solar System a few centuries from now, there don’t seem to be any doctors.
When someone gets ill, they are strapped into a chair and their arm is wrapped in a hardware band. The computer screen tells them what’s wrong and the casing delivers the necessary drug.
Despite various artificial incarnations depicted on screen and page over the years, human doctors have still played key roles in many of science fiction’s space-faring communities, doing their best to preserve life and sometimes – “He’s dead, Jim” – failing. Maybe this is because the idea of artificial entities replacing real-life healers seemed so unlikely, even more unlikely than faster-than-light drive or all the other fantastical trappings of the genre.
In recent years, however, it’s become easier to imagine. Artificial Intelligence isn’t science fiction anymore; it’s transforming our society now. AI has crept up on us in so many different guises – facial recognition, music and streaming recommendations, targeted adverts, smart appliances and travel apps, all slowly infiltrating our world behind the scenes and making themselves indispensable. Of course AI isn’t (yet) perfect – as when you buy a car and get deluged for weeks after by adverts for more cars. But one day, it probably well be.
I don’t worry about being made obsolete, or species-existential crises such as the Singularity, but recent advances in science have made me think about what it is to be a (human) scientist, and how that might change in the future. The release of DeepMind’s AlphaFold database of around 350,000 predicted protein structures is one such advance. Its underpinning neural network has attempted to forecast, simply from the DNA code, how the resulting amino acids bend, fold and cluster in space – which in turn can offer an insight into how those proteins might behave in real life. Such clues not only help us understand how our bodies work in health, but can also aid in the design of drugs to deploy when mutated forms of those proteins cause disease.
Like the superfluous purchasing pattern algorithm that bombards new car owners with more cars, AlphaFold is not perfect – yet. Only about 36% of those predicted structures are thought to be precise enough to enable drug design experiments, and the database is still missing a few proteins. The neural network will continue to learn, however, and the database will grow. There is certainly enough to be getting on with, and many scientists are excited at the prospect.
So there is still very much a place for the human being in this particular endeavour – people referring to the database to inform their real-world experiments, and feeding back their physical results as fuel for the network. As partnerships go, it’s pretty satisfying, and still very much human-driven.
But I do have a niggle of wistfulness about where all this is heading. In the past, behind every “solved” protein structure was an individual, labouring in a lab – as much an artist as a scientist. I can see her now, suspending dissolved proteins as hanging droplets, checking each day to see whether any of the globules had developed a suitable protein crystal. Every day for weeks, the breathless check, the small crash of disappointment – a sensation so common it surprises her that she still feels it, every damned time. And then, maybe if she’s lucky, the flash of triumph that will later propel her down to the pub with her lab mates to celebrate: a tiny flower materializing in one of the droplets, like a shard of ice under the microscope.
She lovingly flash-freezes the one lucky crystal; later she will keep vigil at a faraway synchrotron particle collider, drinking bad coffee at 3 AM as her precious sample is bombarded with X-rays. She will pour over the resulting patterns, run it through databases, try to divine its form and purpose at the atomic level. And despite all this work, the crystal –if it ever grew at all – might still refuse to give up its secrets. The world would remain in the dark about how this protein does its thing. She might carry on, or eventually give up and move on to something more tractable – others might take it up the mantle after her, re-treading the same ground with renewed hope and a reprise of failure.
Surely making such scenes a thing of the past would be a good thing. All of that waste, all of that emotional energy, all of that hard work and heartache, replaced by something more positive and productive. Science would advance more quickly than ever before, freeing up scientists to think bigger and bolder.
The strange thing, though, is that I cannot picture the scientific process without all of its human trappings: the graft, the physical manipulations, the late nights, the crushing failure. And arching above all, that vast, almost celestial feeling of the unknown: that there are things that will always be unknown and mysterious, that we will never be able to understand no matter how hard we try. Pulling away the petals until nothing remains but dust.
But what if our future is now going to be different? Where will we be when everything is solved, outsourced to our own creations? This failure in understanding, this abject absence of enlightenment, will no longer be the spark that drives us forward to try again, to try different, to try better, like humans have done since they first winked into being from a dark universe of unimaginable mystery. How much of our humanity is tangled up in that, and what will be left when it is gone?
Dr Jenny Rohn is a practising cell biologist at University College London. She is also a writer, broadcaster and novelist. She has written three novels, the latest being Cat Zero, released in mid 2018. She was a regular contributor to the Occam’s Corner blog at The Guardian and was a co-founder of the Science is Vital group. She is on Twitter at @JennyRohn.