On Monday, British machine learning startup DeepMind announced that it had solved the 50-year "protein folding" problem.
Scientists rejoiced at the promise of improved drugs for incurable diseases. It was a historical breakthrough which proved artificial intelligence is more than just hype.
Understanding how a protein folds allows scientists to predict what 3D shape a protein will take, judging by its amino-acid sequence. In just weeks, using a relatively modest amount of computing power in lab situations, AlphaFold claimed to have predicted the shape of a sequence that scientists had spent a decade grappling with.
The breakthrough is is a crucial step to making drugs like cancer or Alzheimer’s treatments or vaccines. In fact, earlier this year AlphaFold did predict several protein structures of the Covid-19 virus.
But it is difficult to gauge how useful this technology, which DeepMind has yet to release the scientific paper for, will be in practice. How can it be applied more generally in a way that could assist with labs around the world? Who will benefit from this? Can it be recreated outside of DeepMind’s research hub?
A 3D rendering of a protein cell
The other unanswered question relates to the looming presence of DeepMind’s parent company. Google bought DeepMind, founded by Britons Demis Hassabis and Mustafa Suleyman in 2014 for a reported £400m with cash generated by its digital advertising operation.
This led to much grumbling about selling a British gem to an American company. Later, changes in directors and the absorption of its health research unit into Google proper raised eyebrows with privacy experts who feared its work in the National Health System may be compromised through Streams, a kidney dialysis app.
In response to the criticism that DeepMind was moving its health data to Google in 2018, Hassabis claimed Google would "give us the platform to more rapidly bring our technologies to the wider world at scale".
Google’s millions have allowed DeepMind to hire the best computer science graduates from the UK, Europe and the US. Founding investor Humayun Sheikh recently told CNBC that it would have “probably failed” if it had not been bought.
Whether Google can make money out of the technology remains to be seen. Google’s cloud computing unit is seeking to attract healthcare companies. AlphaFold’s system and the necessary compute power could be packaged up as a viable product, monetising the breakthrough. DeepMind does not allude to this in its blog post, simply leaving an email address with a call to scientists who wish to get involved in its research, along with a vague statement of its mission to explore "how best to provide broader access to the system in a scalable way".
DeepMind is not making money. It lost $154m in 2016, $341m in 2017 and $572m in 2018, according to the most recent accounts.
Yet DeepMind has proven valuable — its systems have improved cooling systems for Google’s massive datacentres and its technology can be found in the Android operating system to help reduce battery drain. It has contracts with NHS departments.
Google co-founder Sergey Brin is reported to have backed DeepMind even when other departments questioned its ability to return all-important shareholder value.
However, there is pressure from inside Google for DeepMind to be more than just a well-funded science project. Outsiders may also fear that further world-changing breakthroughs may be halted if they fail to show a clear path for profit.
Свежие комментарии