DeepMind is on a mission to change everything and solve life’s biggest problems. Google’s deep-learning system began its journey by beating human opponents at old Atari games before progressing to more complex classics such as Starcraft II, Chess, and Go. Once the technology proved it could outwit human opponents, it was time to focus on real-world problems.
Traditionally, medical advances were the result of chance discoveries. The most obvious example was when Alexander Fleming left a petri dish of bacteria in his lab before embarking on a two-week holiday. On his return, he discovered that the dish had been contaminated by a fungus that produced an antibacterial substance. This accidental incident led to penicillin, which went on to save millions of lives.
Here in 2020, it was DeepMind’s AI program called AlphaFold that unveiled a solution to the 50-year-old biological challenge of predicting the shape of proteins. Finding biology’s “holy grail” and deciphering the complex protein processes is expected to empower scientists to better understand the machinery of life and unlock opportunities for drug discovery.
How a love of gaming could change the world
Demis Hassabis, the co-founder of DeepMind, told the BBC that his love of video games inspired the scientific breakthrough. Foldit, an online puzzle video game, challenged gamers with folding protein into a particular shape. The game led to the discovery of important structures for real proteins. That was the moment where videogames collided with science.
Foldit revealed how a group of gamers had trained their intuition and pattern-matching capabilities to do something that brute-force computer systems couldn’t do at the time by coming up with the right shapes. This was the moment that inspired Hassabis to question if AI could mimic the gamers’ same intuitive capability.
In another interview, Hassabis shared how they were using AI to solve intelligence, and then using that to solve everything else. The latest discovery suggests that DeepMind is beginning to play a pivotal role in decoding the language of life. However, the lack of a peer-reviewed paper prompted suspicion and criticism from academics.
If we dare to look beyond the headlines, it quickly becomes apparent that we are still in the early stages and only just beginning to understand the art of the possible. The journey started with exploring the real-world impact technology can have on understanding diseases to improve drug discovery. But at what cost?
The billion-dollar cost of revolutionizing biology
The London-based AI lab was famously acquired in 2014 by Google parent Alphabet for $600 million. Although the company is beginning to deliver ground-breaking results, it has been haemorrhaging money for years.
DeepMind made a $649 million loss last year, and Alphabet has also waived a total $1.5 billion debt.
The recent events could be a bittersweet pill for the UK government to swallow. It has researched and developed something that has gone on to revolutionize biology. But they also appear naive when it comes to capitalizing on it. Selling DeepMind to the highest bidder will come back to haunt them in the future.
University researchers set out to decode the language of life for public good. They built the project on government-funded insights and good intentions. But how long until big tech uses the findings to form an alliance with big pharma in pursuit of profit? Ironically, university researchers could find themselves again using government cash, but this time to pay DeepMind to access a system it helped build.
The risk of de-democratizing knowledge production
Looking at the bigger picture, it’s hard not to predict that DeepMind’s research and product development is destined for great things. Who will profit from the tens of billions that result from the studying of protein folding? There is a danger that the company could drift away from its original mission of public good.
As AI research gathers pace, there’s also increasing concern in the emergence of a compute divide between large firms and non-elite universities.
If only those with access to supercomputing power have the necessary capabilities, many believe we are running the risk of de-democratizing knowledge production.
Recent reports from DeepMind researchers claiming that neural networks can outperform neuro-symbolic models are certain to further divide the community. The research is sure to have implications on the development of machines that will have the ability to reason about their experiences. It’s undoubtedly an exciting time for the industry, but it also feels like we are at a crossroads, and the next move forward will be critical.
For over 50 years, scientists questioned how proteins knew what shape to fold themselves into. DeepMind’s ground-breaking achievement has finally answered the question and is on the right path to revolutionizing medicine. But it has also raised a few more about the long-term impacts for society and if AI will empower scientists or replace them. But those are questions for another day.
