For the last 50 years, the “protein folding problem” has puzzled researchers seeking to untangle the mysteries of human biology. Proteins play a critical role in determining how cells function in the body, which impacts everything from growth to development to the regulation of the body’s tissues and organs. Each protein’s specific function is dependent on its three-dimensional, ribbon-like structure, which is complex because it folds itself up with an astronomical number of possible configurations. A researcher in 1969 estimated it would take longer than the entire known universe to discover all the combinations. The current process of predicting protein structures uses trial and error, often taking years and requiring expensive, specialist equipment to ascertain the shape of a protein. Recent advancements in solving the protein folding problem are a salient example of why investing in AI research and development is so critically important.
In December, the Critical Assessment of protein Structure Prediction (CASP) announced that DeepMind’s AlphaFold had used artificial intelligence to make major strides in accurately predicting protein structures in a matter of days. AlphaFold was trained using deep learning techniques on 170,000 known protein sequences and structures, available from a public protein data bank. According to DeepMind’s press release, this enabled the algorithm to translate the structures into data and use that data to model target protein shapes and predict how accurate they were. DeepMind stated how this “demonstrates the impact AI can have on scientific discovery and its potential to dramatically accelerate progress in some of the most fundamental fields that explain and shape our world.”
There is reason to be cautious, as the DeepMind researchers are yet to submit a peer-reviewed paper, which is key to ensuring validity and giving the findings external credibility. Further, the limitations of the technology are still unclear and the protein folding problem is only one part of the process. For example, while this method could accelerate vaccine development, large clinical trials that prove its efficacy will still take time. Going forward, the scientists will need to collaborate with other multi-disciplinary researchers on how to use this technology to develop new medicines, or where it could be useful in other fields.
If these claims prove to be true, however, this is a significant step in understanding the mysteries of the human body. The technology has the potential to accelerate drug and vaccine development, which could help mitigate against future pandemics, aid researchers to better understand diseases, and, according to the DeepMind team, even assist in environmental sustainability solutions.
DeepMind’s breakthrough appears to be the latest example of AI as a meta technology. Meta technologies are technologies or inventions that have the capacity to aid new discoveries or spur innovation in other areas. Just as the microscope created a medium to study objects too small for the naked eye, and the telescope provided a way to detect objects too distant, AI can help find patterns that aid discoveries in virtually every discipline. This highlights both the potential of AI and the importance of investing in research and development.
The United States government has historically played a large role in funding R&D for the purpose of scientific discovery and experimenting with technology in ways that are not immediately able to be commercialized. This has led to some groundbreaking inventions, such as the internet, barcodes, maps of the human genome, GPS and even baby formula. The U.S. should continue to support research across the entire R&D cycle through investments and partnerships between government, academia, and the private sector.
A BPC report on AI R&D noted a worrying trend, as US investments in basic research have declined as a percentage of GDP from about 1.2 percent in 1976 to about 0.7% in 2018. This decline in R&D funding is particularly concerning when compared with other nations, like China, Israel, and South Korea, who are significantly increasing their own R&D budget. To counter these trends and cement US leadership in AI, BPC’s report outlined six policy principles to guide the federal government’s approach:
- Overall federal AI R&D spending needs to increase significantly
- The country needs to expand and diversify its computing capacity
- The federal government plays a key role in incentivizing private sector R&D
- There needs to be international cooperation to advance AI R&D
- The federal government would benefit from opening up avenues for private talent
- AI standards and measurement are essential to fostering AI technologies that are safe, secure, reliable, and comport with U.S. norms and values
In a promising sign last year, President-elect Joe Biden wrote in an op-ed that research and development would be a “cornerstone of my presidency,” in order to ensure continued US leadership in emerging technology. This sentiment is echoed on the other side of the aisle. Rep. Frank Lucas (R-OK), Ranking Member of the House Committee on Science, Space and Technology, has advocated for increased R&D funding, introducing legislation to double U.S. investment in basic research as a way to secure American leadership.
Whether these efforts will come to fruition is yet to be seen, but the new administration and Congress should be mindful of the importance of AI research and development and the future of innovation. A bipartisan approach that ensures access to adequate resources while protecting privacy and proprietary information can enable the government to maximize the nation’s research potential into AI technologies and their myriad applications, and help unlock AI’s full potential.