Pivotal technologies have helped accelerate the discovery of other new technologies throughout history. For instance, the telescope powered the fields of astronomy and physics, while the microscope led to new innovations in the field of biology.
Artificial intelligence has the potential to be such a technology. For instance, in the field of material science, AI has already aided in the discovery of new materials for a broad range of applications including in clean energy, advanced electronics, and next generation aerospace technologies. A number of research funding agencies—including the National Science Foundation, the Department of Energy and the National Institute of Standards and Technology—have been exploring ways to expand the use of AI technologies to accelerate these discoveries. This suggests AI can greatly boost our understanding of the world and foster innovation beyond the realm of computer science.
In thinking about innovation and ways to accelerate it, three interrelated challenges merit consideration:
- The Burden of Knowledge: The expertise necessary to make original contributions in a field is likely to grow the further we advance technologically, which suggests researchers will require more education and larger teams to innovate.
- Discovering Useful Combinations: Innovation often requires combining different inputs, data, and ideas, but it is difficult to discover what combinations are useful, especially as the amount of information available continues to grow.
- Risk and Uncertainty: The greater the risk and uncertainty around an R&D project, the more costly it becomes to finance.
These challenges are all currently present and will persist for the foreseeable future. However, AI provides a tool to help mitigate them:
- AI can help reduce the burden of knowledge by making knowledge more accessible. For example, many researchers have been using AI as a tool to more quickly gather and synthesize a large number of research papers, which can greatly speed up the literature review process.
- AI can help more quickly and efficiently find combinations that spur innovation. Some of the most pressing challenges today, including climate change, require multi-disciplinary approaches to innovation. The Advanced Research Projects Agency – Energy (ARPA-E) has recently invested in the development of AI tools which combine wider ranges of innovative concepts together which are capable of designing better photovoltaics, turbine blades, and power electronics to enable the decarbonization of the U.S. economy.
- AI can help more accurately calculate the risk of an R&D project, which can reduce financing cost and allow companies to more efficiently allocate resources. For instance, AI is being used by pharmaceutical companies to predict the likelihood of success for a drug development program. If successful, these efforts can reduce the cost of developing drugs, while improving patient outcomes.
AI offers many benefits toward boosting innovation, but it can be overhyped, potentially leading to counterproductive disillusionment. Therefore, it is important to provide a realistic assessment of its potential and highlight some of its limitations:
- AI can be bound by data limitations. A lack of good data can limit the effectiveness of AI and prevent it from reaching its full potential. For instance, an AI trying to discover medical treatments personalized to specific patients would be less effective if it has less data about different patients to train itself with.
- AI can be bound by the availability of computing power. Similar to data limitations, AI can be constrained by how much computing power it has to process data.
- The burden of knowledge can only be shrunk so much by AI. AI can help reduce the burden of knowledge for researchers, but there are limits to how much this can shrink. By analogy, internet search engines have made browsing for academic articles easier because people don’t have to go to a library and find articles manually, but one still needs to have a general sense of what they are looking for and how to weed through search results.
- There are limits to the types of combinations and innovations that AI is good at finding. This could result in a spurt of AI-based innovation at first, but a dramatic slowdown after the “low hanging fruit” are picked.
- AI cannot do everything in the creativity process. For instance, AI is currently not good at making causal inferences, so domain expertise will continue to be important in dealing with issues of causality. AI will be a tool to enable domain experts to work more efficiently, but experts still need to be experts.
Policymakers can help optimize the use of AI for driving innovation, but they need to ask the right questions. Several questions include:
- How can the government better incorporate AI into its R&D roadmaps?
- What can the government do to make datasets more accessible to university and public-sector researchers, while maintaining proper safeguards for sensitive information?
- Can public policy play a role in fostering the effective use of AI in R&D at the private sector and universities?
- How can we increase public and industry awareness about the potential (and limitations) of AI to boost the rate of innovation in the United States?