This piece considers the general state of policy on AI and also gives some numbers to the federal activity encouraging research and application in this field.
Growing interest in Artificial Intelligence (AI) comes from a range of audiences including academia, commercial industry, the entertainment industry, and now the White House Office of Science and Technology Policy (OSTP). On October 12, 2016, OSTP released two new documents on AI. The National Artificial Intelligence Research and Development Strategic Plan is a high level framework for prioritizing and coordinating federal research and development (R&D) to advance AI. A companion piece, Preparing for the Future of Artificial Intelligence, discusses the technologies that can be identified as AI and how these may be used to benefit society. A third document, to be authored by the President’s Council of Economic Advisors and released later this year, will explore how AI might affect employment.
AI—along with related initiatives on topics ranging from precision medicine to the Internet of Things (IoT)—was also showcased at the White House Frontiers Conference. But the winds of change blow fiercely in election years. With the presidential election approaching, one key task for the incoming administration will be to take account of current OSTP policies and determine whether these will be bolstered or recast. The rest of this piece considers: How can the next administration advance current policies? And what can be done in and beyond the White House to ensure that AI R&D remains a national priority in the coming years?
There are many technologies that can be grouped under the designation of AI, from general techniques to specific applications. One general technology is deep learning, which is a set of techniques for gradually structuring data by defining it in layers. A more specific application is machine translation, which powers the familiar service Google Translate. While many AI technologies were initially created in research laboratories, innovation in AI technology is increasingly driven by industry, thanks to (for example) market incentives for investing in digital personal assistants and autonomous driving.
Any administration that makes AI a priority will want to see the US emerge as a global leader. Unfortunately, as a relatively new area of broad public interest the exact levels of financial investment in AI research are not well-defined. Thus, determining where funding originates or how exactly it is spent is challenging. As one starting point, the National AI R&D strategy gives a number of $1.1 billion in unclassified, federally funded R&D for AI.  However, without more statistics for AI investment, it is difficult to compare US federal investments with funding in other countries. Some other approximations of productivity are possible, including comparing the number of relevant journal articles and patents in countries like China and the US (e.g., NSTC’s strategy in the National Artificial Intelligence Research and Development Strategic Plan), or comparing the number of computer science publications by country.
Another approach is to consider general trends in research funding. In 2013 gross expenses of R&D for the US and China were $457 billion and $336.5 billion respectively, which puts China’s R&D spending at 73.6% of US investments  Comparing these figures to the size of the respective economies, China spent 2.1% of GDP on R&D in 2013, vs. 2.7% in the US. Further, Chinese R&D spending is growing rapidly. Between 1993 and 2013, US spending on R&D had a compound average annual growth rate of 3.1%, while Chinese spending grew at 17.1%, adjusted for inflation. These figures reflect both government and commercial spending, and are offered as background on the two global leaders in AI R&D, China and the US.
Assuming parity in the percentage of gross funding for AI research in the PRC – a weak assumption at best – if not the amount spent on research generally, in 2012 funding could amount to $710 million for AI research. . Foreign competition however is created by foreign firms, not just facilitated by government-funded research. For examples of commercial R&D there is PRC based Baidu’s investments in AI last year, amounting to $1.5 billion.
If AI can deliver on the promises of social change and economic growth, this will be easiest when communication among academics, businesses, and government is coordinated but not constrained. Competition may be effective for supporting isolated developments and specific applications of AI—like software designed for autonomous long-haul trucking. But there will be some questions, like the challenge of ensuring safe and equitable AI development, which are by nature national in scale. With this in mind – as suggested by the National Strategic Plan – federal support for AI should establish a specific research portfolio to give federal agencies a clear indication of basic research priorities.
The next administration could encourage OSTP to further coordinate federal funding schemes and also to facilitate more communication with business and academic stakeholders. Regarding specific organizations to handle these efforts, there is the National Information Technology Research and Development (NITRD) subcommittee of the National Science and Technology Council (NSTC) already in place for the former role, and the National Institute of Standards and Technology (NIST) for the latter. In the National Strategy discretion is left to the funding agencies, though with the suggestion of coordinating with NITRD on the exact implementation of allocating funds. Naturally, considering other demands on agencies’ time and resources it is possible that adherence could be uneven.
OSTP should also continue to organize fora, symposia, or other consultation processes to engage the public at large. Activities could build upon the Request for Information (RFI) that OSTP issued in June 2016, which informed the report Preparing for the Future of AI. The new administration should also consider exploring bilateral or multi-lateral cooperation in AI, particularly to support basic research. This will also require further coordination and organizational work. One example of an opportunity for multi-lateral cooperation is creating, endorsing, or spreading standards, which is a task that governments are well suited to. Again, this is something NIST can lead in the US with relatively few additional resources. One benefit of using informal standards or “best practices” for AI is that these reduce uncertainty in planning and allow for more time to be spent on improving efficiency and effectiveness of approaches and methods, rather than just keeping a business or product operational.
Interest in AI comes from not only the federal government, but from states as well, including Nevada and Michigan. The initiative of states to make certain applications of AI – such as autonomous vehicles – legal is itself a process of innovation. Generally the more complex a process is the more difficult it is to align all the elements of it to all work at the same time, never mind to align them all to work well. Given all the moving parts to the federal government setting policy to encourage AI research can require hundreds of people and can take months or perhaps years for an effect to appear. However, experimentation can occur first with state law and policy, which is often easier to formulate and implement given the difference in scale. Letting states take the initiative to experiment with their own policies could offer valuable lessons for the federal government to iterate on at a larger scale. Naturally, there is a risk that 50 states will create 50 different policies, so this approach should not be indefinite.
The Obama Administration’s AI policy offers a vision. Whether the next administration provides for the journey is open question.
 Considering the quoted figure of federal AI spending $1.1bn against total federal R&D spending ~$152bn requested for FY17, it amounts to <1% (~0.7%) of total federal R&D. Focusing on research, of the $72.8bn requested for FY2017, the $1.1bn figure amounts to 1.5% of federal research.
 National Science Board. 2016. Science and Engineering Indicators 2016. Arlington, VA: National Science Foundation (NSB-2016-1), https://www.nsf.gov/statistics/2016/nsb20161/#/, appendix table 5-31.
 For more information see tables 4 – 3, 5, 6, 11, 16 and figures 4 – 8 and 13, Science and Engineering Indicators 2016.
 Table 4-4, Science and Engineering Indicators 2016.
 Appendix Table 4-12, Science and Engineering Indicators 2016.
 Appendix Table 4-12, Science and Engineering Indicators 2016.
 Table 4-6, Science and Engineering Indicators 2016.
 Research includes basic and applied research and omits development and R&D plant.
 The inaugural report from Stanford titled the One Hundred Year Study on Artificial Intelligence (AI|100) section on policy and legal consideration goes into some detail on where the intersections for these applications and policies are.
Peter Stone, et. al. “Artificial Intelligence and Life in 2030.” One Hundred Year Study on Artificial Intelligence: Report of the 2015-2016 Study Panel, Stanford University, Stanford, CA, September 2016. Doc: http://ai100.stanford.edu/2016-report. Accessed: October 12, 2016.
 Just as asking international partners to join in research or product development spreads risk and brings in expertise for companies, informal standards also lower the discovery cost of finding the most efficient method or application for accomplishing tasks.