Why we need to invest more in AI
The White House recently released the NITRD Supplement to the President’s FY2020 budget indicating the budget for research and development in artificial intelligence across all federal agencies. The announcement comes on the heels of the White House Summit on AI in Federal Government that brought together academics, key industry innovators and thought leaders, aiming to re-think government services in the new algorithmic age.
Combined, the requested FY2020 budget for all non-defense AI-related R&D programs in the entire U.S. federal government — across the DOE, NIH, NSF, NIS, IARPA, etc. — is just $973 Million.
This is not enough — just one-sixth of one percent of the federal R&D budget dedicated to AI. Funding for a technology that will be so transformative to the U.S. economy and so critical to our national and physical security, must scale proportionally to the rest of the U.S. R&D budget.
The Donald Trump administration has shown great leadership in championing the cause of AI: developing guidance for the implementation the OPEN Government Data Act, driving the development of the Federal Data Strategy, issuing the Executive Order on Maintaining American Leadership in AI, updating the National AI R&D strategy, and urging NIST to develop a plan for Federal engagement in AI Standards. However, they came up short on this budget.
While fiscal responsibility is a critical constraint, we must carefully consider whether less than $1 billion per year is sufficient to counteract China’s threat of AI hegemony by 2030. As evidenced by the Center of Data Innovation’s extensive research, the United States is still winning the AI race against China — for now. But there is evidence that China is catching up. China’s government spending, its aggressive and well-coordinated industrial policy is allowing them to shrink our lead, which may soon evaporate without appropriate continued investment.
The United States is also falling behind other countries in innovation. According to Bloomberg, the U.S. has fallen out of the top 10 most innovative countries for the first time in the six years their gauge has been compiled.
According to ITIF, the U.S. is likewise behind on its target for productivity growth, which is critical to increasing per capita income and lowering the debt-to-GDP ratio. Investing in robotics, autonomous systems, and AI boosts productivity growth, but federal spending on R&D as a share of GDP has consistently fallen from its levels in the 1960s.
Public R&D spending also encourages private investment through the spillover effect. OECD data shows one public dollar given to private firms results in additional 1.70 dollars of research on average. Lower federal investment in AI research means lower U.S. AI investment across the board.
Further OECD data found that AI start-ups have so far attracted around 12 percent of all worldwide private equity investments in the first half of 2018, a steep increase from just 3 percent in 2011. But since 2016, China’s AI start-up investment has seen an even more dramatic upsurge: from just 3 percent in 2015, Chinese companies attracted 36 percent of global AI private equity investment in 2017.
Great strides are yet to be made in AI fundamental research. Deep learning and reinforcement learning may one day exhaust their utility and open the space for paradigm shifting theories. We know that we need to rethink many fundamental concepts behind what we characterize today as machine intelligence. Today’s supervised deep learning — where most private investment is made — is data hungry, power hungry, lacks versatility, is incapable of thinking rationally, and only learns by observation — which makes it prone to bias.
We may need to re-think machine intelligence that will be inherently adaptable and evolving, that can understand constructs like causality (not just correlation), temporality, open-ended inferences, axiomatic knowledge, logical reasoning, and common sense; intelligence that can be somewhat predictable, transparent and explainable — and more resilient to adversarial attacks.
Finally, public investment in AI is essential to promoting a healthy business ecosystem and a sustainable pipeline of incredible talent. The rise of AI research hubs abroad has put a strain on acquisition of talent in NYC, Silicon Valley, and other U.S. hubs. National labs and academic institutions cannot compete in compensation due to budgetary limits.
Congress ought to take all of this into consideration when appropriating the federal R&D budget. We need to significantly up the ante. Legislative reforms, regulatory approach, national initiatives, and strategic moonshots will keep us in the race. Unfortunately, the goalpost in this race is continuously moving, so if the United States is not agile enough, aggressive enough and spending in proportion to the competitive threat and opportunity we are facing, we will not be in the lead for long.
Mina Hanna is the Chair of the IEEE-USA Artificial Intelligence and Autonomous Systems Policy Committee, and Co-Chair of the Policy Committee of the IEEE Standards Association’s Global Initiative on Ethics of Autonomous and Intelligent Systems. He is a senior software consultant at Synopsys, Inc. and previously worked at Intel Corporation. He received a Master of Science in Electrical Engineering from Stanford University. The views voiced here are his alone.