The White House has recently released its Preparing for the Future of Artificial Intelligence report. The report reviews the history of AI, outlines a few applications, surveys current efforts by federal agencies, and makes some policy recommendations. Like most such government reports, it is more about the present than the future.
The federal government has funded significant research in AI through programs like the DARPA Grand Challenge to develop autonomous vehicles. The report recognizes agencies, such as the Defense Advanced Research Projects Agency (DARPA), the National Science Foundation (NSF), the National Institutes of Health (NIH), the Office of Naval Research (ONR), and the Intelligence Advanced Research Projects Activity (IARPA), as being at the forefront of the AI movement. At the same time, it suggests that some agencies, such as the Department of Labor, could do more. The report wisely avoids the issues of jobs and employment, but makes some sage recommendations, while avoiding obvious implications.
Current Interest in AI
The section titled “A Brief History of AI” largely mirrors Tractica’s Artificial Intelligence for Enterprise Applications report by identifying the three principal reasons for the current interest in AI as being:
- Vastly more data
- Much faster hardware
- More efficient algorithms
Strong AI has now been rebranded as general AI and is not expected for many decades. The report ties robots deeply with the future direction of AI, although our own point of view at Tractica is that the importance of AI in the cloud should not be minimized. The report also identifies human-machine teams, which is another way of saying computer-assisted work in which the human maintains control and responsibility, as the current state of the technology.
Good Recommendations for Open Data Standards
The report includes 23 recommendations, some of which are quite good. One of the best is the second, which states:
Federal agencies should prioritize open training data and open data standards in AI. The government should emphasize the release of datasets that enable the use of AI to address social challenges. Potential steps may include developing an “Open Data for AI” initiative with the objective of releasing a significant number of government data sets to accelerate AI research and galvanize the use of open data standards and best practices across government, academia, and the private sector.
In the world of AI, you cannot be too thin, too rich, or have too much data, and the progress of many projects is hindered by the lack of good data.
A Good Role for Government
From our perspective, one of the most important parts of the report is the “Safety and Control” section in the “Fairness, Safety, and Governance” chapter. It borrows heavily on the recent Concrete Problems in AI Safety report and identifies certain standards for AI, including:
- Avoiding negative side effects
- Avoiding reward hacking
- Scalable oversight
- Safe exploration
- Robustness to distributional shifts
In the future, this may be the most important role of government in AI, because it is clear that the industry will not be able to accomplish this on its own.
Troubling Concept of AI in Weapons Systems
The most troubling section of the report is about the use of AI in weapons systems. As the report points out, automation has been a part of weapons for decades. But with AI, the risks are much greater. If an AI credit rating system either approves everyone or rejects everyone, the implications are different than a weapons system that either kills everyone or kills no one. Inherently, there needs to be a way every system can be turned off without being killed, but does that not create a vulnerability that an enemy could exploit? In Iraq and Afghanistan, official statistics tell us that 21% of fallen soldiers already die from “friendly fire,” although the actual number is undoubtedly higher. One measure of AI’s success will be seeing if this number increases or decreases.
Recognizing the Future Importance of AI
The report has some faults. It does not really consider some of the obvious implications of its predictions. For example, if self-driving cars are going to replace human-driven cars, they are almost certainly going to be electric. This means that, in a few decades, if there are going to be millions of electric cars plugging into the electric grid, upgrading the grid should be a national priority to prepare us for the future of AI.
Between WikiLeaks and Access Hollywood, data has already become a major issue in the current presidential election. AI is likely to grow in importance in government over the years and Preparing for the Future of Artificial Intelligence is another step in that direction.