Policy Recommendations for an Artificially Intelligent Problem

As a car travels down the road, it is faced with a dilemma. Should the vehicle, driven by Artificial Intelligence (AI), swerve to protect multiple pedestrians or continue on its path and protect the passengers of the vehicle? Corporations may choose to value the life of the passenger because the passengers are their customers. This is just one of many issues faced today as AI is developed. To protect public safety, privacy, and interest, the government should introduce policy to help regulate AI. While suggesting policy options, the uses of AI, the problems with AI, and the recommended regulations need to be discussed.

Artificial Intelligence is not a new field of study but for many people it is shrouded in mystery. Knowing the different types of AI used, the current applications of AI, and the future of AI is all important to understanding the issues. According to a government paper the term Artificial Intelligence was coined in 1956, but the concept was introduced earlier by Alan Turing who asked the question, “Can machines think?” (Executive Office, 2016). Despite the concept being around for numerous years, there is still not a strict definition for AI. In his book The Quest for Artificial Intelligence: A History of Ideas and Achievements Neil J, Nilsson (2009) defined AI as “that activity devoted to making machines intelligent, and intelligence is that quality which enables an entity to function appropriately and with foresight in its environment” (p. 13). Although AI is loosely defined, there are two main categories - General AI and Narrow AI. General AI is a system that exhibits “intelligent behavior as advanced as a person across the full range of cognitive tasks” (Executive Office, 2016). This AI is the type often depicted in our media. HAL 9000 from 2001: A Space Odyssey and Cylons from Battlestar Galactica are both examples of General AI. Narrow AI is made to address specific application areas. Narrow AI is what is being discussed currently for regulation as General AI “will not be achieved for at least decades” (Executive Office, 2016).

The current applications of AI are all examples of Narrow AI. Highly Automated Vehicles (HAVs) and machine learning are both major applications. HAVs are vehicles that utilize AI to pilot the vehicle with limited to no human interaction. These vehicles are most commonly seen as self-driving cars, but HAVs also cover airplanes, buses, trucks, and other modes of transportation. HAVs will provide new access to transportation to a wide range of people. Self-driving cars will provide opportunities to “people with disabilities, ageing populations, communities where car ownership is prohibitively expensive, or those who prefer not to drive or own a car” (US. Department, n.d., p. 5). While self-driving cars provide the most benefit to the general public, the application of AI in other vehicles is equally important. The applications of AI in the shipping industry provide enormous benefits. Companies are already testing their ability to transport goods using a completely autonomously driven truck (Santens, 2015). This simply means that the vehicle operates with no human interaction.

Meanwhile, machine learning “is a statistical process that starts with a body of data and tries to derive a rule or procedure that explains the data or can predict future data” (Executive Office, 2016). This AI collects and interprets data to find patterns. The data the AI interprets is often called big data, which simply refers to a large amount of data. Machine learning is commonly used by the government as well as large corporations such as Facebook and Google. Corporations tend to use machine learning to target advertising and services to individuals. They use the AI to gather data and extrapolate patterns in people’s preferences. This allows the AI to form conclusions on the likes and dislikes of any individual therefore increasing an individual’s likelihood to interact with an advertisement. Machine learning can also be used in different ways. For example, credit card companies use machine learning to detect fraudulent credit card charges which is a useful application. Similar to corporations who use AI to target advertising, the government utilizes machine learning to detect behavior patterns to identify possible terrorists. Unfortunately, the application of machine learning in the government is incredibly similar to how corporations utilize AI. Paul Dempsey (2016), journalist for E&T Magazine, states, “Once you start talking about AI innovation that addresses ‘big data’ as noted earlier, applications with commercial and military value could even prove to be identical.” Despite the intended results being different between the government and corporations, it begs the question of if there should be regulations differentiating between government and commercial use.

Although Narrow AI is more visible in our everyday lives, the development of General AI continues. These types of AI are still decades away, but it is worth looking to the future to be prepared. Currently, deep learning AI is an offshoot of machine learning. Deep learning looks to replicate the human brain by “[using] structures loosely inspired by the human brain, consisting of a set of units (or ‘neurons’)” (Executive Office, 2016). Currently the application of deep learning AI fits more into the realm of Narrow AI, the research being done to replicate human processes in software looks to lead to new research related to General AI. The research done on AI will continue to advance with time, but the potential problems at our doorstep require our attention sooner rather than later.

As AI research continues, new problems and questions arise. The future uses of AI, large data collection, and HAVs all introduce issues that need to be addressed with government policy. Due to the rapid rate at which AI development occurs, it causes difficulties in government regulation. There is no way to predict where AI development will go, so any policy looking to regulate future AI is difficult. Even AI that exists currently is under constant development. This requires any regulation introduced to be flexible so it can adapt to new research. For example, HAVs are slowly being introduced to the roads. As testing occurs and new data is collected the functions that HAVs are able to undertake increases. Some predict that HAVs will eventually be linked to a network in charge of regulating traffic. However, this is only a prediction and is currently not integrated into HAVs. This uncertainty is just further reason that any enacted policy must be flexible.

The issues involved in large data collection are commonly debated. From the public outcry after Edward Snowden released NSA documents proving the existence of large data collection by the government to large corporations collecting data on its customers. In both cases, the use of AI is at the forefront. AI gathers large amounts of public data to interpret in both commercial and governmental applications. Both use AI to notice patterns in the data. However, the patterns the AI looks for are different between the two applications. Where the government looks for patterns related to national security, corporations look for patterns to increase their profits. In both cases the amount of data collected is enough to make people question their personal privacy. It is easier to justify the government’s usage of machine learning than it is to justify commercial use. While the government looks for patterns exposing terrorism efforts, corporations look for patterns to best market to an individual. While corporate uses can increase convenience, for many it does not justify the use of data as effectively as the government use. So, should regulations affect corporations and government differently when they are using the same technology?

HAVs are faced with similar issues. There are ethics concerns, questions of fairness, safety issues, and legal issues. A common ethical question when it comes to self-driving cars is the ‘Trolley Problem.’ Larry Greenmeier (2016), an associate editor for Scientific American, described the Trolley Problem as a situation where a car is approaching a group of pedestrians and must make a decision on whether to swerve and potentially harm the driver, or to protect the driver and passengers while choosing to harm the pedestrians. Although the Trolley Problem is only a theory, its practical application raises interesting questions about what sort of decisions HAVs will have to make and how they will make them. Researchers at the Massachusetts Institute of Technology (MIT) created a website to address this issue. The goal of the “Moral Machine” is to “take the discussion further, by providing a platform for 1) building a crowd-sourced picture of human opinion on how machines should make decisions when faced with moral dilemmas, and 2) crowd-sourcing assembly and discussion of potential scenarios of moral consequence” ("Moral Machine," n.d.). An interesting twist that the Moral Machine introduces to its examples are a difference in “social value.” They provide examples with athletic people, doctors, convicted felons, older people, younger people, and more. This raises an interesting question of if HAVs could or should discriminate based on these defining factors. Whatever the answer to this question, the decision would need to be enforced by the government to protect public safety. The U.S. Department of Transportation (USDOT) (n.d.) has already proposed regulation to allow for safe testing of HAVs, but it is currently only a proposition. Artificial Intelligence needs the ability to act safely outside of a controlled environment, which is why standards for safe testing are needed. Beyond that, how should accidents involving HAVs be handled? Should the manufacturer of the car be held responsible for the decisions made by its AI? If there is a lack of standards for an AI to follow then companies have the authority to make decisions similar the Trolley Problem on their own. This could harm public safety as corporations might make the choice to favor the safety of passengers over the safety of the public. Another question of safety is if HAVs should require a licensed driver to be behind the wheel. While initially this requirement is reasonable to ensure the safe operation of HAVs, down the line as HAVs are refined would it still be necessary? While having a human behind the wheel is a good precaution to have, it may not be necessary when every car is a HAV and the risks involved in everyday transportation decrease. Even though not every issue or question can be answered right now it is important to consider policy options to protect public safety and privacy.

What government action should be taken? Currently there are recommendations for regulation, but no definite policy or legislation. These recommendations include the introduction of ethics education, the standardization of safety procedures, and the role of government in these policies. Ethics education is a great first step to take in the discussion of AI. By promoting education on ethical issues it allows future industry professionals to make informed decisions on the moral obligations of an Artificial Intelligence. As mentioned in Preparing for the Future of Artificial Intelligence released by the White House, “Schools and universities should include ethics, and related topics in security, privacy, and safety as an integral part of curricula on AI, machine learning, computer science, and data science” (Executive Office, 2016). Although safety regulation does relate as well to data collection as it does to HAVs, it is still an important part of AI regulation. The proposed “Safety Assessment” from the USDOT outlines areas of importance when it comes to safe operation of HAVs. The USDOT recommends that:

The Safety Assessment [should] cover the following areas: Data recording and sharing, privacy, system safety, vehicle cybersecurity, human machine interface, crashworthiness, consumer education and training, registration and certification, post-crash behavior, federal, state, and local laws, ethical considerations, operational design domain, object and event detection and response, fall back (minimal risk condition), [and] validation methods (US. Department, n.d., p. 15).

Much of this suggestion is targeted towards HAVs, but some of it could be retooled to apply to machine learning and data collection. For example, the recommendation for privacy in the USDOT document discusses the importance for transparency, choice, and security (US. Department, n.d., p. 19). These are aspects that could be the basis for a law regarding the commercial use of machine learning. Other than passing laws regulating AI, the government is expected to be involved in many aspects of the policy. For example, the USDOT provided a framework for the federal and state roles in HAV policy. They recommend actions such as “issuing guidance for vehicle and equipment manufacturers to follow” and having “each state identify a lead agency responsible for consideration of any testing of HAVs” (US. Department, n.d., p. 40).

While these recommendations are a start, there is concern over whether or not future regulation will stifle industry growth. To address this concern it was suggested that “policies should be designed to encourage helpful innovation, generate and transfer expertise, and foster broad corporate and civic responsibility for addressing critical societal issues raised by these technologies” (Stone et al., 2016, p. 42). The White House has adopted this stance in Preparing for The Future of Artificial Intelligence:

Government has several roles to play. It should convene conversations about important issues and help to set the agenda for public debate. It should monitor the safety and fairness of applications as they develop, and adapt regulatory frameworks to encourage innovation while protecting the public (Executive Office, 2016).

This statement along with mentions of wanting to work with the industry show that the government is looking to cooperate, not dictate. Hopefully these suggestions combined with the government’s willingness to communicate will lead to effective and safe regulations.

As Artificial Intelligence grows regulation must grow with it. This does not mean that regulations must restrict AI. Regulations can be put in place to protect public safety without harming innovation. Protections need to be put in place to protect public safety, and I believe that the actions being taken by the government along with their efforts to foster an environment of communication are effective ways to handle an unpredictable industry. As of this writing the Executive Office of the President has not released a follow up document which would “further investigate the effects of AI and automation on the U.S. job market, and outline recommended policy responses” (Executive Office, 2016). I look forward to the release of this document because I am interested to see what AI policy the White House recommends. AI has great potential to help our society and moving forward it must be utilized responsibly to ensure the safety and privacy of everyone. Yes, the unknowns of AI are terrifying, but at the same time they are equally exciting.

References

Dempsey, P. (2016, October 4). The wild west of AI regulation revealed in Westworld remake [Newsgroup post]. Retrieved from E&T website: https://eandt.theiet.org/content/articles/2016/10/view-from-washington-the-equally-wild-west-of-ai-regulation/

Executive Office of the President. (2016, October). Preparing for the future of artificial intelligence. Retrieved from https://www.whitehouse.gov/sites/default/files/whitehouse_files/microsites/ostp/NSTC/preparing_for_the_future_of_ai.pdf

Government overregulation [Newsgroup post]. (2016, November 6). Retrieved from Desert News website: http://www.deseretnews.com/article/865666574/In-our-opinion-Government-overregulation.html?pg=all

Greenmeier, L. (2016, June 23). Driverless cars will face moral dilemmas [Newsgroup post]. Retrieved from Scientific American website: https://www.scientificamerican.com/article/driverless-cars-will-face-moral-dilemmas/

Moral machine. (n.d.). Retrieved from http://moralmachine.mit.edu/

Nilsson, N. J. (2009). The quest for artificial intelligence: A history of ideas and achievements [PDF]. http://doi.org/10.1017/CBO9780511819346

Pultarova, T. (2016, June 3). Humans will need to be upgraded to keep up with AI says Musk [Newsgroup post]. Retrieved from E&T website: https://eandt.theiet.org/content/articles/2016/06/humans-will-need-to-be-upgraded-to-keep-up-with-ai-says-musk/

Santens, S. (2015, May 14). Self-driving trucks are going to hit us like a human-driven truck [Newsgroup post]. Retrieved from Medium website: https://medium.com/basic-income/self-driving-trucks-are-going-to-hit-us-like-a-human-driven-truck-b8507d9c5961#.rgzr003bl

Stone, P., Brooks, R., Brynjolfsson, E., Calo, R., Etzioni, O., Hager, G., . . . Teller, A. (2016, September). Artificial intelligence and life in 2030: One hundred year study on artificial intelligence. Retrieved from https://ai100.stanford.edu/sites/default/files/ai100report10032016fnl_singles.pdf

US. Department of Transportation. (n.d.). General automated vehicles policy. Retrieved from https://www.transportation.gov/AV/federal-automated-vehicles-policy-september-2016

Worstall, T. (2016, October 12). Exactly what we don't need - regulation of AI and technology [Newsgroup post]. Retrieved from Forbes website: http://www.forbes.com/sites/timworstall/2016/10/12/exactly-what-we-dont-need-regulation-of-ai-and-technology/#70afe6d51216