Free beta access. 50 free AI tokens each week. Optional top-ups available. See what is included
← SearchGeorgetown2022Affirmative
Opening Word document…

Forcing all NATO countries to adhere to uniform ethical guidelines hamstrings innovation and the competitiveness of the DIB.

Castro & McLaughlin 19 --- *Vice President, ITIF, and Director, Center for Data Innovation. **Research Analyst Information Technology and Innovation Foundation

Daniel & Michael, 2-4-2019, "Ten Ways the Precautionary Principle Undermines Progress in Artificial Intelligence," ITIF, https://itif.org/publications/2019/02/04/ten-ways-precautionary-principle-undermines-progress-artificial-intelligence/

1. Slower and More Expensive AI Development

Policies based on the precautionary principle both slow and make the development of AI more expensive. For example, if all fifty U.S. states had laws such as New York’s, which requires autonomous vehicle firms to perform road testing under the paid supervision of police, testing such vehicles would be more expensive. Moreover, proposals to require even non-medical algorithms to undergo pre-market trials would hurt the development of AI because such trials are time-consuming and expensive. Such proposals may also make AI systems that use machine learning, and thus may change frequently and need more testing, significantly less viable because such systems could constantly need to go through a new approval process.96 Finally, policies that increase the cost of developing AI would likely discourage innovation in AI by creating a substantial barrier to entry for startups that lack sufficient funding to cover the cost of proving their AI system is safe. For example, the GDPR has dampened investment in European technology startups and led to a 30 percent decrease in the market share of small online advertising firms that lack the resources to easily comply with the regulation.97

Restrictions on one AI technology can also limit ways to develop another AI technology. For example, researchers in Germany are using drones hovering hundreds of meters above highways to record the movements of vehicles. This data can help develop simulations to test autonomous vehicles; such simulations are important tools for improving the safety of autonomous vehicles because otherwise they would need to travel billions of miles for safety validation.98 While this novel method of collecting data to validate the safety of autonomous vehicles may or may not prove valuable, implementing it in the United States would be would be difficult to do at scale until the FAA implements its new rules that allow out-of-sight drone flights and flights over people.99

2. Less Innovation

AI will spur innovation so policies that limit the development of AI will limit innovation.100 For example, proposals to ban or limit the introduction of autonomous vehicles would also limit the generation of new businesses, business models, and ways to do deliver services through the “passenger economy.” The passenger economy, a term coined by Intel and research firm Strategy Analytics, “is the economic and societal value that will be generated by fully autonomous…pilotless vehicles.”101 The firms envision a world where a significant portion of vehicle ownership is replaced by fleets of autonomous vehicles that provide on-demand transportation. Productivity would also increase as autonomous vehicles free employees to work during their commutes and autonomous trucks to operate more efficiently. The firms estimate the value of this economy could be $7 trillion by 2050.102 Nations that ban autonomous vehicles will not experience the benefits of such an economy.

3. Lower-Quality AI

There is often a negative correlation between making an AI system more explainable and its accuracy.103 As a result, any policies that require AI to be explainable could lead to less accurate AI. For example, researchers at Mount Sinai Hospital in New York developed an AI system called Deep Patient that can predict whether a patient is contracting any of a wide variety of diseases.104 The researchers trained Deep Patient on the health data from 700,000 patients, using hundreds of variables, such as test results, which allow it to predict diseases such as schizophrenia—which doctors struggle to predict—extremely well.105 Even though its operators can verify its accuracy by measuring outcomes, such as if a person is developing a disease, it is difficult for its own developers to know why it made a particular decision.106

Many sophisticated forms of AI pose a similar problem. Developing an AI system capable of explaining itself or justifying its decisions is an incredibly challenging technical feat, so much so that the U.S. Defense Advanced Research Projects Agency (DARPA) devoted $75 million in 2017 to research how AI could achieve it.107 Some groups are skeptical that requiring explainability would chill innovation. They cite DeepMind, a British company owned by Google parent-company Alphabet, developing an AI system in 2018 that can analyze eye scans to predict diseases while also providing doctors a map of the features of disease it sees, such as hemorrhages.108 However, the fact that one of the world’s leading AI companies could achieve a form of explainability in a system it worked on for nearly two years is not evidence that all other operators should or would be able to achieve explainability for their AI easily.109 To be clear, it is legitimate for companies, such as IBM, to create internal requirements for AI explainability.110 Requiring all firms to meet such a standard, however, would create a barrier to adopting AI, because not all AI systems are alike and not all businesses have a similar level of expertise.

Nonetheless, it is important for AI operators to continually assess their AI system’s accuracy to ensure it is generating or predicting the correct outcomes. The other option is to allow only AI applications that operators can explain; this would lead to AI systems that consider fewer variables and that use simpler algorithms to make decisions.  In turn, this would reduce the effectiveness of AI that can generate significant impacts such as identifying a terminal illness before a doctor can.

4. Less AI Adoption

The right to human review illustrates how attempts to mitigate the impact of AI could also stifle its adoption. One of the reasons firms use AI is because it increases productivity as it can analyze large amounts of data significantly faster and cheaper than humans. For example, LawGeex, a firm that uses AI to automate the review and approval of contracts, created an AI system that outperforms lawyers in identifying risks in non-disclosure agreements (NDAs). During a test in which 20 lawyers and LawGeex’s AI were each given five NDAs to review, the lawyers took an average of 92 minutes to review the contracts and had a mean accuracy score of 85 percent. LawGeex’s AI, however, achieved 94 percent accuracy and only took 26 seconds to review all the contracts.111

A right to human review would require firms to review significant decisions made by algorithms. Such a requirement is particularly problematic because the complexity and amount of data used by some AI systems to make accurate decisions can make it nearly impossible for firms to explain exactly why a system made one decision, even though they may be able to provide a general explanation of how the system works. Thus, it would take significant time and expertise for a firm to explain many decisions made by AI, which then makes using AI more expensive—negating one of its benefits. Firms subject to a right to human review can make one of three choices.  They can: 1) use sophisticated AI, but face litigation if they cannot properly explain a decision, 2) implement simple, and therefore more explainable but less useful, forms of AI, or 3) leverage no AI at all. The first option is not viable over the long term, leaving firms with only the latter two options. And if firms choose either of these options, the economy will be less productive.112

5. Less Economic Growth

PricewaterhouseCoopers predicts AI can boost global gross domestic product by 14 percent by 2030.113 Unfortunately, policies based on the precautionary principle often discourage the use of AI out of fears that AI will eliminate jobs. For example, Amy Webb, founder of the Future Today Institute, which researches emerging technologies, professes, “We need to address a difficult truth that few are willing to utter aloud: AI will eventually cause a large number of people to be permanently out of work...”114

But policies that discourage the use of AI due to the prevalence of such fears rob economies of ways to become more productive, something that all developed nations will desperately need over the course of the next three decades as populations age and dependency ratios increase. If productivity growth really eliminates net jobs, then developed nations should be in depression-like conditions, as productivity over the last 50 years has increased in most nations by over 75 percent.115 The reality is that productivity leads to cost savings, most of which are passed on to consumers in the form of lower prices or to workers in the form of higher wages, both of which spur more spending, which in turn spurs job creation. Consequently, virtually all economic studies show that productivity gains lead to more jobs, even if there is short-term job loss.116 Policies that aim to stem the introduction of AI, and thus automation, will reduce per-capita income growth.

6. Fewer Options for Consumers

Biometric laws show how passing legislation to address hypothetical problems can discourage the use of AI, such that consumers have access to fewer services or products. For example, Illinois users of Facebook, Shutterfly, Google, and Snapchat have all sued the companies for scanning their faces without consent, which is illegal under the state’s Biometric Information Privacy Act.117 Regardless, the companies were typically sued for relatively innocuous uses of AI, such as for scanning individuals’ faces to tag them in photos or to add alterations to photos.118

Such threats of legal action do and will lead to fewer services for consumers. For example, Illinois and Texas’ biometric laws led to Google blocking individuals in those states from using its Arts and Culture app.119 Millions of individuals have downloaded the app, which scans users’ faces and compares those images to those of paintings in Google’s database to find users’ doppelgängers in famous art.120

Similarly, some lawmakers have already passed precautionary legislation related to autonomous vehicles that limit consumers’ options. For example, Washington, D.C. enacted a law in 2013 that requires a licensed human driver in the front seat of autonomous vehicles who is prepared to take control of the vehicle at any moment. This requirement means people with certain disabilities, who would like the independence that would come from using autonomous vehicles but do not qualify as a capable human driver under this law, are unable to use autonomous vehicles, even if they can safely operate them.121 The requirement might be reasonable given the current state of the technology, but locks in a standard that is unhelpful over the long term. Instead of a broad restriction that requires a capable individual in the driver’s seat of autonomous vehicles, government regulators, such as the National Highway Traffic Safety Administration (NHTSA) in the United States, should develop and enforce safety standards that preempt local laws, but allow the operation of fully autonomous vehicles that meet safety standards.

7. Higher Prices

By raising the costs of using AI for operators or by banning forms of AI, policies based on the precautionary principle also keep prices high for consumers. For example, some policies would require businesses to get express consent before using facial recognition. Yet, U.S. stores lose nearly $50 billion every year due to shoplifting, and facial recognition could reduce that figure by helping catch repeat offenders.122 Shoplifting costs consumers because money lost from shoplifting leads to higher prices—shoplifting cost the average U.S. family over $400 in 2009—instead of being put towards increased investments in customer experience improvements.123

Another way the precautionary principle keeps prices artificially high for consumers is by limiting the ways firms can offer their services or products through bans. For example, delivery robots can perform the “last mile” of a delivery—where transporters move packages from a central hub to an individual’s residence. Today the process is time-intensive and can be up to 28 percent of a product’s transportation cost.124 As a result, efforts to ban robots on sidewalks, which could reduce these costs, would rob consumers of faster and cheaper deliveries.

Likewise, slowing the introduction of autonomous trucks hurts consumers. There was a shortage of 51,000 truck drivers in the United States in 2017 which grew to 63,000 in 2018.125 Truck driver turnover rates are also 94 percent, meaning employers in the for-hire trucking market need to replace the vast majority of employees they hire every year.126 The situation has already led to delayed deliveries and higher prices for consumers and may only get worse because there will be 900,000 truck driver openings in the United States over the next decade due to retirements.127

8. Inferior Consumer Experiences

Policies that require firms to get prior consent before using commercial applications of AI, including facial recognition, can actually delay improvements in consumer experiences. For example, AI may be able to reduce the effects of implicit bias—the stereotypes that affect human actions in an unconscious manner. These stereotypes lead to people of color getting falsely accused of theft by store employees.128 Indeed, employees for Nordstrom Rack, Staples, and Finish Line have all wrongly accused African-Americans of theft in 2018.129 But AI technologies can improve the consumer experiences of all people, including people of color, by replacing or complementing human decision-making. Amazon Go, one of Amazon’s cash-register-less stores, uses cameras, sensors, and computer vision technology to “see” who takes items off shelves, adding these items to a virtual shopping cart so that checkout is seamless. Many retailers are moving in this direction, with over 150 companies, including 7-Eleven and two of China’s largest e-commerce businesses, Alibaba and JD.com, experimenting with using facial recognition and other biometrics to eliminate the need for cashiers. As a result, store employees are not looking for potential shoplifters because the technology automatically charges customers for what they take when they leave the store.130 After visiting Amazon Go, former CNET senior associate editor Ashlee Clark Thompson, an African-American journalist, wrote “No one cared what I was doing. Is this what it feels like to shop when you're not black?”131

9. Fewer Positive Social Impacts

AI can and already is generating positive social impacts, from mapping poverty to measuring literacy rates to helping doctors treat deadly infections.132 It is also helping make society safer, but the demonization of such AI applications as facial recognition and proposals to phase in AI could derail its benefits. While privacy advocates are stoking fears of future mass surveillance, law enforcement is already using facial recognition for several positive purposes. These purposes include identifying uncooperative suspects, such as the Capital Gazette shooter.133 In addition, some airports, such as the Washington Dulles International Airport, employ the technology to catch individuals using false documents.134

Law enforcement also uses AI to find victims. For example, the Fort Worth Police Department uses a combination of AI tools from Marinus Analytics, which builds AI tools to fight human trafficking, including facial recognition, to identify victims of human trafficking. With thousands of escort ads appearing online each day, AI can significantly reduce the time it would take a detective to go through the ads manually.135 While no government should use facial recognition to undermine personal freedoms and rights among its citizens or to unfairly target certain demographic groups, nations can mitigate negative uses without creating bans that curtail the use of beneficial ones.

Unfortunately, misguided proposals to curtail negative impacts from AI can create other negative impacts. For example, phasing in autonomous trucks to lessen job loss would be detrimental to the environment. Tractor trailers account for a disproportionate amount of greenhouse gas emissions, but autonomous trucks can take advantage of platooning, a form of driving where the trucks drive closer together than humans can by using vehicle-to-vehicle communication and sensors to automatically break and accelerate together.136 The trucks following the leader experience less wind resistance, which improves fuel efficiency.137 Limiting the number of autonomous trucks on roads, however, keeps emissions from trucks higher than necessary. In addition, banning autonomous vehicles in the United States would rob the nation of a potential $900 billion in yearly savings from fewer crashes.138

10. Reduced Economic Competitiveness and National Security

Nations that slow AI adoption will metaphorically tie one hand behind the backs of their companies competing in global markets. Moreover, for nations such as the United States, finishing behind China in the global race to be the leader in AI not only limits its ability to influence the development of AI, but also raises national security concerns due to the many potential national security applications of AI and the reduced competitiveness of the defense industrial base.139

Sign in to recut, update, or save this card.