Who's Listening? the Fourth Amendment in the Age of Siri

• Publication year: 2022
• Author: By Harold Cohen

WHO'S LISTENING? THE FOURTH AMENDMENT IN THE AGE OF SIRI

By Harold Cohen¹

INTRODUCTION

Innovation is the heartbeat of humanity, and the pace of our progress has hastened exponentially. In less than a century, the computing power that was once only available to top scientists in smoke-filled rooms trying to keep astronauts alive has been dwarfed by what most of us carry around in our pockets. Nowadays we even have computerized assistants. The introduction of voice recognition technology through platforms such as Apple's "Siri" and Amazon's "Alexa" has changed the way we live. This progress has led to innumerable improvements in our lives, but our advancement has come with a price.

The voice-recognition technology that powers our personal assistants is dependent upon a mixture of machine listening and artificial intelligence. Large amounts of data must be gathered, stored, and compared in order for these programs function and advance. This process, along with the rapid adoption of these devices by the public, has resulted in a massive amount of the population being listened to and recorded by at least one of these devices at all times. In order to aggregate the data gathered by these devices, users are asked to grant permission to relay their information back to companies. This constant sharing of information raises the question: What are the Fourth amendment implications of this process?

The Fourth Amendment does not apply when private companies interact with citizens. Therefore, when customers give consent for their devices to listen and communicate their data, no direct Constitutional issues arise. As a practical matter, this has resulted in a host of private conversations and activities being shared, many times without the knowledge of those being monitored. Further, the sharing of this information strips it of Constitutional protections.

Under the Fourth Amendment's so called "third-party doctrine" the constant monitoring and sharing of personal user information is not a problem. The third-party doctrine generally holds that when private communications are shared with third parties, they become available for the authorities to access without the need for a warrant.² The third-party doctrine is usually employed to access inadvertent disclosures by parties, but in these cases the disclosure isn't even inadvertent. It is explicit in the allowances often granted by the user at the point of purchase and buried in terms of service agreements. The third party doctrine, along with the endless terms of service agreements that users consent to, have combined to erode the Fourth amendment protections of the citizenry.

Concerned by this lack of protection, some companies, including IBM, have banned the use of voice recognition devices in their offices.³ Additionally, groups like the American Civil Liberties Union have posted warnings about possible dangers for all users that have enabled both the sharing function and voice recognition feature(s) of their device(s).⁴

Yet, despite the attention from companies and civil rights organizations this issue has yet to be fully explored. So far, only a handful of scholars have written about

[Page 14]

this technology's erosion of our Fourth Amendment protections.⁵ Even among those who have, they have offered little in the way of proactive solutions.

This paper seeks to add to the conversation by exploring the privacy implications of voice recognition technology, considering the available solutions, and introducing one its own. Part I provides a foundation for the analysis by explaining the current interplay of artificial intelligence and the third-party doctrine. Part II examines current and proposed solutions. Part III proposes a new solution by taking a proactive approach to the problem and focusing on user education. This proactive approach will allow users to make fully informed decisions when engaging with AI platforms.

ARTIFICIAL INTELLIGENCE AND THE THIRD-PARTY DOCTRINE: A MATCH

In order to understand the problem, one must first understand how both artificial intelligence and the Fourth Amendment third-party doctrine works. This section provides explanations of both. I start by exploring the inner workings of Artificial Intelligence. Then turn to an explanation of the third-party doctrine. Finally, I put the two together, examining what happens when the third-party doctrine overlaps with the working of Artificial Intelligence.

ARTIFICIAL INTELLIGENCE

Most Americans would recognize "Alexa" or "Siri" and are comfortable using these platforms.⁶ However, for something so ubiquitous it is surprising how few truly understand how AI works. There could be no AI without a process called "machine learning,"⁷ which enables devices to "learn" from their users. In order to "learn," each device records user data that is later shared with a main data collection center. Here data from all devices is analyzed, compared, and processed in order to advance the platform. Mass data collection not only makes AI possible, but is crucial to its performance.⁸

This process is generally the same on any machine-learning enabled device, and was acknowledged in early 2018 when, in a Congressional Hearing in which the VMware's⁹ CEO, Sanjay Poonen, was asked by Representative Jeff Duncan, "But that smart TV is monitoring all your viewing habits?"¹⁰ Mr. Poonen replied, "Exactly."¹¹

Mass data collection and machine learning may sound complicated, but in reality, it has become a part of our daily lives. For example, imagine you're on a road trip and you want something quick to eat. In order to keep your eyes on the road, you ask your "Siri" to search for the closest "Wendy's." Then, your phone starts dialing, you look down and see that "Siri" has called your ex-wife Wendy. Terror grips you as you quickly end the call, but still hungry, you ask "Siri" to find a McDonalds. The McDonalds is found easily, and you and "Siri" have no further interactions for the rest of the day.

You may think that all you've done is avoid an unpleasant conversation, but you've also advanced AI through machine learning. When you engaged with "Siri," the phone monitors your next actions to try and determine whether "Siri" properly responded to your request. When you suddenly changed your command, the phone recognized that the response given to you rendered a negative outcome-you ended the phone call before completing it. You then re-stated your command, this time for a similar restaurant that could not be confused with a first name, and the phone took note that no correction was made to "Siri's" second response.

This entire interaction will be shared with the main database that powers "Siri" in order to teach it that "Wendy" can both be a person's name and a restaurant. Now, the next time any "Siri" user asks for "Wendy's" the platform will not immediately call "Wendy." "Siri" will now try new responses to this command, sometimes performing different actions and sometimes asking follow-up questions for user clarification. For example, the next time you ask for "Wendy's" "Siri" may begin to list the people named "Wendy" in your contact list or ask a clarifying question as to whether you are looking for contacts or restaurants named "Wendy." These interactions will be shared with the database and once its error rate drops to an acceptable level (whatever that is coded to be),¹² the machine will have "learned" the proper response.

Conducted by independent devices, machine-learning is a painstakingly slow process of trial and error. However, when scaled, the pace of machine learning is limited only by the number of users it engages with. In this model, the user plays the role of both consumer and teacher. The AI is advanced by its exposure to new users and requests, and the user is rewarded by having access to an ever-advancing AI platform at no additional cost. The exponential growth in the everyday use of AI enabled devices has resulted in an exponential advancement in AI capabilities.

[Page 15]

THE THIRD-PARTY DOCTRINE

The third-party doctrine is a Fourth Amendment doctrine used to determine whether a search (as defined by the Fourth Amendment) has occurred when a subject has knowingly or inadvertently shared information with a third-party. The doctrine generally holds that if a subject has shared information with a third-party, then they have no Fourth Amendment protection for that information. This rule has evolved over time, but the basic principle has been upheld and still applies today.

The foundation of the third-party doctrine was set forth in the Supreme Court's decision in Katz v. United States.¹³ Katz held that whether a search as defined by the Fourth Amendment has occurred depends on whether a subject had a reasonable expectation of privacy.¹⁴ The decisions rendered defining when individuals have a reasonable expectation of privacy have helped to shape the third party doctrine.

The Supreme Court's decisions in both United States v. White¹⁵ and United States v. Miller¹⁶ planted the seeds of the third party doctrine. In White, the Defendant argued that his statements to a police informant should be suppressed.¹⁷ The Court held that there was "[no] indicat[ion] in any way that a defendant has a justifiable and constitutionally protected expectation that a person with whom he is conversing will not then or later reveal the conversation to the police."¹⁸ In Miller, the Court held that Defendant had forfeited Fourth amendment protections in documents he had voluntarily shared with the bank despite the documents being subject to the Bank Secrecy Act of 1970.¹⁹ The Court rephrased the holding in White, stating, ". . . the depositor takes the risk, in revealing his affairs to another, that the information will be conveyed by that person to the Government."²⁰ Both White and Miller solidified the Supreme Court's view that subjects assume the risk of information they divulge to third-parties ending up with the police.

In 1979, the third party doctrine came...