When Artificial Intelligence and Big Data Collide—How Data Aggregation and Predictive Machines Threaten our Privacy and Autonomy

Abstract

Artificial Intelligence and Big Data represent two profound technology trends. Professor Alben’s article explores how Big Data feeds AI applications and makes the case that necessity to monitor such applications has become more immediate and consequential to protect our civil discourse and personal autonomy, especially as they are expressed on social media.

Like many of the revolutionary technologies that preceded it, ranging from broadcast radio to atomic power, AI can be used for purposes that benefit human beings and purposes that threaten our very existence. The challenge for the next decade is to make sure that we harness AI with appropriate safeguards and limitations.

With a perspective on previous “revolutionary” technologies, the article explains how personal data became profiled and marketed by data brokers over the past two decades with an emphasis on dangers to privacy rights.

The article observes that it is critical to adopt an approach in the public policy realm that addresses the bias dangers of a technology, while enabling a fair and transparent implementation that allows our society to reap the benefits of adoption. It advocates solutions to improve the technology and adopt the best versions, not cut off development in early stages of the new technology’s evolution.

Drawing on the author’s work as a state-level Chief Privacy Officer and a high-tech executive, the article concludes with four policy recommendations for curbing the flow of personal information into the Big Data economy: 1. Regulating data brokers; 2. Minimizing data by default; 3. Public Records Reform and 4. Improving personal data hygiene.

_______________

Introduction: Taking the Long View on New Technology

At the outset of a new decade, we have been promised that Artificial Intelligence (frequently abbreviated as “AI”) will solve a host of problems facing our society, ranging from economic disparity to political equity and social injustice. In parallel, we are also told that Artificial Intelligence will create a host of the problems, mapping to the same societal challenges. Both perspectives are correct.

This article examines the relationship between AI and “Big Data.” Specifically, it observes that without the fuel of data, the nascent AI industry could not possibly have grown as rapidly or proliferated into the myriad technical implementations we are witnessing today. It also poses the question of whether, as a society, we still have the ability to reign in the flow of Big Data before it truly enables profiling, tracking and prediction of individual human behavior to get out of control, creating more harm than good and threatening personal autonomy. The concluding section offers public policy approaches to curb the flow of the tidal wave of data that is being generated for corporations and data brokers to exploit.

Like many of the revolutionary technologies that preceded it, ranging from broadcast radio to atomic power, AI can be used for purposes that benefit human beings and purposes that threaten our very existence. The challenge for the next decade is to make sure that we harness AI with appropriate safeguards and limitations.

for decades. Anyone who has been a passenger on a jet plane, taken an Uber ride or made a mobile banking deposit has already utilized a form of AI. AI is widely used in e-commerce to identify shopping patterns, prevent fraud and predict future consumer needs.

Despite these widespread uses of AI, a quotient of fear has been introduced into the public discussion, especially in the realm of the threat that AI or even simple “algorithms” pose to civil rights. When Wired Magazine ran an article in 2018 citing an ACLU study of Amazon’s facial recognition software that erroneously matched 28 of the 435 members of the U.S. Congress with a database of law enforcement mugshots, civil liberties organizations alleged a strong racial bias in the Rekognition software, given that individuals with darker skin tones were twice as likely to be matched with the arrest database with a setting of an 80% confidence level. Amazon noted that other settings in Rekognition could correct for a 95% confidence level in results, yet both privacy advocates and computer scientists chimed in declaring that the software was too likely to make mistakes along racial lines.

Without follow-up, this type of article leaves the indelible impression that some types of AI technologies are inherently prone to error and could result in miscarriages of justice, especially in mistaken identification of suspects. Yet instead of public calls for transparency and better data, many commentators have jumped to the preemptive conclusion that AI for facial recognition should be indefinitely banned. Yet instead of public calls for transparency and better data, many commentators have jumped to the preemptive conclusion that AI for facial recognition should be indefinitely banned. This led to the City of San Francisco to declare a moratorium on the use of AI in 2019. Other jurisdictions have followed suit. A more recent and comprehensive study of the accuracy of facial recognition conducted by the National Institute of Standards and Technology showed that demographic factors do skew AI results and concludes that more caution is needed in the deployment of the technology for use in specific contexts.

The ultimate promise of accurate facial recognition technology, however, must be that it will prevent racial bias by focusing on the characteristics of individuals and not racial traits. Instead of using photography to confirm bias, AI holds out the promise of correctly identifying specific actors as a sorting tool for the implementation of public policy. In the midst of the brouhaha over moratoriums and claims of bias, this larger promise has been lost. This is why it is critical to adopt an approach in the policy realm that addresses the bias dangers of a technology, while enabling a fair and transparent implementation that allows our society to reap the benefits of adoption. We want to look for solutions to improve the technology and adopt the best versions, not cut off development in early stages of its evolution.

Whenever an accident occurs on a self-driving automobile or an algorithm is shown to result in an apparently unfair result, critics contend that all predictive technologies utilizing algorithms are deeply flawed and many contend they should be discontinued pending further study. In short, we are witnessing a frenzy in the technology world over a new set of technologies that we don’t completely understand, can’t easily define and have reason to both fear and respect.

At the beginning of the computer age in the 1960’s, we experienced a similar kind of ambivalence about the coming age where computers would make complex decisions for us, freeing us of painful labor, solving the energy crisis and transforming our economy. Yet today, we witness not only the rapid adoption of a new suite of applications utilizing AI, but the marriage of such technology to the explosion of Big Data. It is this combination that makes the issue of accurate AI especially salient and relevant to organizations seeking to deploy AI and to their customers and stakeholders.

As outlined in this article, the data aggregation industry grew in an unregulated fashion and has given rise to a lucrative data broker industry fueled by personal information. While such data profiles are largely used today in commerce, these individual data troves can also be harnessed by governments and other entities. Once created, they are difficult to control. Consequently, lawmakers, scholars and members of the public must become more conscious of the dangers of an unregulated data industry and seriously consider means to regulate the flow of data that will fuel AI applications going forward.

What Are We Talking About? The Challenge of Defining AI

A good working definition of Artificial Intelligence was floated over ten years ago by Stanford University Computer Science professor Nils Nilsson, a pioneer in the AI field:

“Artificial intelligence is that activity devoted to making machines intelligent, and intelligence is that quality that enables an entity to function appropriately and with foresight in

its environment.”

Writing in Forbes Magazine, Bernard Marr provides more historical perspective on the notion of defining what constitutes AI:

John McCarthy first coined the term artificial intelligence in 1956 when he invited agroup of researchers from a variety of disciplines including language simulation, neuron nets, complexity theory and more to a summer workshop called the Dartmouth Summer Research Project on Artificial Intelligence to discuss what would ultimately become the field of AI. At that time, the researchers came together to clarify and develop the concepts around “thinking machines” which up to this point had been quite divergent.7

With this useful working concept of a Thinking Machine, a further refinement of the types of AI is still desirable, given the panoply of technologies that are currently deployed or are on the proverbial drawing board. For the purposes of this article, an Artificial Intelligence program or application will include at least the following elements: 1. The ability to identify data, either through computer language or audio-visual and other “real world” inputs; 2. The ability to store data or seek out data from networked sources; 3. A logic function that allows the program to sort, filter and build hierarchies of data; 4. A machine learning algorithm giving the program the ability to make predictions and to change results based on past experience.

People who have used the United Airlines robotic voice assistant, Ted, have experienced a form of AI that can recognize human language and learn from aggregated chats how to direct consumer queries. PayPal, banks and other financial services use AI programs to detect patterns in commerce that suggest credit card fraud.8 We are not talking about the types of AI on display in sci-fi movies such as 2001: A Space Odyssey, where the computer HAL seeks to take over a space mission,9 although such types of sophisticated programs may become real in our lifetimes.

Given the evolving status of the AI industry, it’s interesting how quickly we have come to expect perfection from thinking machines. We seem to live in an environment where every mistake made by a robot or automated vehicle resulting in human injury is widely chronicled and publicized, leading the public to mistrust new technologies that appear to be held to “zero tolerance” standards.10 Yet another dimension of data is not simply quality, but the sheer number of inputs. The balance of this article focuses on the “bigness” of “big data” and poses whether size itself can result in societal problems when such pools of data are harnessed by AI applications.

How Data Got “Big”

All AI machines benefit from the new world of Big Data, because thinking machines need data in both training and operation. A thinking machine starved of data will not become smart. The proliferation of data types has gone hand in hand with the evolution of computers and the Internet. Embedding cameras in cell phones about 15 years ago has given rise to the creation of more photographs each year than were taken in the previous history of photography.11 Current estimates for the total number of web sites exceed 1.5 billion, and Google has indexed at least 4.45 billion individual web pages.12 GPS technology has given rise to tracking data for anyone who keeps their cell phone location data turned on—and even limited tracking when the phone is off.13

We have come to tolerate vast aggregations of data as a necessary byproduct of the Internet and connected devices. In the early days of the World Wide Web, companies worried about the cost of data storage. The advent of cloud services has truly served as a game changer in making Big Data even bigger.14 Cheaper and cheaper storage has driven the cost of retaining data to zero or below, creating a paradox that it is now cheaper for most firms to keep data than to delete it. If true, this will only accelerate data proliferation.15

Artificial Intelligence technologies have existed for decades, yet only in the past ten to fifteen years have they been married to pools of large data, enabling them to accomplish both useful and invasive tasks.

In 2011, IBM’s Watson computer made headlines by defeating two accomplished Jeopardy Champions, Ken Jennings and Brad Rutter, in a three-game match. The playful “Smarter Planet” logo that viewers saw on television masked ten racks of IBM Power 750 servers sitting in a separate room. When Watson cogitated host Alex Trebeck’s questions, wavy green lines animated his “face.”

Unlike many AI applications that can process natural language, Watson did not actually listen to Trebeck, but received his inputs via text messages that translated the host’s verbal questions. However, like his two human contestants, Watson had to frame his questions in terms of answers and had to “buzz in” to gain priority to answer correctly. He did so with stunning accuracy, despite the fact that he had no Internet connection as many viewers assumed. In fact, Watson had been fed over 200 million pages of data, ranging from sports to entertainment trivia. By besting Jennings and Rutter by over $50,000 Jeopardy

dollars, Watson claimed the one-million-dollar tournament prize.

Watson’s AI architecture had taken IBM scientists over three years and thousands of practice rounds to develop. According to TechRepublic:

IBM developed DeepQA, a massively parallel software architecture that examined natural language content in both the clues set by Jeopardy and in Watson's own stored data, along with looking into the structured information it holds. The component- based system, built on a series of pluggable components for searching and weighting information, took about 20 researchers three years to reach a level where it could tackle a quiz show performance and come out looking better than its human opponents.17

Watson’s victory sparked widespread interest in AI technologies, even though Watson itself would be further developed by IBM for commercial applications. Health insurer Wellpoint and The Memorial Sloan Kettering Medical Center began utilizing Watson for health care problem solving by 2012.18 To do so, Watson not only was put online, but had to learn to properly ingest medical taxonomies and two million pages of medical data from over 600,000 sources. The value of using this type of AI to sort through presentations of facts and make diagnoses had not been lost on the oncologists at Sloan-Kettering. While used only as a backup to human diagnosticians, Watson came to make accurate diagnoses and was able to incorporate patient history and DNA testing in devising its predictions of what course of treatment would be appropriate for cancer patients.

Watson has also been adapted for use in the finance industry and for customer service applications. It has already been tested by some banks that use Watson to recommend financial services to customers. Sites developed by law firms have trained Watson’s Natural Language functions to help answer basic legal questions. The platform is even finding its way into retail settings, attempting to influence the course of consumer purchasing decisions. Like all powerful technologies developed before it, AI will also be harnessed for entertainment and less than life-or-death human pursuits. Edge Up Sports, a Fantasy Football start-up, for example, has employed Watson to give its Fantasy Fans better data recommendations than they might otherwise develop by consumer sports stats on their own.21

We can surely expect AI platforms, such as Watson, to be trained to tackle challenges in healthcare, finance, customer service, and myriad verticals over the next decade. With human supervision and backup, these powerful programs might improve diagnostic accuracy and speed up critical processes that make a difference in human lives. At least that’s the promise of AI, yet the collision of the AI and Big Data train have already led to the challenges of technology quickly getting out of control, akin to the metaphor of opening Pandora’s Box.

Big Data Meets Social Media Platforms

One of the first pools of data harnessed by AI algorithms has been the personal information that billions of users of Twitter, Google, Amazon, Microsoft and other leading Tech platforms have generated in pursuit of communications and social interaction. The irony that users don’t adequately value their own data has not been lost on a growing number of economists and legal scholars, who have suggested that the fundamental business models of these social networks is flawed from a consumer perspective. The defining paradox of the age of social media may turn out to be that while each user is willing to trade their most personal data for free software, the value of such data in aggregate is exploited by some of the most profitable enterprises that the world has ever seen.

At the moment, TikTok stands out as a prominent example of an application that utilizes algorithms and vast troves of user data to create a compelling entertainment product. Owned by a Chinese parent company, ByteDance, the app has raised the ire of the Trump Administration due to potential data sharing of user information with the Chinese government. TikTok’s global reach is difficult to dispute. As of July 2020, the song and dance-centered social application had over 689 million users across the planet and has been downloaded over two billion times. In the U.S., it has reached approximately 100 million monthly users, an increase of over 800% in two years.22

While the Trump administration did not cite any specific instance of user data transferred to Chinese authorities, the company has been criticized for violating Google’s Android Platform privacy policy regarding the capture of user device MAC addresses. In August of 2020, the Wall Street Journal outlined how TikTok violated the Android policy:

TikTok skirted a privacy safeguard in Google’s Android operating system to collect unique identifiers from millions of mobile devices, data that allows the app to track users online without allowing them to opt-out, a Wall Street Journal analysis has found. The tactic, which experts in mobile phone security said was concealed through an unusual added layer of encryption, appears to have violated Google policies limiting how apps track people and wasn’t disclosed to TikTok users. TikTok ended the practice in November, the Journal’s testing showed.23

Clearly, the potential exists for consumer data to find its way from a TikTok consumer’s device to a parent company and then to entities that are not identified in TikTok’s end-user license agreement. Highly popular with children and teens, the specter of unauthorized collection and transfers of user data also raises serious questions relating to children’s privacy, compliance with COPPA and similar legislation in the European Union and other jurisdictions. The fact that user-generated videos also feature images of TikTok users, also creates possibilities of utilizing facial recognition and other identifiers, so that a person appearing in an apparently harmless homemade karaoke may in fact by spotted and tracked for nefarious purposes.

Like any software product, TikTok is powered by algorithms, which lie at the heart of the struggle between the perceived interests of the American government and the autonomous operation of ByteDance, TikTok’s owner. Based on a blog post by the company, Wired Magazine reported a few details of how the “For You” function works, determining which videos a user will see in their TikTok app:

When a video is uploaded to TikTok, the For You algorithm shows it first to a small subset of users. These people may or may not follow the creator already, but TikTok has determined they may be more likely to engage with the video, based on their past behavior. If they respond favorably—say, by sharing the video or watching it in full— TikTok then shows it to more people who it thinks share similar interests. That same process then repeats itself, and if this positive feedback loop happens enough times, the video can go viral. But if the initial group of guinea pigs doesn’t signal they enjoyed the content, it’s shown to fewer users, limiting its potential reach.24

While this type of affinity algorithm has been in use for decades, suitors for the company apparently consider it to be of great value and have insisted that ownership of the algorithm be part of the resolution of an acquisition of the company. Similarly, Facebook has created controversy and criticism by weighting user “News Feeds” with factors that stimulate polarized views on political issues, although we lack a definitive scale of what type of speech might be purely factual or neutral. Nevertheless, critics of Facebook decry the “Filter Bubbles” that determine exposure of content to its user base and argue that its AI technology has disrupted political discourse and caused societal harm:

Where Facebook asserts that users control their experience by picking the friends and sources that populate their News Feed, in reality an artificial intelligence, algorithms, and menus created by Facebook engineers control every aspect of that experience.

Facebook can draw upon the user profiles and activity generated each second by Instagram and Facebook and WhatsApp platforms, allowing it to slice and dice data in ways that cater to user preferences and are driven by user behavior on the site. In its first five years, when activity primarily revolved around the social experiences of its college-age members, Facebook attracted little criticism. Once the “Like” button was added in February of 2009, the company gained a powerful input to understand user behavior and serve content more closely tailored to the needs of its advertisers.26 “Likes” fueled the growth of the platform from 400 million to over two billion global users, illustrating how a smart company can harness the potential of Big Data.

Information that a user provides Facebook isn’t limited to elements such as likes, posts and photos, but can include the location metadata inside photos, and even what is seen through the camera in its apps. Facebook uses a person’s address book, call log or SMS log to suggest people that the user may know. The company can collect a user’s phone number and additional information from other people uploading their contacts.

Whenever possible, Facebook logs each individual’s phone’s battery level, signal strength, even available storage. On a computer, Facebook logs a user’s browser type and its plugins. It also tracks whether a window is in the foreground or background, and the movements of a mouse. While Facebook can obtain location data when provided access to GPS, the company doesn’t stop tracking an individual’s location when they turn off location services. It also tracks location from other data points, including IP addresses and nearby Wi-Fi access points and cell towers.28

Facebook also gathers information about other devices that are nearby or “on your network.” The policy says it is to make it easier, for instance, to stream video from your phone to your TV.29 Because Facebook provides proper Amazon not only knows a person’s purchase history, but how often they leave your shopping cart open. The public may be aware of these practices, but very few consumers object or even dial down the privacy settings made available by these companies.33

This level of comprehensive data gathering is akin to the “mosaic theory” of Fourth Amendment surveillance law, which holds that an isolated photo or video might not constitute an intrusion or a “search,” yet stringing together multiple photos or videos could most likely yield a detailed account of an individual’s pattern, habits and life.34

The harm created by broad data profiling is not the collection of random data points about a person, but the aggregation of such points to paint a complex profile of an individual and her history and predilections.

This is the danger of Big Data when harnessed to AI programs such as machine learning.

How Our Data Became Brokered

Social media platforms acquire even more information about their users from data brokers such as Acxiom and Oracle. These firms specialize in the collection of data and monetize it through sales to marketers, to both traditional firms and online. One might observe that the White Pages published by local telephone operators was an early form of data collection and in the pre-digital era, records were regularly compiled by the county clerk to measure births, deaths, mortgages and property sales. Yet before the advent of widely available search algorithms, the collection and analysis of these written records required considerable investment of time and labor. With the advent of Big Data, the data broker industry struck the mother lode. Disparate databases could be harnessed together. Different data types could be sorted, analyzed and recombined. Individuals could be tracked across hundreds and even thousands of data repositories. Acxiom compiles data on individuals to track their: religion, health interests, alcohol and tobacco consumption, banking relationships, social media usage, medical insurance, size and type of home, family size and likelihood of having another baby, loans, income, personal net worth, relationship status, media consumption, political views and, of course, age, gender, education and employment.35

Other leading data brokers include Oracle, Experian, Trans Union, Lifelock, Equifax, Moody’s and Thomson- Reuters. Each systematically gathers personal information from public sources, purchases other personal data from private sources and monetizes their different user profiles to meet the needs of data-hungry customers. To date, over 400 firms have identified themselves under the data broker provisions of California’s new privacy law. When Vermont passed the first data broker registry law in 2018, over 120 data brokers eventually signed up and paid the $100 registration fee.36 While it is difficult to arrive at a single definition of “data” or “information” broker, it is thought that the global industry rakes in between $200 and $300 billion annually.37 Thus, personal data has become the fuel for a secondary market that simply trades in a widely available resource—our personal data.38

Four Solutions to the Big Data Problem

The operation of the data broker industry is perfectly legal in the United States and data brokers have had no obligation until recently to even tell an individual what data it has collected about them. (In January of 2020, California’s new privacy law established a right to request access to personal data from certain firms, including data brokers.) AI sits atop the iceberg of Big Data. Whether we are talking about image recognition or other forms of machine learning, data-hungry AI machines thrive on

ingesting data to form and perfect their ability to navigate the “real world” and to solve problems they are trained to solve. One could even go so far as to posit that the AI industry could not exist in a meaningful way without huge depositories of data on which to be trained and produce results. In the perfect storm of this new decade, the AI industry doesn’t need to worry about finding data, just as a 6th grader doesn’t have to worry about finding an online source for a book report.

Can Big Data Be Controlled?

Like King Canute attempting to hold back the advancing tide, proposals to curb Big Data might strike most people in the technology industry as whimsical or quixotic. There are practical measures, however, that can influence the course of data creation and data retention which we need to more fully explore in order to ascertain when they are effective in giving users more control over their personal information:

1. RegulationofDataBrokers

With regard to the regulation of data brokers, efforts are underfoot in several states to identify data brokers and create more consumer transparency around their practices. Vermont became the first state to pass a data broker registration statute in 2018. The new Vermont law defines a “data broker” as a business that collects and sells personal information from consumers with whom the broker has no direct relationship. Thus, the Vermont law begins to address “third party” data mining (that is, data mining by companies that have no direct relationship with consumers).39

In the wake of its ground-breaking new privacy law, the California Legislature also passed a bill requiring data broker registration at the end of 2019.40 Other states are considering data broker registration. When coupled with the CCPA’s requirement that a consumer can request access to and deletion of data from a company that garners more than half of its revenue from the sale of personal information, California has now jumped the queue in terms of transparency.

Transparency is the fundamental principle in this regime. Once people are aware of the actual practices of data brokers and how those practices impact their personal lives, they may act to curtail certain types of data sharing and surely will become more sympathetic to legislative efforts, perhaps even a national law, to restrict data broker practices that freely trade their personal data without the need to seek their consent.41 Third party data sharing provides the oxygen for the fire of the data industry, further enabling applications that make predictions about user behavior.

2. Deleting Content by Default

Companies love to keep data, yet if our society is to bring Big Data under control, the collection and retention of

data should be purposeful and driven by either an identified company need—which could be monetization—or consumer benefit.

Companies keep data because they can and the machines are set by default to chronicle records of time spent on site, pages visited, pages hovered over and other metrics. While such data can be useful for analysis of consumer behavior, that does not justify the retention of all data from all users.

Keeping all data for indefinite periods of time poses hazards for companies, opens them up for claims of unjustified tracking and contributes to the big data problem. Leading private sector actors with sophisticated engineer programs clearly are capable of doing better and doing more with limited data sets, yet their feet have never really been held to the fire, either by regulators or the public at large. With the growing recognition of the lead of data to unintended places and growing legislative calls for data scrutiny, now would be a good time to begin by shifting the default setting to the deletion of data and to conscious and transparent justification for data retention.

3. Public Records Reform

In several states, public records acts have become the third rail of politics and unintentionally have contributed to the flow of data from public records to data brokers and other commercial actors. Originated in the 1970s, most public records acts seek to spread “sunshine” in the workings of government by mandating that administrative records be retained for set periods of time and be made available to the public at little or no cost.43 Newspapers and broadcasters rely on public disclosure to uncover news about the workings of state and local government and to conduct investigations relating to individuals. No one questions this legitimate use of the public disclosure system. Yet, as described below, other actors have utilized the mountain of public data for their own uses with no regard to the broader public interest. Unfortunately, efforts at public records reform to address specific abuses and abusers, are usually met with criticism that such reform aims to limit “the public’s right to know.” With the evolution of Big Data, this type of argument has become increasingly divorced

from the reality of how public records are actually consumed.

In the era of file cabinets and clerks, the flow of public records relating to real estate transactions, births and deaths and criminal records grew at a reasonable pace.

However, as in the wider industry, the advent of new data types such as audio, video, GPS and social media communications, vastly exploded the scope of public records, with most courts holding that a new data type qualified as such.44 Further, the evolution of online search made these records readily available, both to interested members of the public and to commercial entities seeking to scoop up data on a routine basis. Much of the data industry, in fact, relies on “scraping” public databases for millions of individual records.45 The brokers then combine this public information with personal data gleaned from other sources, creating rich and more economically valuable profiles.

Commentators have pointed out that the long tail of personal data often results in distorted profiles for individuals, especially people who have gone through the criminal justice system.46 Further, the compilation of public records has enabled bad actors to utilize the public records act to harass victims, such as estranged spouses or people whom they bear a grudge against.47

This was not the intent of the 1970s sunshine era laws, yet the proliferation of data created a wave of information that has swamped the resources of states and local government.

A modest and targeted reform of the Public Records Act could achieve the goals of limiting the flow of data while preserving the rights of journalists and members of the public. First, a public records requester should have to state a reason for the request, such as their personal need to find information about another individual. The recordskeeper can then have some basis for evaluating a request and flag suspicious requests. Second, data brokers should have to purchase data from a government according to an approved agreement. Many of these exist in the realm of departments of motor vehicles for vehicle data and such agreements should be duplicated across other state functions, such as taxation and business records. Third, states should increase penalties for those seeking commercial use of public records and proactively try to stop such requests when they are made. For example, a request for “all addresses of homeowners in a water district,” should give rise to suspicion that the requestor is seeking the information for a commercial purpose. A request to see the recorded video of a woman’s

movement in and out of a public building should be treated with suspicion, especially in the context of a sexual harassment inquiry. At present, most state laws place the burden on the government to establish that the request is suspect.

Many of the new data types created in public and private sectors are “transient.” For example, GPS location data from a state vehicle is captured and recorded, but not necessarily kept by the wireless carrier or intermediary for longer than a day or two. However, strict reading of public records statutes calls for the retention of such data. One logical reform would be to more carefully define and examine so-called “transient” data. Limiting the mandatory collection of transient data will protect the privacy interests of such civil servants, while also narrowing the funnel of data that has contributed to the overload and increasing misuse of public data by private actors with motives that do not meet the legislative goals of public transparency.

Finally, retention periods for public data should be reviewed. Most of these retention periods were set in a pre-digital pre-search era, where there was a greater justification to keep data around longer. If data retention periods for public records can be shortened across the board, those seeking timely news and information will be favored As it stands, the system rewards massive scoops of public data that end up in deep personal profiles.

4. Personal Data Hygiene

A promising way to limit data collection is for individuals to develop better “data hygiene” by taking simple steps to filter the personal information shared with third parties. Five simple suggestions for the average digital device user:

1. Turn off location services in phone settings (and in other devices) when a specific application is not being used. Don’t worry, you can always turn the location service back on when needed.

2. Don’t allow “contact sharing” between applications. When prompted to share your contact or email list with a new program, follow Nancy Reagan’s time- honored advice: “Just say no.”

3. Dial back advertising settings in your major social media platforms such as Twitter, Instagram, Gmail and Facebook. All of these platforms have “Privacy Settings” tabs, allowing the user some degree of control over data sharing with third-party advertisers.

4. Limittheuseofthird-partycookies.Youcandothis by going into the settings of your web browser and moderate the dropping of cookies on the browser by third parties. It’s also good hygiene to wipe all cookies after long periods of time. This may disrupt your log-in of seldom used sites, but that is the tradeoff.

E. Finally, exercise one’s privacy rights. The new California Privacy Law, the CCPA, allows you to request a company to tell you the data they have gathered about you, delete some of that day and to opt out of data sharing with third parties. Each company with either $25 million in revenue or 50,000 consumer names doing business in the state of California is now obligated to provide a prominent “Opt Out” button on its website. Any California resident can avail themselves of this window into the data collection practices of myriad firms.

Conclusion

Where Do We Go From Here?

At the outset of this new decade, we find ourselves standing at the intersection of two incoming trains—the explosive growth of Big Data and the rapid development of AI technologies. Will we get caught in this intersection or will we figure out as a society how to harness both trends and use them for the benefit of our culture, economy and planet? In order to achieve the later outcome, our leaders need to thoughtfully define AI without clouding the debate with erroneous fear. Consequently, we need to ask the precise question of how best to implement AI technologies with a view toward enhancing our civil rights and promoting economic progress. At this early stage, it’s also appropriate to ask: Who should be framing these questions and making these decisions?

Perhaps this suggestion borders on a truism, but it behooves those interested in addressing this question to think about how we might assemble the best minds and forward-looking thinkers on this topic, drawing not only from the tech world, but from civil society leaders, sociology, economics, politics, law and pure sciences. Coping with a “brave new world” of AI is akin to Huxley’s dystopian novel.48 We can either take concerted action to understand and productively apply the new world of AI, or we will find ourselves flattened by the steamroller of technology under the guise of “progress.”

Members of the legal profession have a special obligation as custodians of data, because we are viewed by society as “arbiters of truth” and our 21st-century truths increasingly rely on models drawn from data. If the quality of a conclusion is only as good as the quality of the data inputs to be interpreted by an algorithm, then attorneys trained in concepts of evidence and civil procedure must energetically exercise their powers in this role of determining what facts constitute admissible “evidence” for a given case. As any law student taking Evidence might observe, admitting a fact into evidence is not a simple matter as facts must run a complex legal gauntlet before they find themselves before a jury or trier of fact.

As crafters of many of the algorithms that increasingly determine public benefits and detriments, computer scientists and engineers also should feel a heightened sense of responsibility for the outputs of their work. It is inherently difficult for an individual to recognize his or her own “bias” with respect to a matter, even an element of an algorithm that appears to be neutral on its face. Yet the inclusion of certain sets of “neutral” elements can sway a result one way or another. A data point that includes an individual’s age, location or education is not so much a single marker as a collection of aggregate facts. Parsing such facts poses a great challenge for the data scientists of the next decade as they train AI programs to incorporate information into their programs. Finally, public policy makers must avoid the temptation to jump to conclusions based on news reports or incomplete

studies about the nature or track record of AI programs. As discussed in this article, AI will be imperfect so long as data inputs are flawed or incomplete. AI will yield inaccurate results so long as its programmers incorporate biases, both hidden and overt. Yet the promise of AI to improve human decision-making and to crunch data at scales not possible by humans cannot be ignored, as so many of our global problems, ranging from water shortages to climate change to a quest for better energy sources cry out for us to harness all of the tools in our arsenal of thinking, including AI.

At the close of World War II, the advent of a new technology proved that it could both extinguish humanity and perhaps also benefit the world through peaceful harnessing of atomic fission.49 Seventy-five years later, we are still engaged in debate as to how to balance the destructive power of nuclear arms and nuclear waste with the benefits to society. Artificial Intelligence is not unlike atomic power in that respect, as we are only at the beginning of the long journey toward wisely incorporating AI into our decision-making processes. Just as the incorporation of the computer took decades to integrate into our working systems and personal lives, we should recognize that this perilous road will be marked by both triumphs and mistakes. Utilization of bad data could lead to catastrophic consequences for humans and for our environment. Allowing AI control over vital functions that might be manipulated in ways that promote human suffering and disparity could result in damage to people that cannot be reversed. To that end, readers of this article must reflect on the roles they might play to control, moderate, and influence the evolution of this powerful technology, treating it with the awe and gravity that it deserves.