Report on Artificial Intelligence (AI) in the Securities Industry | June 2020 1 Contents Introduction 1 SECTION I: Overview of Artificial Intelligence Technology 2 Key Components of AI Applications 4 SECTION II: AI Applications in the Securities Industry 5 Communications with Customers 5 Investment Processes 7 Operational Functions 8 SECTION III: Key Challenges and Regulatory Considerations 11 Model Risk Management 11 Data Governance 13 Customer Privacy 15 Supervisory Control Systems 16 Additional Considerations 18 Request for Comments 20 JUNE 2020 Introduction Artificial Intelligence (AI) technology is transforming the financial services industry across the globe. Financial institutions are allocating significant resources to exploring, developing, and deploying AI-based applications to offer innovative new products, increase revenues, cut costs, and improve customer service. 2 First developed in the early 1940s, AI technology has gained significant momentum over the past decade and become more mainstream due in part to the availability of inexpensive computing power, large datasets, cloud storage, and sophisticated open-source algorithms. In a recent survey-based report, executives at financial institutions noted that “AI is expected to turn into an essential business driver across the Financial Services industry in the short run, with 77% of all respondents anticipating AI to possess high or very high overall importance to their businesses within two years.” 3 Broker-dealers are exploring and deploying AI-based applications across different functions of their organizations, including customer facing, investment, and operational activities. In July 2018, FINRA solicited comments from the industry on the potential challenges associated with using and supervising AI applications at broker-dealer firms. 4 In response, commenters recommended that FINRA undertake a broad review of the use of AI in the securities industry to better understand the varied applications of the technology, their associated challenges, and the measures taken by broker-dealers to address those challenges. Based on this feedback, FINRA, through its Office of Financial Innovation (OFI), engaged in an active dialogue with the industry over the past year and held meetings with over two dozen market participants, including broker-dealer firms, academics, technology vendors, and service providers to learn more about the use of AI in the securities industry. A REPORT FROM THE FINANCIAL INDUSTRY REGULATORY AUTHORITY Artificial Intelligence (AI) in the Securities Industry 1 1 This paper is not intended to express any legal position and does not create any new requirements or suggest any change in any existing regulatory obligations, nor does it provide relief from any regulatory obligations. While this paper summarizes key findings from FINRAs outreach and research on the use of AI applications in the securities industry, it does not endorse or validate the use or effectiveness of any of these applications. Further, while the paper highlights certain regulatory and implementation areas that broker-dealers may wish to consider as they adopt AI, the paper does not cover all applicable regulatory requirements or considerations. FINRA encourages firms to conduct a comprehensive review of all applicable securities laws, rules, and regulations to determine potential implications of implementing AI-based applications. 2 PwC, Crossing the Lines: How Fintech is Propelling FS and TMT Firms Out of Their Lanes, 2019, PDF (reporting that financial services executives expect their AI efforts to result in increased revenue and profits (50%), better customer experiences (48%), and innovative new products (42%).). 3 Ryll et al., Transforming Paradigms: A Global AI in Financial Services Survey, Jan. 2020, PDF. 4 FINRA, Special Notice on Financial Technology Innovation in the Broker-Dealer Industry, July 30, 2018, ArticleReport on Artificial Intelligence (AI) in the Securities Industry | June 2020 2 This paper summarizes key findings from FINRAs review in three sections: 0 Section I briefly defines AI and its scope as it pertains to the securities industry for the purposes of this paper. 0 Section II provides an overview of broker-dealers use of AI applications related to: (i) communications with customers, (ii) investment processes, and (iii) operational functions. 0 Finally, Section III discusses key factors including potential regulatory considerations, securities market participants may want to consider as they develop and adopt AI-based tools. The discussion below is intended to be an initial contribution to an ongoing dialogue with market participants about the use of AI in the securities industry. Accordingly, FINRA requests comments on all areas covered by this paper. 5 FINRA also requests comments on any matters for which it would be appropriate to consider guidance, consistent with the principles of investor protection and market integrity, related to AI applications and their implications for FINRA rules. I. Overview of Artificial Intelligence Technology Definition The term artificial intelligence broadly refers to applications of technology to perform tasks that resemble human cognitive function and is generally defined as “the capability of a machine to imitate intelligent human behavior.” 6 AI typically involves “the theory and development of computer systems able to perform tasks normally requiring human intelligence, such as visual perception, speech recognition, decision-making, and translation between languages.” 7 John McCarthy, one of the founders of AI research, “once defined the field as getting a computer to do things which, when done by people, are said to involve intelligence.” 8 Scope 9 While the definitions for AI discussed above provide a general outline of the meaning of the term, there is no single universally agreed upon definition of AI. In practice, AI is used as an umbrella term that encompasses a broad spectrum of different technologies and applications, some of which are described below. 0 Machine Learning (ML) Machine learning is a field of computer science that uses algorithms to process large amounts of data and learn from it. Unlike traditional rules-based programming, ML models 10 learn from input data to make predictions or identify meaningful patterns without being explicitly programmed to do so. There are different types of ML models, depending on their intended function and structure: 5 See Request for Comments section on page 20 of this paper. 6 Artificial Intelligence, Merriam Webster, Article. 7 Artificial Intelligence, Oxford English Dictionary, Article. 8 Shukla Shubhendu updates on social media and other public websites; browsing history on the firms website and mobile apps; and past communications (e.g., from emails, chat messages, and meeting notes). All this information is analyzed using AI tools to provide the registered representative with a broader picture of customer needs, along with tailored suggestions of what investment products the customer may be interested in. Industry participants indicated that registered representatives use this information to augment their existing knowledge and expertise when making suggestions to their customers. Industry participants noted taking a cautious approach to employing AI tools that may offer investment advice and recommendations directly to retail customers, citing several legal, regulatory, and reputational concerns. 17 0 Customized research Firms have also indicated growing use of AI tools to provide curated market research directly to customers to share relevant information on various investment opportunities. For example, as noted in the earlier section, AI-based tools may offer customers social media data and related sentiment analysis on investment products and asset classes. While these AI tools offer the potential to customize investment suggestions for customers, firms should be cognizant of potential concerns and challenges related to data privacy, use of corrupt or misleading data, and adapting to each customers unique circumstances. 18 Portfolio Management and Trading Broker-dealers are also exploring and using AI applications within their portfolio management and trading functions. 0 Portfolio management Within portfolio management, firms noted the use of AI applications to identify new patterns and predict potential price movements of specific products or asset classes. These applications tap into vast amounts of data available from internal and external sources, including from non-traditional sources like social media and satellite imagery, which is used as proxy data for economic activity to identify insights that may signal price movement. Some broker-dealers that are also investment advisors aim to incorporate these predictions into their investment strategies to generate alpha for the portfolio. 17 In the U.S., digital investment platforms (commonly referred to as “robo-advisors”) that offer investment advice directly to clients via online platforms, currently largely use rules-based models to develop those recommendations. See FINRA, Report on Digital Investment Advice, Mar. 2016, PDF. Firms may wish to review their AI-based investment tools to determine whether related activity may be deemed as offering discretionary investment advice and therefore implicate the Investment Advisors Act of 1940. 18 FINRA, Social Sentiment Investing Tools Thing Twice Before Trading Based on Social Media, Apr. 2019, ArticleReport on Artificial Intelligence (AI) in the Securities Industry | June 2020 8 0 Trading Securities industry participants are also exploring AI tools to make their trading functions more efficient by maximizing speed and price performance. Examples include using ML for smart order routing, price optimization, best execution, and optimal allocations of block trades. Firms should bear in mind that use of AI in portfolio management and trading functions may also pose some unique challenges, particularly where the trading and execution applications are designed to act autonomously. Circumstances not captured in model training such as unusual market volatility, natural disasters, pandemics, or geopolitical changes may create a situation where the AI model no longer produces reliable predictions, and this could trigger undesired trading behavior resulting in negative consequences. In addition, some industry participants have expressed concern that AI trading models across the industry may start to learn from each other, potentially leading to collusive activity, herd behavior, or unpredictable results. 19 Operational Functions Compliance and Risk Management In conversations with FINRA staff, industry participants noted that they are spending significant time and resources in developing AI-based applications to enhance their compliance and risk management functions. This is consistent with FINRAs 2018 research on the use of regulatory technology (RegTech), where we observed that “market participants are increasingly looking to use RegTech tools to help them develop more effective, efficient, and risk-based compliance programs.” 20 According to an April 2018 research study conducted by Chartis Research and IBM, which surveyed more than 100 relevant risk and technology professionals, 70% of respondents noted using AI in risk and compliance functions. 21 Broker-dealers have to keep pace with complex and evolving domestic and international regulations, as well as a rapidly changing risk landscape (e.g., cybersecurity, internal threats, and financial risks). At the same time, they now have access to vast amounts of data, inexpensive computing power, and innovative technologies that present opportunities for them to develop automated compliance and risk-management tools. Below are some examples that firms shared of how they are incorporating AI in their compliance and risk management tools. 22 0 Surveillance and monitoring AI technology offers firms the ability to capture and surveil large amounts of structured and unstructured data in various forms (e.g., text, speech, voice, image, and video) from both internal and external sources in order to identify patterns and anomalies. This enables firms to holistically surveil and monitor various functions across the enterprise, as well as monitor conduct across various individuals (e.g., traders, registered representatives, employees, and customers), in a more efficient, effective, and risk-based manner. Market participants noted that these tools could significantly reduce the number of false positives, which in turn, free up compliance and supervisory staff time to conduct more thorough reviews of the remaining alerts, resulting in higher escalation rates. Firms indicate that these tools offer the ability to move beyond “traditional rule-based systems to a predictive, risk-based surveillance model that identifies and exploits patterns in data to inform decision- making.” 23 For example, some firms noted the use of AI-based surveillance tools to monitor 19 R. Jesse McWaters and enable human users to understand, appropriately trust, and effectively manage the emerging generation of artificially intelligent partners.” 35 Data Governance Data is the lifeblood of any AI application. AI applications benefit from large amounts of data to train and retrain models, conduct comprehensive analyses, identify patterns, and make predictions. Accordingly, the quality of the underlying dataset is of paramount importance in any AI application. Industry participants noted that one of the most critical steps in building an AI application is to obtain and build the underlying database, such that it is sufficiently large, valid, and current. Depending on the use case, data scarcity may limit the models analysis and outcomes, and could produce results that may be narrow and irrelevant. On the other hand, incorporating data from many different sources may introduce newer risks if the data is not tested and validated, particularly if new data points fall outside of the dataset used to train the model. In addition, continuous provision of new data, both in terms of raw and feedback data, may aid in the ongoing training of the model. 34 Several research studies have noted that data mining can lead to incorrect or misleading results because of the identification of spurious correlations. See, for instance, Hou, Kewei and Xue, Chen and Zhang, Lu, Replicating Anomalies, Working Paper No. 2017-03-010 presented at the 28th Annual Conference on Financial Economics and Accounting, Fisher College of Business, June 12, 2017, Article. 35 Matt Turek, Explainable Artificial Intelligence (XAI), Defense Advanced Research Projects Agency, ArticleReport on Artificial Intelligence (AI) in the Securities Industry | June 2020 14 When reviewing and modifying data governance policies and procedures to address potential data- related risks that may emerge in AI