AI and the SEC: 6 Steps to Get Ahead of the AI Rule

Artificial Intelligence as an investment concept has exploded this year with the increasing commercialization of products like Chat GPT and Dall-E. As companies and consumers have adapted the new technologies, the SEC is taking notice and beginning to develop their regulatory posture, releasing over the summer a 239-page document outlining their current perspective on use of AI among investment professionals (yes, I read all 239 pages and consumed a few pots of coffee along the way)[i].

SEC Chair Gary Gensler has been making public comments about his concerns regarding the increasing use of artificial intelligence in world financial markets. Particularly, he worries that if many market participants are relying on the same artificial intelligence tools, it could spark a financial crash through herding effects and positive feedback loops (e.g., automated selling causes automated selling begets more automated selling and so on)[ii].

Other concerns raised by the SEC in the extensive rule document include:

• • Lack of human insight into why certain recommendations are being made by AI systems and lack of human oversight over the implementation of recommendations (e.g., automated trading systems or so-called “robo advisors”);
  • Conflict of interest if variables are being used in the system that would result in recommendations that benefit the firm at the expense of the client (e.g., optimizing for firm revenues rather than realized client performance);
  • Reliance on low quality datasets or datasets that could disappear or have significant errors;
  • Inappropriate use of personal identifying information in the modeling process; and
  • Use of AI to perpetrate fraud.

Preparing for New Regulations

Even if you do not use AI tools for modeling or processing investment decisions, it is still possible to fall under the new AI rules. For instance, using ChatGPT to summarize research on securities markets that is then used to make investment decisions or passed along to clients would seemingly fall within the definition of AI use outlined in the SEC’s rule document.

While no official rule has yet been adopted, there are some clear areas where investment professionals can begin to prepare for an AI regulatory regime. Adapting firm policies that address the following seem like a great first step:

1.  Describe how your company is using AI, especially where human beings have the ability to verify and override AI-sourced recommendations for clients.

 

2.  Describe how employees at your company may NOT use AI in interactions with clients.

 

3.  Identify potential or existing conflicts of interest related to the firms’ use of AI and steps/action plans to eliminate or mitigate those conflicts.

 

4.  Identify data sources used as inputs in artificial intelligence systems, including whether the data is licensed, contains personal identifying information, and any backup sources in the event that the primary source fails.

 

5.  Detail your disaster recovery and continuity plans in the event that a critical AI system fails and/or cannot be accessed for a certain period of time.

 

6.  Establish an AI oversight structure within your company if the technology is widely used, and particularly if it is used directly with clients.

Our View

BCM supports the SEC’s efforts to regulate use of AI in the financial sector insofar as those regulations protect individual clients without imposing an undue burden on investment professionals. We believe the current scope of oversight laid out in the rule document seems to achieve that balance.

Our company has been using AI to power our investment products for over a decade. We find ourselves somewhat amused with the recent hype over artificial intelligence, as many of these technologies have been around and in use for decades (see for example Arthur Samuel’s 1959 paper exploring the concept of neural networks[iii]). Some of these tools are presented to the public almost as if they are magic when, under the hood, they are ultimately implementing fairly standard mathematical operations albeit organized in an innovative algorithmic structure.

We caution against using AI tools for critical decision making when (1) the mathematics and methodology underlying those tools are not well-understood by the user and/or (2) the framework that the AI system uses for making decisions or recommendations is opaque. Stephen Wolfram wrote an overview of ChatGPT and large language models that is a great starting place for understanding how some of these new technologies work[iv].

We strongly caution against using the output of AI systems for making complex decisions (such as whether or not to invest client funds in a particular security) without any human oversight or verification.

That said, we are pleased that there is increasing awareness and adoption of a technology we see great promise in and are excited as new artificial intelligence applications spark our own creativity.

[i] Proposed Rule, Conflicts of Interest Associated with the Use of Predictive Data Analytics by Broker-Dealers and Investment Advisers. File No. S7-12-23 (July 26, 2023). https://www.sec.gov/files/rules/proposed/2023/34-97990.pdf.

[ii] Palma, Stefania and Patrick Jenkins. “Gary Gensler urges regulators to tame AI risks to financial stability.” Financial Times. October 15, 2023. https://www.ft.com/content/8227636f-e819-443a-aeba-c8237f0ec1ac.

[iii] Samuel, Arthur L. “Some studies in machine learning using the game of checkers. IBM J Res Develop. 1959 ; 3: 210–229. https://people.csail.mit.edu/brooks/idocs/Samuel.pdf.

[iv] Wolfram, Stephen. “What is ChatGPT doing…and why does it work?” February 14, 2023. https://writings.stephenwolfram.com/2023/02/what-is-chatgpt-doing-and-why-does-it-work/