The Education Advisory Group hosted a virtual panel discussion on October 5, 2023, discussing Artificial Intelligence (AI) and Machine Learning in Investment.
The panel featured three distinguished leaders in this emerging field:
- Peeyush Shukla, CFA, Chief Information and Technology Officer and a member of the Global Management Committee at Heitman, LLC, a global real estate investment management firm. In this role, he is directly responsible for leading Heitman’s Information Technology initiatives and Infrastructure Group in support of Heitman’s global investment management business. Working with the Executive Leadership team and business unit heads, Shukla defines and implements strategies that ensure the firm’s technology and systems are able to support the firm’s business objectives.
- James Boudreault, CFA, Global Head of Data Science at CME Group. In his role, he leads the department that provides analytics relating to customers, products, and industry landscapes in which CME Group operates. He is responsible for data science and developing AI-enabled client-facing tools. He has expertise in data, analytics, cloud technologies, leadership, innovation, and economic thought leadership.
- Lee Davidson, CFA, chief analytics officer for Morningstar, Inc. In his role, he leads the global analytics group responsible for Morningstar’s analytics products and services. Davidson sets the company’s AI and analytics vision and roadmap, aligning with the firm’s strategic business goals.
The panel was moderated by Monique Thanos, PhD, CFA, FRM, a Director at Lukka working in crypto. The following presents a synopsis of the questions and answers during this session.
What is the current state of AI in investments and how have you implemented AI?
Peeyush noted that his firm has built a data platform for AI and machine learning based on Microsoft Azure. They have been working to extend the use to predictive learning on occupancy and other important metrics in real estate using generative AI.
Lee noted that Morningstar collects significant data at scale and have been using machine learning to generate data insights. They have also used generative AI for some content creation.
How would you create a technology framework for AI?
Peeyush emphasized the need to start with data as it is the backbone of an AI model. Next you should pick a strong partner to support your efforts. There are many cloud computing offerings, but Heitman selected Microsoft Azure to provide the data storage and computing ecosystem. Then they selected workstreams to feed the models and then selected the tools to use, broken down between supervised and non-supervised AI. Their initial use case surrounded decision trees and regression trees to analyst their investments.
Can you provide any examples of AI in use?
James noted a recent case of an NFL game that was broadcast using AI to transform the players into characters from Toy Story and how that broadcast enabled him and his children to watch and enjoy the game together. At the CME, they are using AI to develop tools to examine and benchmark execution costs and suggest potential actions by clients to reduce slippage costs. He also noted that some banks are using AI for mortgage applications to suggest items to address to improve the chance of approval. In a similar way, they are using AI to guide clients in trading patterns to improve overall costs.
Lee noted that the human research function at Morningstar could not effectively scale to meet the growth in the market. By utilizing a human supervised AI model, they could apply their research and rating model to a larger universe. The result was a 10X increase in analyst throughput.
Do you think AI is more about productivity enhancement or cost reduction?
James noted that AI should really be known for “Augmented Intelligence” since it makes humans more productive, making people faster and with better insights to enhance productivity.
Peeyush noted he thought it was more about productivity while Lee thought that while it could reduce cost, there was also an optionality in better productivity solutions. When thinking about AI eliminating jobs, we need to start thinking about jobs as groupings of tasks and we are using AI to evolve those tasks.
What would you do to improve your AI skills?
James noted that he used to tell people to learn to code in Python or a similar language, but now he encourages people to be curious and embrace AI. Try out new things and network in the space.
Lee thought that AI skills can assist in being a translation bridge, or effective communicator, between the business needs and the technology.
Peeyush suggested becoming familiar with cloud computing and the related ecosystem, begin querying data and applying algebra and statistics to gain insights.
How do you measure success in your AI efforts?
Peeyush noted that his firm looks for levers for how they generate investment alpha. They look at repetitive, predictive tasks where AI and machine learning can add value. These are opportunities to achieve new capabilities that would not be possible without AI. They also look to leverage AI in innovation.
James offered that CME is leveraging AI on optimization problems such as evaluating which clients are the priority for 1-on-1 interaction and which could be handled more digitally. Using AI to prioritize their actions.
How can you communicate the benefits to those who are still skeptical of AI?
Lee urged listeners to tell success stories as too often communications focus on the cons and not the pros of AI. We also need to build faith in the process and justification for using AI. In addition, using explainable AI is important for building that faith – turning the black box to gray and eventually to white. Using decision science for model validation can also help.
James reiterated the need to explain models to build trust. He noted two approaches to implementation, the bottom-up approach of using initial efforts on smaller tasks to show the value of AI. The top-down approach was essential for long-term success when the CEO takes the lead and begins prioritizing the effort to drive adoption throughout the organization. He cited the importance of a customer-centric approach. They see the value of AI in the 10X ideas that increase speed, with better information to make better decisions. Finally, he noted that using AI to replace tedious, repetitive tasks often leads to increased employee satisfaction as team members can focus on more valuable work.
Peeyush emphasized the need to align with the business-critical use case was essential. Natural use cases that align with the needs of the business. His firm uses generative AI to assist in generating fact sheets and presentations.
Can you share any personal AI stories?
James used it for Google searches on healthcare, which often provides thousands of links but few answers. Instead, he uses ChatGPT to wade through the data to find answers faster. He has also used it to determine starting lineups for his fantasy football team.
Lee takes a novel approach of using ChatGPT to create bedtime stories for his kids. He also developed an algorithm several years ago for fantasy football and won all six leagues he joined, but then he failed to update the algorithm and did not win at all the next year.
Peeyush noted that he uses AI for testing his ideas and thoughts to validate them.
What is the future of AI?
James proffered that the future of AI and work would involve everyone having their own chatbot, rather than using the company’s chatbot. This would evolve into hiring not just the person, but the person and the chatbot, so as individuals we would focus on maximizing our combined value.
Peeyush emphasized the importance of how you train the models using human knowledge to enhance outcomes, but you need to focus on the data.
Lee thought that AI would be most effective in separating the signals from the noise in investment management and as a result, sifting through data would be more interesting.
How do smaller firms with fewer resources remain competitive?
The group noted that machine learning and AI are becoming services of cloud providers making them more affordable. With off-the-shelf solutions, start-up costs are decreasing, and AI offers the opportunity to work with open source material so smaller companies can do more with less.
What are the risks of the use of AI?
Lee noted that his firm developed a beta version of an AI digital research assistant. He listed three steps they took to mitigate the risk: 1) implement guardrails on what keywords the AI was allowed to use, and which were prohibited; 2) limiting data sources; and 3) forcing the AI to cite sources for the information used.
Peeyush suggested opening a limited use case with a small data set to mitigate potential risks.
James suggested starting with low risk use cases to train the model and implementing human checks and human reviews of the output.
Are there particular programs that might be useful?
James suggested using AI called “Codie” to learn how to use Python, referring back to his curiosity point earlier. Many times it might provide an incorrect answer but it can help narrow a starting point. AI combined with Human Intelligence can be a powerful combination, but there must be a human in the loop.
The program concluded with comments by CFA Society Chicago President and CEO Chris Vincent on the CFA Institute Research Foundation work on AI and big data and the Society’s commitment to bringing additional programming on these topics to benefit our members.