It probably won’t come as a surprise that quants have begun to play around with ChatGPT, OpenAI’s new chatbot, which combines information available online with artificial intelligence to answer questions and explain things.
In the process, they’ve noticed features and bugs that have already been well documented by others, such as ChatGPT’s inaccurate answers and its potential to totally disrupt search engines and more. But they’re also trying to figure out how to leverage ChatGPT, along with Google’s “Language Model for Dialogue Applications” (LaMDA) and other similar tools, to improve their businesses.
At PanAgora Asset Management, the $44 billion Boston-based quantitative investment firm, that includes the time-intensive task of filling out reams of questions for institutional investors trying to decide between asset managers. The quant firm has long used natural language processing — a branch of AI — in its research, but George Mussalli, chief investment officer of equity investments at PanAgora, told Institutional Investor that the firm has recently begun experimenting with other machine learning tools.
Allocators’ requests for proposals, or RFPs, ask asset managers many of the same questions. However, RFPs have gotten longer and have begun to ask more specific questions. PanAgora has a team dedicated to the completion of RFPs, and while Mussalli doesn’t see AI replacing them, he does think it could meaningfully improve a necessary part of being an asset manager.
“I want to use it not necessarily to save time, but to give [us] a better answer,” he said.
Mussalli recalled a colleague asking him about all the papers he had written about ESG. He remembered a lot of them, but when he began to compile a list, he rediscovered some relevant ones that he had forgotten. He’s hopeful that a chatbot will not only eventually help PanAgora remember its own research and responses, but will also help it identify the most relevant ones for an RFP, and do it faster than someone can say the acronym.
Richard Tan, managing director and head of stock selector equity investments at PanAgora, is impressed with ChatGPT and other tools — he likened them to an elementary school student who can read at an incredible rate and never forgets anything. He said that with some training (for example, by allowing the bot to review all of PanAgora’s research papers or responses to questions in the past) he thinks they could become a powerful, useful addition to the RFP team.
Other asset managers were less willing to talk about whether or how extensively they were using chatbots or similar tools to help them complete RFPs. But it’s clear that there’s growing interest in the new technology.
“A lot of firms are exploring it. I’m not sure whether there are a huge number of firms that have experimented or adopted it yet,” said Russell Andrews, global head of advice solutions at SEI Asset Management. Nevertheless, he feels that the use of chatbots and other tools could be valuable to both asset managers and allocators.
While the technology is still in its infancy, it’s probably a good bet that other asset management companies will follow PanAgora’s lead, and not just to ease the laborious process of completing RFPs. Allocators could also use AI to analyze the RFPs they get back and, based on the answers, begin their evaluations with the most relevant responses. Andrews also feels that using software to tackle those tasks might help keep biases toward certain managers or strategies in check. For example, a lesser known manager’s RFP, which might normally be one of the last to be considered, could be bumped to the top of the stack.
Like Mussalli, Andrews doesn’t envision chatbots totally replacing the people at asset management firms who work on RFPs. Canned “copy and paste jobs,” even human ones, can be obvious, he said.
“I think it’s fairly early but, like everything, technology evolution moves very quickly these days,” Andrews said. “I think more asset managers [than you might imagine] are exploring these kinds of concepts. And I think it could develop pretty quickly.”
But PanAgora is tempering its excitement. Like any other tool, chatbots come with their own advantages and limitations. They can be used the wrong way and, like market signals, require knowledge and testing. Users must be able to ask a lot of the right questions, and they must also be aware of confabulation, a situation in which the chatbot begins to make up details to help fill out its responses, Tan said.
“It might not be right, it might not be perfect, but it’s a start,” Mussalli added.
