On Thursday, Morningstar launched a beta version of its artificial intelligence chatbot, which is designed to pool data from Morningstar’s extensive independent research platforms.
Named “Mo,” the generative AI tool is powered by the Morningstar Intelligence Engine and Microsoft Azure’s OpenAI Service, which is the technology behind Microsoft’s viral AI chatbot ChatGPT. Mo is available across Morningstar’s Research Portal, Investor, Direct, and Advisor Workstation platforms.
“AI has huge potential to support a better investor experience at scale, and this is just the beginning,” said Morningstar chief technology officer James Rhodes in a statement.
Morningstar first introduced Mo in April at its annual conference. Morningstar CEO Kunal Kapoor spoke with the chatbot on the main stage, and according to the company, conference-goers asked Mo more than 1,000 questions, on topics ranging from investing terms and saving for retirement to Morningstar’s take on emerging asset classes.
“Investors face an overwhelming amount of information amid a growing sea of more than 750,000 investment options. With the power of large language models and other AI tools, navigating large volumes of information can be made dramatically easier,” the company said in a statement. “Instead of working with keyword searches and pages of results, Morningstar customers can now direct their questions to Mo and receive a concise response within a matter of seconds.” However, the company added that Mo does not provide investment advice.
Mo is one of several applications that Morningstar plans to build on the Morningstar Intelligence Engine platform. Rhodes told RIA Intel in an e-mail that in the future, the company plans to use Mo to automate tasks, generate charts, and create tables. “[We] believe there’s a lot more that can come by thinking big about how our content and data can be delivered with the help of AI,” Rhodes said.
Early criticisms of ChatGPT centered around the bot’s tendency to pull data indiscriminately from the internet, regardless of source, which often resulted in inaccurate information being communicated as fact. Rhodes said they chose to use Microsoft’s Azure OpenAI Service because of its commitment not to allow user data to train the OpenAI large language model. “In keeping with Morningstar’s commitment to privacy, users are instructed not to input personal or confidential information into Mo.”
Mo also draws from Morningstar’s extensive content library of equity and managed investment research, editorial, and ratings. “[Mo] does leverage OpenAI’s large language model, gpt3.5turbo. But the difference is that it’s paired with a constrained content pool,” Rhodes said, although he added that Mo, like most AI, can generate inaccurate responses. “We are committed to innovating responsibly and taking steps to mitigate those risks relative to what people see in publicly available tools like ChatGPT.”
To mitigate any inaccuracies, Rhodes said that Mo also uses prompt engineering to set underlying rules and instructions. The company restricts the prompt character count, ensures that every session starts fresh without memory, and blocks Mo from providing financial advice.
The Morningstar statement also explained that responses are tested for relevance and responsiveness before being fed back to the user. “When Mo receives a user question, the Morningstar Intelligence Engine identifies the most responsive content to provide to OpenAI’s large language model, [so that it can] construct a response.”