AI roles vary across the industry from firms looking to upgrade technology and create platforms that AI tools can be built off to hiring chief technology officers, chief information officers and heads of AI.
“There's a whole amalgamation of different titles that are out there,” she said.
'Not just the quants'
Gary Collier, chief technology officer at $178.2 billion Man Group, likes to think about technology through concentric circles with AI as the broadest circle representing anything that can mimic human intelligence, then on to machine learning, deep learning and other subcategories including large language models.
While Man Group has been working on AI for over a decade, Collier said the last two years have brought changes.
“What has made the last year or two noticeably different is it’s not just the quant business with deep interest in generative AI … there is an enormous amount of interest from discretionary businesses,” he said, adding that other areas including operations and legal are also interested in the tools. “It’s not just the quants at the leading edge of technology this time.”
Man Group employs approximately 500 people in technology roles, and technology teams focused on AI and machine learning sat in the quant space.
When Robyn Grew took the helm as the firm’s CEO last year, a dedicated machine learning technology team that cuts across divisions was created and Collier stepped into the new CTO role while Tim Mace serves as the head of data and machine learning.
They’ve worked to build out Man GPT, the firm’s version of ChatGPT, an interface anyone at the firm can use. And an enablement team of engineers works to look at use cases and experiments with teams across the business, he added.
Collier points to use cases including using chatbots to talk with a wide array of data and documents, foreign language translations for documents such as Japanese to English and coding co-pilots that suggest lines of code. Collier said they’ve seen approximately 20% of suggested co-pilot lines of code get used.
“We are talking about co-pilots, augmentation, not sure anyone is using the term autopilot … it’s very much human augmentation, that’s where I see the edge,” he said.
From 2 days to 30 minutes
Charlie Flanagan, Balyasny’s head of applied AI, said there’s been an evolution over the last 15 months, showing both the opportunities and limitations with new generative AI tools.
“AI ML (machine learning) as a term has evolved,” he said, adding “there’s a large focus on these general purpose, large language models and what they can actually do for us.”
Balyasny had a natural language understanding team for six years. As ChatGPT came on to the scene, the firm got a privately hosted version of the model and immediately ran a hackathon last year to see different use capabilities, Flanagan said. Since then, the firm has built its own ChatGPT-like system.
Flanagan and others saw an opportunity to form a centralized team to work across investment and business functions and the applied AI team was created about nine months ago. The team has 14 members, with an even split between researchers and engineers, and is focused on building tools and platforms and educating people, including running more hackathons and trainings.
“We’re really focused on helping empower everybody in the firm to leverage AI in their everyday work,” he said. “That’s the mission of our team.”
And Flanagan argued now is the moment for everyone, experts included, to explore new methodologies and use cases. Balyasny has over 1,000 people on an internal chat sharing use cases, papers and tools, he added.
Flanagan pointed to several use cases, including one that is saving Chris Pulman, head of macro research and chief economist for the macro team at Balyasny, days of work. Previously, Pulman would spend two days a month putting together a central bank preview report to send to investment teams.
With help from the applied AI team, Pulman wrote code and has generative AI tools go through a range of sources from speeches of bank officials to the firm’s own proprietary views and extract relevant information and then put together a report that now only takes 30 minutes.
AI is also helping investing teams ask better questions, with the applied AI team working with investment teams to create AI-generated reports. Flanagan said in one case, the system suggested an analyst ask about GLP-1 drugs, gym memberships and related spending and this helped shape an investment strategy.
'Custom problems'
When AQR’s head of machine learning, Bryan Kelly, thinks about AI, he is focused on portfolio research and views it as “very heavily parameterized prediction models.”
“Organizationally, the way we think about AI is these are tools that are general purpose tools. They can be valuable for a lot of our portfolio research processes,” he said.
But asset management faces a lot of “custom problems that we have to solve, and that these are not problems that can be solved with off-the-shelf methods, typically. So we need to have capabilities to solve these problems ourselves, and that's what the machine learning team focuses on,” he said.
AQR formed its machine learning team in early 2019 and a big area of focus has been using machine learning to extract information from textual data sources, Kelly said. That along with pre-trained language models have become “table stakes.”
But Kelly added that training custom language models for asset management is key.
“That last mile of the portfolio construction problem using large language models or other preexisting AI, is where all the finance expertise needs to come in. That’s where the differentiation is going to happen,” he said.
Kelly has also been spending time exploring the theoretical behaviors of AI models, which he said can save an enormous amount of resources and guide how to structure models.
Needed skills
Korn Ferry’s Egol expects that in a few years, AI skill sets will be part of an individual's experiences vs. a stand-alone function. “It’s going to be so intrinsic in what a quant does, or what the head of data or technology does, you don’t really want to kind of keep it as a stand-alone,” she predicted.
Salaries for heads of AI run the gamut from $550,000 to $750,000 to packages that hit $1 million and up, Egol said. With the expenses involved, firms are having to decide whether to bring talent in-house — and if so, where to place them and how to support them — or outsource aspects of AI.
Man Group’s Collier said hiring bright people who are passionate about technology is still key. And he advises people who want to go into finance to learn the programming language Python.
He scoffs at predictions that software engineers will not be necessary because of generative AI.
AQR’s Kelly advises that students should “develop a balanced expertise between technical machine learning, statistical skills and economic expertise, grounding in theoretical economics, theoretical finance. These two together are the ideal combination for success as a quantitative researcher in the asset management industry,” he said.
Kelly added that “there are surprisingly few people” who have both technical and economic skills at “a high degree of sophistication” and that the combination is “the home-run recipe for being a successful quantitative researcher.”
Kelly expects there will be “a ramp up” in the industry in expertise on the machine learning side, but “it's going to flatten out in the not-too-distant future.”
Balyasny’s Flanagan said people who can adapt are critical and everyone needs to be learning some elements of AI, and he encourages everyone to take an engineering and coding class, especially Python. “I don't think that engineering goes away. I just think that the most tedious parts probably get a little bit easier,” he said.
Flanagan said he sees AI tools as replacing tasks, especially "the most monotonous or the most frustrating in a lot of cases," but not replacing jobs.
A revolutionary moment?
Man’s Collier said a view may be emerging that the peak of initial hype cycle around generative AI has passed and that some people may be disappointed that there is no “killer application.”
He argues that is the wrong way to look at it since generative AI will be broad and incrementally additive to many parts of an organization.
“The technology is very additive in a lot of different places as opposed to a single wow, killer knockout app,” he said, adding he expects it to become “table stakes.”
Balyasny’s Flanagan thinks the current moment in time is a “platform shift” akin to the internet in the late 1990s and the introduction of the Apple iPhone.
“I do think this moment in time, we'll look back on and say that that was the moment that things changed,” he said.
