Cracking the AI Code

By Dan Perlin, Alex LaPlante, Sal Vella, Susom Ghosh
Published April 14, 2023 | 4 min read

How can financial institutions harness the potential of artificial intelligence and make their investments pay off? Experts from the frontline of this technology shared their perspectives at RBC Capital Markets’ Global Financial Institutions Conference.

Key Points:

  • Collaboration between humans – both tech specialists and business experts – is key to building viable AI products for enterprises.
  • As the technology’s potential evolves at pace, it’s critical to tackle risk and ensure transparency before deployment.
  • Recent developments in ChatGPT and other language models have powerful potential for business in areas such as content generation.
  • Large enterprises need to invest early and heavily to avoid being left behind; business leaders should engage personally to develop their understanding of AI.

How to crack the code and deploy AI at scale

“A lot of large organizations invest heavily in AI, but never see the benefits,” noted Alex LaPlante as she articulated the challenges faced by institutions seeking to exploit this exciting, but expensive and risky, vein of technology.

As Interim Head of Borealis AI, a research and development institute at RBC, LaPlante believes her team has “cracked the code” of how to operationalize AI at scale. She identified some of the keys to success, including the “often overlooked” central element of AI deployment: human input.

“AI is inherently human-centric. You can’t build a great AI product without people,” she explained to the attendees at the conference in New York.

While tech teams were important, they needed to work in collaboration with domain experts within the business, as well as regulatory and privacy specialists. This minimized the risk of experts creating “really cool technology in search of a problem.”

“You need to focus on the client from the very beginning and really understand their pain points,” LaPlante said. Programs often began on a narrow basis, she added, but matured after a year or two as technologists became fully aligned with business needs.

“A first step is translating technical performance metrics to business metrics that meet clients’ needs,” LaPlante Said. For example, “what does 95% accuracy mean in a chatbot? Is that good or bad? It depends on where it fails.”

“AI is inherently human-centric. You can’t build a great AI product without people.”

- Alex LaPlante, Interim Head, Borealis AI

 

Tackling risk and ‘explainability’

Panel moderator Dan Perlin, who is also the Payments, Processors and IT Services Analyst at RBC Capital Markets, acknowledged that the topic of AI is controversial as well as timely. Some of the potential risks were outlined by Sal Vella, Vice President, Technology & Operations Strategy and Innovation at RBC: fears about bias and security, IP issues and the tendency of some AI programs toward “hallucination.”

With the potential applications of AI evolving “every week and every day,” the RBC team is careful to explore all the risks between the development of a program and its use, said Vella. “We’re at the very forefront of implementation, in terms of proof of concept,” he said. “The question is, when do we deploy?”

Under a concept called Responsible by Design, said LaPlante, developers consult with experts to explore mitigation of bias in any new idea. “Explainability” is also critical – finding a way for the end users to comprehend what an AI program is doing and to trust its output.

This was key to the launch of Aiden, RBC’s AI-powered trading platform.

“We’re dealing with traders; you have to convince them you’re giving them a better end outcome; one they can understand.”

- Susom Ghosh, Global Head, Quantitative Sales Training & Strategic Initiatives, Digital Solutions & Client Insights, RBC Capital Markets

Aiden is powered by reinforcement learning, a more dynamic form than the supervised and unsupervised learning that has trained previous forms of AI. Ghosh drew an analogy between reinforcement-led AI and the Mario character in a Super Mario Bros game. Mario might play the game poorly at first, but “let him play it 1,000 times and not only will he learn how to beat the game, but he’ll also figure out the most efficient way to beat the game.”

 

The power of language models for business

The panel hailed the potential of language models such as OpenAI’s ChatGPT. Near-term applications include summarization and content generation, said Vella, while the technology’s ability to write code would dramatically change the software development world.

Within the past few months, he said, services such as Bing AI have used the language power of GPT-3 technology to add new data sources, such as graph databases, to search facilities.

“When you ask a question, it parses it out and then figures out where to get the data,” Vella explained. “That’s where it becomes really powerful and useful for enterprises.”

For Ghosh, one of the most important uses is the ability to accelerate the generation of content by sales and trading staff and “not taking away the human element from the process but making the outcome more scalable.”

One note of warning on verifying the output of language models was sounded by LaPlante. “What’s really interesting about large language models is that they can lie so convincingly that even for a human to discern when they’re lying is quite challenging. That’s a big hurdle that we will have to get over,” she warned.

 

Invest early – but focus on solving big problems

Asked by Perlin if financial institutions had the understanding and fortitude to get behind AI development for the duration required for success, the panel was largely positive.

A technologist by background and new to RBC, Vella said he had been pleasantly surprised.

“Curiosity is encouraged at all levels. When I go in and pitch, I’m welcomed with open arms.”

- Sal Vella, Vice President, Technology & Operations Strategy and Innovation, RBC

He said business leaders should be engaging personally with accessible technologies such as ChatGPT, to help them understand the use cases for AI.

LaPlante emphasized the importance of assessing not only whether an AI project was feasible, but also whether it would be worthwhile. “AI is very expensive to develop, deploy and maintain; you’re only going to benefit if you focus your top talent on the big, meaty problems that are most important to your business,” she said.

But she urged enterprises to invest early and heavily: “The pace of change in technology is only accelerating. My guess is that will create gaps between the people who have invested and those who haven’t.”

This content is based on commentary and analysis from RBC Capital Markets' Global Financial Institutions Conference hosted in New York, NY on March 7-8 2023. For more information about the conference, please contact your RBC representative.


Dan Perlin

Dan Perlin
Payments, Processors and IT Services Analyst, RBC Capital Markets


Alex LaPlante

Alex LaPlante
Interim Head, Borealis AI


Sal Vella

Sal Vella
Vice President, T&O Strategy and Innovation, RBC


Susom Ghosh

Susom Ghosh
Global Head, Quantitative Sales Trading and Strategic Initiatives, RBC Capital Markets


AI/MLAidenArtificial IntelligenceBorealis AIMachine LearningTechnology