Balancing Innovation and Oversight Key to Leveraging the Benefits of Gen AI 

Generative artificial intelligence (Gen AI) and Large Language Models (LLMs), the engines that power it, have raised interest in artificial intelligence (AI) to a whole new level. Their ability to respond to a user request and generate human-like responses has captured the imagination of all industries, including Canadian pension funds. The world has since entered a rapid state of flux, trying to understand how best to harness this revolutionary technology while being sensitive to the potential data security, legal and regulatory questions that it may raise.   

Canada is the birthplace of modern AI, thanks to Geoff Hinton and Yoshua Bengio, who laid the groundwork with their research into neural networks and are often referred to as the ‘godfathers of AI’. These networks were architected to model the biological neuron structure and function of the human brain. Following this pioneering research, interest in deep learning and LLMs has grown significantly, resulting in the current rapid transformation of AI.

Today, organizations across the world are utilizing AI daily, and many, including pension funds, are beginning to explore the limitless power of Gen AI. They are looking into how Gen AI can help sort through vast amounts of data to improve reporting, information collection and process automation to make workdays more efficient.  In the future, Gen AI may also have the potential to optimize pension fund investing.

Large Language Models Explained

LLMs are a new form of AI, known as generative AI (Gen AI), due to their ability to create (generate) content. They are based on Deep Neural Networks (DNNs) and utilize a specific neural “transformer” architecture which is the key to their ability to predict and generate content. This is different from past forms of AI which were typically used to forecast a value, classify some data or detect an anomaly. An LLM’s ability to ingest a human generated question or prompt and create a relevant response based on the question’s intent makes it unique in AI. It is exactly this capability that has prompted many business leaders, including heads of pension funds, to analyze the potential impact and commit to the strategic development of this technology.

While LLMs hold tremendous potential, concerns regarding over reliance on this technology have arisen.  As capable as LLMs are, they do not possess human abilities like general reasoning, planning or cause and effect determination. LLMs in their current form are not capable of human-level intelligence and lack true understanding. Human oversight remains a critical component of the use of Gen AI. 

The Power of Generative AI in Financial Services

Gen AI has the power to change the face of every industry. From a broad perspective, there are three potential areas of impact for financial services: automation, acceleration and augmentation. Gen AI can help automate operational processes, accelerate productivity and efficiency and augment information gathering. For example, an asset manager could use Gen AI to filter market data that aids them in their investment decision making process. Alongside market data, Gen AI can combine esoteric datasets such as sentiment analysis or keywords searches that it has deemed to be relevant. It can unearth hidden trends or ‘black swan’ events, providing a unique view of the investment horizon faster than had been previously possible. In the shorter term, Gen AI will most likely support existing research and data gathering and will not be a driver of decisions.

Gen AI has potential within fund analytics as well. Traditional ex-ante risk models or performance attribution may be overtaken by Gen AI that follows continuous learning paths to tweak and optimize models based on investment style and market events, making the models more relevant to a pension fund’s investment strategy.

How Pension Funds Can Use Generative AI

Like many investors, pension funds have been actively researching the many applications of Gen AI. One way they are considering using it is to sort through vast amounts of data. Gen AI has the ability to take disparate data sets and organize them into dynamic reports. This can allow pension funds to rethink their reporting and make it more customizable. It can help them quickly sift through data input and user preferences to create tailored reports and summaries that meet individual needs. 

Pension funds are also looking to Gen AI for process automation, plan administration and back-end tasks. This has the potential to make pension funds more efficient and reduce the need for manual, time-consuming processes. Automating internal processes can also make the onboarding experience faster for plan sponsors and participants.

In addition, Gen AI has the potential to support the investment decision making process, allowing investors to shift from being intuition-based to data-driven. Given Gen AI’s ability to understand human generated text, the world of unstructured data, such as documents, emails and social media, all become new sources of insight for the investment process. Though there are already ways to optimize pension portfolios, most methods rely more on historical data and focus on comparing returns with the level of investment risk. Gen AI could refine this process. With inputs based on an asset manager’s style and investment philosophy, Gen AI could suggest asset allocations overlayed with an investor’s investment and ethical policies to assist with portfolio optimization. 

Data Oversight and Security Concerns

While Gen AI has many benefits, it has raised data oversight and security concerns. AI hallucinations, in particular, are one such concern. Hallucinations are non-sensical or incorrect outputs generated by AI systems. These inaccuracies can range from factual errors to biased or misleading information. This potential misinformation has become a paramount concern for organizations and regulators, especially if the Gen AI use case involves high levels of risk, such as investment decision-making.

Data privacy and protection is yet another concern. AI systems ingest and leverage vast amounts of data, making security a crucial focus, particularly when involving sensitive information. 

Due to the possibility of inaccurate outputs and security concerns, organizations would be well served to implement appropriate AI rules and guidelines, with a monitoring program to ensure the continuous accuracy of LLM responses. And as this technology continues to advance and permeate the industry, oversight regulations may be implemented, focusing on the transparency and accountability of AI systems.

Balancing the Power of Gen AI

Gen AI is a disruptive technology with the potential to transform industries around the world. The financial services industry, for one, continues to experiment with its many applications. Pension funds are beginning to explore how it can be used to improve reporting, data analysis, and operational processes. In these early days, the complex and powerful nature of LLMs necessitates careful data security and oversight considerations. As the industry considers how best to harness the power of Gen AI, a balanced approach that combines innovation with attention to oversight will be essential.

^{1 Three AI godfathers, two of them Canadian, selected for the 'Nobel Prize of Computing' | Financial Post}

^{2 Generative AI could be next big thing for defined contribution plans | Pensions & Investments (pionline.com)}