A Year In, the Outlook for Generative AI in FS

Just over a year ago, ChatGPT launched. The excitement, anxiety and optimism associated with the new AI shows little sign of abating. In November OpenAI CEO Sam Altman was removed from his position, only to return some days later. Rishi Sunak hosted world leaders at the UK’s AI Safety Summit, interviewing Elon Musk in front of gathering of world leaders and tech entrepreneurs. Behind the scenes, AI researchers are rumoured to be close to even more breakthroughs. 

What does it all mean for those industries that want to benefit from AI but are unsure of the risks?

Some form of machine learning – what we used to call AI – has been around for a century. Since the early 1990s, those tools have been a key operational element of some banking, government, and corporate processes, while being notably absent from others.

So why the uneven adoption? Generally, that's down to risk. AI tools are great for tasks like fraud detection where well-established and tested algorithms can do things that analysts simply can’t by reviewing vast swathes of data in milliseconds. That has become the norm, particularly because it is not essential to understand each and every decision in detail.

Other processes have been more resistant to change. Usually, that’s not because an algorithm couldn’t do better, but rather because – in areas such as credit scoring or money laundering detection – the potential for unexpected biases to creep in is unacceptable. That has particularly acute in credit scoring when a loan or mortgage could be declined due to non-financial characteristics - including racial biases.

While the adoption of older AI techniques has been progressing year after year, the arrival of Generative AI, characterised by ChatGPT, has changed everything. The potential for the new models – both good and bad – is huge, and commentary has divided accordingly. What is clear is that no organisation wants to miss out on the upside. Despite the talk about risks with Generative and Frontier models, 2023 has been brimming with excitement about the revolution ahead.

Two Objectives

A primary use case for AI in the financial crime space is to detect and prevent fraudulent and criminal activity. Efforts are generally concentrated around two similar but different objectives. These are 1) thwarting fraudulent activity – stopping you or your friend or relative from getting defrauded - and 2) adhering to existing regulatory guidelines to support anti-money laundering (AML), and combatting the financing of terrorism (CFT).

Historically, AI deployments in the AML and CFT have faced concerns about potentially overlooking critical activity compared to traditional rule-based methods. That has changed over the last 5-10 years, with regulators initiating a shift by encouraging innovation to help with AML and CFT cases - declaring that innovators will be judged by their overall results not by some missed alerts.

However, despite the use of machine learning models in fraud prevention over the past decades, adoption in AML/CFT has been much slower with a prevalence for headlines and predications over actual action. The advent of Generative AI looks likely to change that equation dramatically.

One bright spot for AI in compliance over the last 5 years, has been in customer and counterparty screening, particularly when it comes to the vast quantities of data involved in high-quality Adverse Media (aka Negative News) screening where organisations look for the early signs of risk in the news media to protect themselves from potential issues.

The nature of high-volume screening against billions of unstructured documents has meant that the advantages of machine learning and artificial intelligence far outweigh the risks and enable organisations to undertake checks which would simply not be possible otherwise.

Now banks and other organisations want to go a stage further. As Generation AI models start to approach AGI (Artificial General Intelligence) where they can routinely outperform human analysts, the question is when, and not if, they can use the technology to better support decisions and potentially even make the decisions unilaterally.

AI Safety in Compliance

The 2023 AI Safety Summit was a significant milestone in acknowledging the importance of AI. The Summit resulted in 28 countries signing a declaration to continue meetings to address AI risks. The event led to the inauguration of the AI Safety Institute, which will contribute to future research and collaboration to ensure its safety.

Though there are advantages to having an international focus on the AI conversation, the GPT transformer models were the primary focus areas during the Summit. This poses a risk of oversimplifying or confusing the broader AI spectrum for unaccustomed individuals.

AI is not just Generative and different technologies provide a massive range of different characteristics. For example, while the way that Generative AI works is almost entirely opaque or "black box", much of the legacy AI can showcase the reasons for its decisions.

If we don't want to go backwards with AI panic, regulators and others need to understand the complexity. Banks, government agencies, and global companies must exert a thoughtful approach to AI utilisation. They must emphasise its appropriate safe, careful, and explainable use when leveraged inside and outside of compliance frameworks.

The Road Ahead

The compliance landscape demands a review of standards for responsible AI use. It is essential to establish best practices and clear objectives to help steer organisations away from hastily assembled AI solutions that compromise accuracy. Accuracy, reliability, and innovation are equally important to mitigate fabrication or potential misinformation.

Within the banking sector, AI is being used to support compliance analysts who already struggling with time constraints and growing regulatory responsibilities. AI can significantly aid teams by automating mundane tasks, augmenting decision-making processes, and enhancing fraud detection.

The UK can and should benefit from the latest opportunities. We should cultivate an innovation ecosystem with is receptive to AI innovation across fintech, regtech, and beyond. Clarity from government and thought leaders on AI tailored to practical implementations in the industry is key. We must also be open to welcoming new graduates from the growing global talent pool for AI to fortify the country’s position in pioneering AI-driven solutions and integrating them seamlessly. Amid industry change, prioritising and backing responsible AI deployment is crucial for the successful ongoing battle against all aspects of financial crime.