AI's game-changing role in managing content in the finance sector

To the point

• Large Language Models (LLMs) like ChatGPT will improve natural language information processing in the financial services industry, providing significant competitive advantages. 

• LLMs employ Natural Language Processing (NLP) and neural networks to understand human language and perform tasks, such as translation, summarization, paraphrasing, question-answering, and natural language generation.

• Potential financial services applications include knowledge gathering, automated authoring, content transformation, (summarizing or translating), and insightful content analysis.

• Challenges to adopting NLP in financial services include technical challenges, ensuring reliability and security, navigating regulation, and managing changes in existing processes and employee roles.

• Despite these challenges, the role of NLP in financial services content is poised to be game-changing, and with the right balance of technology and human insight, the future looks promising.

Information is a key asset and resource in the financial services industry. The ability to obtain better, faster, and more accurate information brings competitive advantages, as information exchange underpins all parts of the industry, including financial markets. Yet, while there has been a revolution in quantitative data tools, such as algorithmic trading, a revolution of the same magnitude relating to natural language information has not yet occurred.

Natural language information is written or spoken information in any human language, pertinent examples being an annual report, a meeting transcript, a clinical trial report, a news article, a prospectus, or a training video. The challenge has been that, until very recently, the ability of machines to understand natural language has been poor, and therefore we have relied on humans for the production and analysis of this information.

However, the advent of Large Language Models (LLMs), such as ChatGPT, presents the stunning possibility to overcome human-scale limitations and exponentially change the magnitude, speed, and accuracy of human-language information processing.

There are significant incentives in the financial services industry to embrace these technologies quickly, to gain competitive advantage and boost productivity. Below, we set out, at a high level, some of the potential opportunities, but also considerations and challenges of adopting these new technologies.

To many, ChatGPT may have appeared to burst onto the scene from nowhere. It is seriously impressive, but it’s worth understanding its context. The ChatGPT application uses artificial intelligence known as Natural Language Processing (NLP). NLP is a field of science that has been around for many decades but has made incredible progress in the past five years due to the switch from statistical models to self-learning neural networks.

When trained on large volumes of high-quality data, neural networks learn to ‘make sense’ of human language and can perform several tasks. These tasks include relatively simple actions, such as keyword and named entity recognition, tagging, text classification, and parsing, through to more complex actions, such as translation (Neural Machine Translation or NMT), summarization or paraphrasing, as well as the significantly more difficult tasks of question-answering and natural language generation. It is the progress in these last highly complex tasks, achieved by training on vast datasets with billions of parameters, that have “completed the set” and generated excitement (not to mention, launched a few thousand VC pitches).

Furthermore, NLP is increasingly combined with Automatic Speech Recognition and synthetic voice for text-to-speech, speech-to-text, and speech-to-speech. With the right training data, it is now theoretically possible to build almost any language-related application using the blocks described above, presenting significant benefits in the financial services industry. These could be applicable for the following:

• Knowledge gathering from a much greater variety of sources, in any language or format
• Assisted or fully automated authoring
• Content transformation or repurposing, such as summarizing, re-phrasing, or translating
• Content analysis and insights, including contextual insights such as reliability scoring

Imagine if all company announcements released on a given day could be analyzed in seconds, with summaries created in any language. Or consider the ability to scan a prospectus for anomalies and risks, contextualized within other companies in its sector or the market overall. An ESG report would be automatically generated by pulling data directly from the company's systems, with the correct taxonomy applied for the jurisdiction, or a report could be generated to explain a set of statistics to a non-expert.

What is stopping us? There are probably five key areas where we must pause for thought:

Technical challenges. There is good news: the NLP community is open-source by nature, and indeed open-source algorithms and engines are advancing extremely quickly. This means that organizations are not going to be reliant on just a few providers. Nevertheless, there is a scramble to assemble the rare skill sets required to build these new applications, which require not just technical skills, but also domain-specific subject matter expertise.

Reliability. NLP can produce extremely fluent, convincing-sounding sentences, and it is usually optimized to do so. However, it can be wildly wrong and has no “audit trail” regarding how it came up with a particular response. In the translation industry, Machine Translation is still largely used as an assistant and is edited by humans – and despite its significant progress, no one would yet rely on it for high-stakes content. Reliability and accuracy will improve over time, but caution must be exercised.

Security. Many organizations have already created policies prohibiting or constraining the use of public LLMs at work, to prevent information leakage and comply with data protection regulations. Organizations must ensure that their applications and data are well protected.

Regulation. In addition to the data protection considerations above, AI is likely to become increasingly regulated in the coming years. With the benefits of NLP, additional layers of compliance will emerge. Interesting conundrums around intellectual property and inside information may also need to be dealt with.

Change management. Enhancing existing processes with AI or fundamentally re-engineering them will be a significant change management exercise, impacting not only a company's systems but also the redefining of employee roles. Mapping this path and executing any such change is going to be high on the list of many businesses, presenting medium to long-term challenges.

Conclusion

The game-changing role of NLP in financial services content is just unfolding. As we navigate the challenges and risks, staying focused on the immense potential and opportunities that AI brings is crucial. With the right balance of technology and human insight, the future of financial services content looks promising and exciting. Financial institutions can look forward to a more efficient and globally interconnected future by embracing AI's capabilities and preparing for the changes it brings.