Head of AI and Quantum

Sergio Gago, Head of AI and Quantum

As the financial industry stands on the cusp of a major transformation, generative AI and large language models (LLMs) like ChatGPT are poised to revolutionize various aspects of finance, from risk management to investment all the way to customer service. However, it is essential to understand how these technologies work, where they fall short and where they can augment humansas "copilots". Many of us are already using generative AI at the personal level - how do we bring it to the enterprise level in a safe and scalable way?

Artificial Intelligence itself has been around for decades. Most financial institutions, including Moody’s, have Machine Learning algorithms embedded in their products providing great advantages to customers. Like “being in the cloud”, using Machine Learning is table stakes.So, what has changed? How has ChatGPT up ended the way we look at AI?

Training Machine Learning algorithms has never been easy. Traditionally we required enormous amounts ofdifficult to find, qualitydata,and people with very specific technical skillsets, such as data scientists and engineers,who were in high demand and expensive. Without a guarantee of a working algorithm, it was a risky investment. Now, Chat GPT has brought about a “no-code” version of ML,and anybody can create their own classifier, fraud identification machine, or data normalizer without specific skills. We call it “Prompt Engineering”, or “knowing how to ask the right questions to a chatbot”.

Note, the essential skill remains knowing how to ask the right questions - humans remain crucial for providing context and creativity, not to mention validation and verification of data. In this new era of collaboration between humans and AI, we lean into the concept of "centaurs" – human-AI teams as coined by the chess master Gary Kasparov. By leveraging human intuition and creativity combined with AI's processing power, centaurs have the potential to outperform purely human or machine-driven solutions.

"The Key Lies in Embracing New Skills that Complement Emerging Technologies. Generative AI will Help us do More and Better and the Demand for Most of the Products in the Finance Industry will Simply Increase"     

For example, an investment manager can utilize AI for real time updates and benchmarked information about clients or investments, and a risk manager could assess operational or investment risk using AI but both will rely on human expertise for final decisions. A customer service agent will be able to produce fully customized responses to users based on the whole history of that relationship as well as expert knowledge, in a way a human could not. KYC processes can be greatly improved by having a vast array of information – from PEP to UBO or ESG data - digested and processed for you. And, despite their promise, generative AI models come with certain challenges that must be overcome. Ensuring these models produce accurate and reliable results is essential for building trust in their capabilities. ChatGPT is famous for its “hallucinations”, the technical name for when a bot gives a random answer with 100% confidence, but absolute inaccuracy. Additionally, bias present in training data can lead to unintended consequences when applied in real-world situations. On top of that, existing models are trained with public content up to a certain date (2021 for ChatGPT), so you can’t ask questions about recent events or any piece of information that is licensed or behind a paywall.

Those challenges aside, large language models(LLMs) present an opportunity to be as transformational to the finance industry as the Internet. But it is a daunting task that pulls in all areas, from technology to change management.

To effectively integrate LLMs into production workflows, companies must be systematic - first identifying specific use cases and business areas where AI can add value, such as automating routine RFP tasks or generating data insights for asset managers. Firms must work with industry partners e.g. cloud providers with their own LLM models available, but still invest in research and development to fine-tune AI models for their unique requirements. Finally, deployment requires ongoing monitoring and maintenance to ensure optimal performance and adaptability.

And don’t forget that anything one company is doing to find efficiencies through generating text are likely to be matched by competitors. The real game changer will be user facing use cases in which a company’s real competitive advantage can be enhanced through LLMs. This could be in the form of proprietary data assets or access to exclusive financial models and resources.

The rise of generative AI has sparked concerns about job displacement- but history shows us that technology also creates new opportunities. The key lies in embracing new skills that complement emerging technologies. Generative AI will help us do more and better and the demand for most of the products in the finance industry will simply increase.

Specialists such as data scientists, MLOps (for defining the data pipelines) software engineers and now "prompt engineers” will still be required. But embracing cultural change and agility is key - every team member should have a “hacker” methodology where speed and the ability to break traditional corporate barriers are encouraged. As generative AI becomes more sophisticated, we can expect the emergence of an "everywhere co-pilot" strategy: individuals will have access to personalized AI agents that provide real-time guidance and support across various aspects of their lives – both personal and professional. This ubiquitous presence will enable humans to make more informed decisions and optimize productivity.

In the race to capitalize on generative AI's potential, companies with access to vast amounts of data will have a distinct advantage. Data is the new currency – providing essential fuel for training and refining AI models. Firms that can harness this resource effectively will be well-positioned for success in an increasingly competitive landscape. LLMs will be a commodity and will be everywhere. Some will be small – domain level – agents, others will be generic and multipurpose (like ChatGPT). If all these models feed on real, verifiable data, this will be the real trade in the "AI economy”.

As generative AI continues to evolve, its impact on the financial industry is poised to be transformative. By adopting a human-centric approach, companies can leverage these advancements while mitigating potential risks. The future belongs to those who embrace this powerful synergy between humans and machines, unlocking unprecedented opportunities for growth and innovation.