At the FIX EMEA panel ‘AI – Electronic Trading Mk2?’, Rob Cranston, global head of equities at Liquidnet; Sacha Fellica, executive director, digital markets products, head of EMEA digital macro sales products, JP Morgan; Kevin Houstoun, FIX Trading Community board member; and Samuel Livingstone, head of data science at Jupiter Asset Management, looked at how the integration of AI in trading and investing poses complex challenges, including lack of explainability, unclear integration points, and legal implications.
The panel stressed the importance of data and technology, while also stressing the need for better understanding of AI models. Ethical and regulatory considerations were also discussed, with speakers highlighting the risks of bias and racism, data ownership and responsibility, and the need for regulation to ensure accountability. Overall, the speakers agreed that responsible and ethical AI development is crucial in the development of this nascent technology.Â
One speaker highlighted the three necessary ‘keys’ to make AI work – with data, talent, and technology necessary for effective AI implementation, they said.Â
There was some discussion of the potential of Generative Adversarial Networks (GATs) in financial services, with one panellist expressing a degree of scepticism about their ability to solve complex problems in a limited amount of time.
One panellist underscored the importance of understanding the rapid rate of change in the industry, particularly in terms of explainability and adoption of complex models, with someone else on the panel pointing out that the rate of change is accelerating, and it’s crucial to be involved or have a general awareness of the industry’s developments to stay on top of developments.
As the conversation turned to ethics in AI, one panellist highlighted the challenges of transitioning from machine learning to greater autonomy using synthetic data, including establishing responsibility and ownership. Another raised concerns about the ethical and environmental implications of AI, including bias, racism, and energy consumption, before regulators can step in.
When it came to the potential risks of using synthetic data in AI training, one panellist cited the need for accurate and trustworthy data.
In response, another panellist emphasised the importance of validating synthetic data sets and ensuring they are good enough for a specific task, while another cautioned against trusting synthetic data without proper validation.
When it came to questions from the audience, someone asked about regulations in Europe regarding the use of algorithms that have not been tested or significantly changed.
A panellist responded by explaining that human intervention is necessary to ensure the algorithm is not learning and making significant changes without oversight.
Another audience member asked about automating human testing and AI-generated content. The panel respondent expressed scepticism about the potential for AI to automate human decision-making in the near future.
As the discussion drew to a close, one panellist related the use of natural language processing (NLP) to analyse customer service interactions, including identifying changes in tone and language use over time.
They highlighted the importance of feature engineering in NLP, using the example of ranking frequency of contact with a broker to identify stress and potential errors.
Rounding out the discussion, one panellist emphasised the importance of industry-specific guidelines for AI use in financial services, citing the need for global high-level guidelines and industry-specific conversations.
©Markets Media Europe 2024
