I am a user experience guy.
My career has been spent trying to work out how to make complex technology and digital experiences consumable by users.
The last three decades have seen rapid advancements in interaction design, graphics, and multimedia capabilities. But while the digital interfaces of 2023 are leagues ahead of those of 1993, the primary challenge remains: how to reconcile an increasingly sophisticated user experience with the limitations of technology.
The result has been that individualisation (or personalisation) has yet to be achieved, i.e. delivering digital experiences based on how an individual wishes to consume that information. That is until last year, with the launch of ChatGPT-3.5.
As a user experience person, I believe that Large Language Models (LLMs) and conversational interfaces completely change how information is integrated, delivered, and consumed. We truly stand at the start of something momentous.
For this article, I asked a GenAI expert I have been working with, Lee Mallon, for his thoughts on what is happening. Lee is based in the UAE and is working with ADGM, Abu Dhabi’s financial free zone (among other companies). Based on his projects and some personal R&D, Lee believes that the future of AI is about dynamic knowledge rendering.
According to Lee, “Generative AI enables the conversion of raw content and data into meaningful, digestible, and personalised content for audiences to absorb. It is already possible to feed structured and unstructured data into LLMs that natural language prompts can interrogate. Existing models can generate text, images, videos, and audio. In the not-too-distant future, by stitching these models together in a workflow, LLMs will be able to deliver a content experience that is truly individualised.”
The implications of this are profound for digital experience. The same information could be rendered in multiple ways to suit the needs and aptitudes of the end consumer.
Dynamic knowledge rendering is already happening, and as it happens, it is happening in financial services. A case in point is LTX, a subsidiary of Broadridge focused on electronic trading in the fixed-income market, which has started down the path.
I spoke with Jim Kwiatkowski, its CEO. What I thought would be a conversation about AI’s impact on trading platforms and liquidity clouds turned into a conversation about user interfaces. Happy days!
Jim told me that one of the key challenges LTX faced was “designing an effective user interface that could present a wealth of data to fixed income traders and portfolio managers without consuming too much screen real estate”.
Clients wanted abundant information yet had conflicting requirements about what that information should be. Additionally, they did not want the interface to be too cluttered. This problem was particularly acute given the complexity of the fixed-income market, where traders need access to an extensive array of details, from CUSIP numbers to price, direction, and size, among other variables, and the need to make quick decisions which might rely on any of it (or all of it) at once.
To navigate this dilemma, LTX began experimenting with GPT-based technology, starting with GPT-3.5 and then transitioning to GPT-4.
They found that a conversational, natural language interface could condense a massive amount of information into a very small space.
Introducing BondGPT. By employing a text box where traders could ask specific questions, they managed to offer a user experience that was both data-rich and spatially efficient.
This approach removed the need for traditional menus and complicated user interfaces, which typically require far more screen real estate.
This natural language interface allowed traders to quickly access complex insights based on a blend of raw data, with the prompt simply being “Ask me anything about bonds”. As a result, traders could bypass the cumbersome step of manually gathering different pieces of raw data before combining them for actionable insights. Instead, they could ask directly for the required insights, receiving a quick and precise response.
According to Jim, “LTX employed a carefully designed rule framework to ensure their GPT-based system was accurate and quick, while still being compliant with financial regulations.”
The rules include:
- Accuracy: The foremost concern is to ensure that the information provided is precise, particularly given that the financial markets have no room for errors or so-called “hallucinations” that AI can produce. To this end, LTX incorporates curated data and uses its own models for bond arithmetic to minimise the risk of errors. They also turned down generative features in the GPT model to prevent it from producing inaccurate or irrelevant information.
- Compliance: As a broker-dealer, LTX had to ensure that the AI doesn’t breach any compliance rules. For instance, it cannot offer investment advice, so questions like “What’s the best bond?” cannot be answered. To enforce this, it built an “adversarial agent” to act as a final checkpoint. This agent vets the AI’s responses against a pre-loaded set of compliance rules before releasing the information to clients.
- Speed: Traders require immediate answers, so the system was optimised for quick response times. It would lose its utility if it takes too long to generate a response, regardless of how accurate it might be.
- Custom rules: The adversarial agent can also be configured with custom rules specified by the customer, adding an additional layer of flexibility and adaptability to the system.
LTX exemplifies the possibilities of GenAI. From the demo that I saw, the company is also evolving fast. For its customers, they are already getting a personalised experience. LTX is also already going beyond text, with graphs and images forming responses.
I wonder how long it will take before it includes videos and audio.
I knew there was something I should have asked Jim!
The development of LLMs like GPT-3.5 and GPT-4 is a watershed moment for those of us in UX. These technologies offer a powerful way to break down the longstanding barriers between what is technologically possible and what is desirable from a user experience perspective.
As LTX demonstrates, we are not just on the cusp of significant change – we are in the midst of it.