AI Solutions for AI Problems.
Artificial intelligence and machine learning (AI/ML) have been used in the financial services industry for more than a decade, enabling enhancements that range from better underwriting to improved foundational fraud scores. Yet these have largely fallen short of their promise of reducing friction internally in finance and more importantly — reducing friction for customers. But what about safety, service and actually helping people ?
How many of these widgets that sit on your phone actually help you ? How many are proactively making us safer, predicting what’s best for us or recommending better paths for us to take ? Or are they really just gadgets?
Generative AI via large language models (LLMs) represent a monumental leap forward in reducing friction. But their limitations on data supply will hamper them delivering ‘real’ help. The Long-Tail Problem in AI is very real — especially for LLMs.
While LLMs are really still in their infancy, they are transforming healthcare, education, commerce, and significant cost structures within institutions. While traditional AI/ML is focused on making predictions or classifications based on existing data, generative AI creates net-new content. This is wonderful for artwork (thanks Midjourney) but plain dangerous for when you really can’t chance the outcome — eg. your health.
It’s clear that the ability to train LLMs on vast amounts of unstructured data, combined with essentially unlimited computational power, will yield the largest transformation the services market has seen in decades. This is not confined to finance, but the underlying framework for organisational behaviour itself.
Combined with the composable nature of Blockchain architecture, Gen AI will unlock the ability to share and store data privately in ways that were previously impossible. The combination between cryptography and neural networks will change the way data is stored — for the better. It will also unlock previously impossible frameworks for how people interact with each other. More of a symbiotic relationship between participants than a ‘conductor & orchestra’. This ‘direction of the chairs’ analogy perfectly reflect the dilemma for most LLMs. Really, LLMs are only as good as the conductor — or the orchestra (underlying data). But what LLMs lack today is the tailored data propositions that users desperately need.
General solutions are great for general problems but personal solutions are imperative for personal problems. LLMs simply can’t cater for the Long-Tail of AI problem solving. We need new architecture.
Unlike what came before
Unlike other platform shifts — internet, mobile, cloud — where the services industry lagged in adoption, the best new companies and incumbents are embracing generative AI right now. The combination between creating net new paradigms, within the deep learning architecture of the neural network — also affords institutions of all categories tremendous advantages. These include bio-science, pharmaceuticals, intelligence architecture and ofcourse finance.
But with new platforms will come new problems.
Generative AI will make the labour-intensive functions of pulling data from multiple locations, understanding unstructured personalized situations and unstructured compliance laws, exponentially more efficient. The term 1000x is simply too narrow to explain just how many humans it would take to compose what an LLM does in seconds. It will help synthesize, summarize and suggest potential answers for solutions that have always beguiled financial institutions.
AI Problems
But with Generative AI solutions comes Generative AI fraud, risk and deception. We will need new artillery to combat the waves of AI generated fraud that will iterate at the click of a button.
AI Solutions
On the positive side, the first fintech players to correctly apply blockchain with neural networks will have a significant advantage over almost all incumbent financial institutions. Why ?
Because the fundamental architecture of incumbent financial institutions was never built to handle data at speed, to provide instant responses to rapid questions. Like the horse and cart — they were never designed to travel at the speed of light (aka. a neural network).
Traditional financial instutions were designed to serve a purpose. When the first programmable functions were enabled (eg. COBOL) they formed the building blocks for most of these institutions. They also provided scaffolding for safety, security and privacy that ensured customer safety. They were never designed to feed artificial neural networks.
That’s why most financial institutions simply cannot achieve real AI results that help people. Their cognitive abilities are hamstrung from the outset. Furthermore, most will be mere observers in the race to provide real-time solutions to real-time problems.
Peter Toumbourou
Charleston Advisory Group
2024
For more click here
