The evolution of AI & lending: Parallel journeys shaping the future

The timeline of  AI development

The diagram below shows the parallel timelines of developments in AI and lending technology. Let’s look first at the key milestones through the decades of AI development. 

1. The origins of AI: Neural networks and early systems

AI’s journey began in the 1950s and 1960s, when the foundational concepts of neural networks and artificial intelligence emerged from the work of mathematicians. Bringing together rule-based and expert systems to offer new approaches to solving problems offered a lot of promise. Excitement around using these new concepts to solve a variety of problems, however, waned as some of the promises fell flat. AI’s “winter” in the 1980s eventually gave way when researchers turned their attention to machine learning (ML) in the 1990s. They focused efforts on using data and algorithms to uncover patterns independently, with very little human intervention.

2. The rise of machine learning and computational power

In the 1990s and early 2000s, advances in computational technology transformed the landscape. Distributed computing became mainstream, and the launch of public cloud infrastructure such as AWS increased computing power. The increased capacity allowed systems to handle larger datasets, so researchers revisited neural networks to focus on deep learning, which is essentially thousands of neurons stitched together in a specific architecture. This launched the “age of neural networks” starting around 2006 and lasting about a decade. Companies began creating practical applications, like Apple’s Siri and Amazon’s Alexa. These machines could understand voice instructions and connect to different databases or systems to assemble useful information to return to the user.

3. The transformer era and generative AI

The next breakthrough came in 2017 with Google’s introduction of the Transformer neural network architecture. Transformer gave a lot of new capability and power to the neural networks, providing the ability to look at large amounts of information and connect the context of that information. Earlier models could only analyze small snippets of text, but with self-attention and other innovations, Transformers enabled AI to process entire documents.

This Transformer architecture really took off. Combined with the increased computational power, it spurred researchers to expose the architecture to as much data as possible. This innovation paved the way for Open AI’s ChatGPT and other large language models (LLMs), which leverage vast internet-scale datasets to generate human-like text and insights that are similar to the input data but slightly different. With this capability, generative AI has moved beyond text to include images, voice, and video, creating entirely new possibilities for application.