But what exactly is GPT-3? Why are people so excited about it? And what does it mean for firms on a practical level? By digging into how the program works, we can better understand AI’s promise for financial services.

Just the beginning

GPT-3 is a computer program created by Open AI. It’s a neural network that takes a massive amount of language datasets from the internet (including sources like Wikipedia and news sites), and does incredible things with them.

The program is able to produce lines of text that resemble human-written content. But this is just the beginning. The significance of GPT-3 for businesses lies in its potential to automate tasks. This has implications for everyday processes like due diligence, onboarding and reporting, but it also promises to change the way that firms interact with customers.

A powerful model

Language models have been around for some time. They compute the conditional probability of words, calculating the likelihood of words to appear in a passage given other words in the texts. But GPT-3 is different. It’s a very powerful language model that’s trained to repeat the next character-base on thousands of previous characters. With a single sentence, the program can create an entire essay consisting of contextualised information.

GPT-3 is fed huge samples of text, which it compresses into numeric representations called vectors. It then unpacks this text into valid sentences. Once the model has been trained, it can accurately predict what words will come next in a sequence when stimulated by a person typing an initial word (or series of words). But the program’s expertise goes beyond writing. It can also perform reasoning tasks that involve natural-language responses. And it can even play chess!

The program’s somewhat functional name refers to General Purpose Transformer, meaning that the program generates sequences of original text, rather than simply spitting out numeric scores or binary answers like previous language models. In other words, it can complete domain-specific tasks without being programmed with domain-specific knowledge.

Reducing bias

Machines have always been good at logic, but they have lacked creativity. GPT-3 demonstrates how computers are changing – they are becoming less limited, more general, more human. And because the program is proficient in text, it has the potential to spawn a huge number of applications. Right now, GPT-3 is only available in controlled beta to approved parties, and a small number of developers are doing extremely cool things with the program.

Furthermore, the Open AI team have been keen to explore the application of Ethical or Responsible AI within GPT-3 in order to assess and understand pre-existing sentiment biases which exist in language. Studies have been conducted to observe linguistic bias in terms of racial, religious and gender bias.

Initial analysis has shown that a degree of linguistic bias is present which is, perhaps, to be expected due to models making associations with terms that are already linguistically commonplace.

The team is actively working to reduce bias and increase fairness, for instance highlighting that “there is a need for building a common vocabulary tying together the normative, technical and empirical challenges of bias mitigation for these models.” It is positive then that Open AI has committed the resources and skillset required to reduce bias and does so as an ongoing activity.

This is clearly an area which demands constant attention to avoid issues arising at the scaled AI level.

Glimpsing the future

Within the context of financial services, GPT-3 promises great things. Perhaps the most obvious application that springs to mind is that of robo advisors. Not only can the program intelligently infer intent – it can also produce more human-like responses to users. It has the potential to work incredibly well for triaging online customer enquiries.

Another exciting application is sentiment analysis. GPT-3 analyses hundreds of thousands of data points from the web, enabling financial services organisations to check positive and negative reviews of their offerings over a long time period. This could be helpful for identifying opportunities across product lines, even entire franchises.

Only a few years ago, neural networks were built with functions tuned to a specific task, such as translation or question answering. Now we have a neural network that is general – it doesn’t need task-specific functions, or dedicated datasets. We’re talking about a new chapter in the story of machine learning.

Accelerating transformation

It's still early days for GPT-3 and it has a long way to go until it can be commercialised on a wide level. The company’s CEO Sam Altman has said,

"The GPT-3 hype is way too much. It's impressive but it still has serious weaknesses and sometimes makes very silly mistakes. AI is going to change the world, but GPT-3 is just a very early glimpse. We have a lot still to figure out."

But GPT-3 is already moving the needle for firms who are interested in embracing the power of data. It hints at what’s possible for AI in the years ahead, illustrating how the right technology can reduce production time and augment workflows without the need for armies of data scientists.

GPT-3 enables practitioners to engage with different use cases in a very short period of time, engaging key stakeholders with the art of the possible. Visualising the future no longer requires complex PowerPoint presentations; in the space of 20 minutes it’s possible to demonstrate multiple use cases. GPT-3 also helps to determine next steps for deployment in the real world after delivery of a successful proof of concept, driving innovation by accelerating the data maturity journey for less mature businesses. Creating use cases is simple, taking hours rather than days, weeks or months as it did previously.

The art of the possible

When working with organisations to help them leverage the power of data, one of the core issues that Mudano, part of Accenture faces is determining how to deploy AI. In other words, where should it be used? Exploring the Art of the Possible is a potential unblocker in determining next steps.

For organisations that haven't embraced data, automation and AI, GPT-3 can provide a high level of intelligence and ability on a standalone basis. It doesn’t require massive upfront investment. Instead, the firm can create intelligent apps that make API calls to GPT-3, avoiding the need for expensive data warehouses, pipelines and production systems.

By illustrating what’s possible, machine learning technology is helping technologists and business leaders everywhere to imagine a future that is more automated, scalable and above all, user-centric.

An article by: Boris Power

Boris Power is Lead Data Scientist at Mudano, part of Accenture. Boris specialises in prototyping, Deep Learning and Machine Learning algorithm tailoring to solve a variety of business problems for Mudano’s clients. He has over a decade of experience of bringing innovative data science into real-world applications to improve process efficiencies for clients that deliver value growth.

{{ article.HEADLINE }} {% if article.external_mode %}{% endif %}

{{ article.HEADLINE }} {% if article.external_mode %}{% endif %}

{{ article.HEADLINE }} {% if article.external_mode %}{% endif %}