InvestTech 2023 - ChatGPT: The new age of AI

By Jayson Forrest

ChatGPT is one of a new breed of AI language tools that are poised to transform businesses and their relationships with clients. Dr Michael Kollo (Qurious Analytics) explains how ChatGPT is likely to impact the financial advice sector, helping improve productivity and making advice more accessible to a greater number of Australians.

There are many industry commentators who’d have you believe that the latest iteration of artificial intelligence (AI) language tools will be the panacea to the problems plaguing the financial advice space - including reducing costs, building efficiency, and tackling compliance. Some would even suggest these new technologies could even threaten the dominance of content aggregators like Google.

However, while this new breed of technology is exciting for advisers, it’s unlikely to replace the human interaction advisers have with their clients. Instead, this technology is shaping up as another tool that advisers can use to improve productivity and enhance their value proposition.

This was the overarching message delivered by Dr Michael Kollo - Founder and Chief Executive Officer of Qurious Analytics - who used his presentation at the IMAP 2023 Virtual InvestTech conference, to explain how AI language technologies, like ChatGPT, will strengthen the advice process.

According to Michael, ChatGPT is a powerful new AI chatbot system that can be used by businesses across all industries. Essentially, it is a language algorithm that is able to transform and translate language from one place to another in different formats. He says the technology is a bit like a “master linguist”, with a reported linguistic IQ of 148.

ChatGPT uses technology in the GPT-3 language model series, which is based on 175 billion parameters and a 570 GB training dataset comprising of books, articles, websites and more. GPT-3 stands for Generative Pre-trained Transformer 3 – it’s the third version of the tool to be released. The next version, GPT-4, is likely to be based on parameters in the trillions.

Developed by San Francisco-based OpenAI, ChatGPT was launched in November 2022. It enables the user to type natural language prompts - like, ‘What is a ‘managed account?’ - and ChatGPT’s algorithm trawls its dataset of text and then responds to the question in a conversational way.

“This algorithm is particularly good for language related problems. The dialogue format makes it possible for ChatGPT to answer followup questions, admit its mistakes, challenge incorrect premises, and reject inappropriate requests,” says Michael. “Essentially, it transforms written language (both human and code) into different formats: which can be longer and shorter, formal and informal, summarised or expanded, funny or serious - depending on what the user requires.”

Michael says this technology is suitable for ongoing conversations by the user, as it is able to remember the context of the user’s previous questions and answers. He adds that ChatGPT learns from a wide set of online sources, and will increasingly become more aware of an adviser’s documents, making it ideal for compliance purposes.

“ChatGPT can also be used to take on a persona, like a personal assistant, to provide ideas, outlines and responses,” he says.

In addition, it is a safe technology and heavily controlled to ensure it is suitable for a wide range of users and industries. For example, the GPT-3 algorithm that ChatGPT uses has been restricted from not saying negative things, or be abusive, racist or sexist, allowing it to be utilised within a corporate setting.

As part of the ongoing client review process, Fraser believes it’s important for advisers to have conversations with their clients about how they feel about their data being held by the advice business on their behalf.

“Clients are hearing a lot in the media about cyber breaches and are likely to be fearful about the possibility of their identities and personal information being stolen. Clients are justifiably nervous. Therefore, it’s important for advisers to get on the front foot and educate their clients about the cybersecurity in place to protect their data.”

Fraser adds that it’s equally important for advisers to train their team to talk about what they’ve learnt about cybersecurity, so they can remain informed and provide clients with helpful information about this topic. Cybersecurity audits are also an additional opportunity to discuss issues with both the advice team and clients.

“Whilst advisers think that asking clients to do things that requires an extra step - like multi-factor authentication - is going to be annoying for them, in the vast majority of cases, clients are actually going to be receptive and supportive that you’re taking that extra step to protect their data.”

While there is a lot to like about this new technology and its application for use by advisers, Michael cautions that ChatGPT is not without its faults. First among these is its inability to reason or create new and original information and content. However, it can mimic ‘thinking’ processes, particularly when explaining the rationale behind coding and other technical approaches.

“It’s important to remember that ChatGPT is an AI language tool. Therefore, it does not feel, nor does it have a ‘will’ or opinion or experience. It is not sentient, so it is unable to perceive or feel things, and it does not have common sense. Instead, it needs to be specifically calibrated in order to provide accurate references and content. Without this specific calibration, it can infer ‘truth’ by accessing a range of different sources, and by doing so, can get things wrong,” says Michael.

In addition, ChatGPT does not easily provide counter-arguments for contentious topics.

“Therefore, when putting together information, like a financial strategy, ChatGPT requires specific context, in order to put all the relevant information pieces together. However, without context, it will just be information that’s out of place for that particular topic.”

Citing research that shows 50 per cent of workers spend more time looking for information than working, with 19.8 per cent of their time spent searching for information, and an average of three hours a day dedicated to writing emails, Michael believes that AI language tools, like ChatGPT, will be hugely advantageous for the financial advice sector, particularly in terms of improved business efficiency and productivity.

He points to four areas where AI language tools can be particularly useful for any advice business. They are:

1. Email generation

Michael believes email generation will be one of the more popular uses of ChatGPT. With multiple clients to service, using ChatGPT will enable advisers to improve their time efficiency and boost productivity.

2. Summation of information

Advisers will be able to take the technical information they receive, such as reports from investment committees or fund managers, and use ChatGPT to summarise this information and reformat it in a style that clients will find more user-friendly to read and importantly, understand.

3. Reports and user manuals

According to Michael, ChatGPT is “insanely powerful” for assisting advisers to generate reports and internal user manuals.

“For example, it has the ability to take onboard your compliance documentation and reorganise it, summarise it, query it, reformat it, point out inadequacies with it, test you on it - it does it all,” says Michael.

“It can generate regular reports for your team or clients. If an adviser wants to write up notes from a meeting, ChatGPT can be used to reorganise those notes and summarise them, thereby saving the adviser considerable time.”

4. Books

ChatGPT can also be used to generate content for a book, but within the scope of the context provided to it. ChatGPT can also write content in the style of a particular writer, however, Michael cautions there may be legal repercussions in doing so.

“The algorithm has a very detailed understanding of different thematic and linguistic styles of writing. There are 25 different ways you can characterise a piece of written text in terms of the structure of it. By specifying those 25 different ways, you can use the technology to write a book in the style of a particular author you like.” 

While AI language tools are still an evolving technology, it does raise a number of issues, particularly in relation to content copyright and client data privacy.

According to Michael, ChatGPT has been set up as a user account. Each conversation is kept within that account and although it is able to recall previous questions and answers, its memory is relatively short.

“Ultimately, any kind of memory and context will be based on how it gets applied in an Application Programming Interface (API),” he says.

As it currently stands, Michael says it would be inappropriate for an employee to use ChatGPT with a client’s private information. That’s because technically this information is not domiciled in Australia. Instead, it’s sent to another server offshore, before coming back to the business in Australia. However, he believes that in the future, you’ll be able to keep your data in a private cloud and then use an API service, like Microsoft Azure, which will process that information and return it to you within your local environment.

“In order for this technology to be compliant for financial services companies, and any other regulated sectors, information cannot leave their premises, including their cloud systems. So, for the moment, think of ChatGPT as a helpful tool for everyday use of things like emails and client communication.”

Another issue emerging is over copyright infringement and ownership of content. This is a very complex and evolving area of law. As the datasets of AI language tools include content that is subject to copyright protection, it means that the output generated by ChatGPT might be similar or even identical to works already in existence. This potentially gives rise to the risk that the use of this output, without permission, could constitute copyright infringement. Therefore, it’s something users need to be mindful of.

Despite some teething issues around privacy and copyright, Michael remains optimistic about the future of AI language tools and their application for the financial services sector.

“What we are seeing with the likes of ChatGPT is the birth of conversational AI, where improvements will ultimately result in this technology being indistinguishable from humans. The combination of a ‘persona’ and language skills will give this AI a more conversational credibility.”

He adds: “Increasingly, we’ll see this digital technology used within businesses to help them make better decisions - both internally for staff training and onboarding, and externally for customer service management.

“I believe you will also see a new age of personal versions of this technology that will be used as digital companions. For example, when somebody joins an organisation, the digital companion will be able to answer the person’s questions around numerous topics, like HR and corporate policy. This type of AI technology will become the new digital avatars and companions for businesses, governments and individuals.”

Michael predicts the next step in the evolution of this technology will be the addition of voice capabilities, and possibly even further in the future, a physical shell.

“When it comes to AI language tools, we’re seeing an exponential of Moore’s law apply to the sophistication of the algorithms. This will mean that future versions of this technology will have increasingly less problems and faults with them.

“We’re going to see enormous amounts of improvements across these algorithms, as well as with their ability to help us with our daily work and productivity. However, I don’t think this technology will completely replace humans. Genuine communication and human interaction is still very important for a lot of companies,” says Michael.

Dr Michael Kollo is Founder and Chief Executive Officer of Qurious Analytics. The session, ‘The new age of AI language tools’, was moderated by Fraser Jack - Founder of The Cyber Collective.