There has been a lot of hype surrounding a new technology released at the end of November 2022 called ChatGPT. I believe ChatGPT is an embryo, not even fully born… yet!
ChatGPT is a natural language processing (NLP) chatbot utilizing OpenAI’s Generative Pre-Trained Transformer 3 (GPT-3) language model, created in May 2020. The model predicts the next token value of a response based on previous token values, leveraging a technique that utilizes unsupervised pre-training of the model with specific supervised fine tuning. It has been used to generate written paragraphs, website content, and papers that are difficult to distinguish from human-written prose. OpenAI published a paper, “Improving Language Understanding by Generative Pre-Training,” which describes the technique in further detail.
After ChatGPT came out, the educational establishment raised concerns that students could use it to “cheat” on term paper creation. Of course, other technology is already being developed to detect AI-generated prose. Notably, a new tool developed in January 2023 called GPTZero can read the prose and assign a score to indicate how likely it is that selected prose is bot-generated. Unsurprisingly, the student who developed the tool was interviewed on CNN on February 1, 2023.
Given the hype, I wanted to experiment with ChatGPT to see what it can do now and how it could evolve over the next few years. I logged in, created a test account, and cut and pasted my resume, the link to my LinkedIn profile, and the link to the Aite-Novarica Group website. From there, I asked a few simple questions and received answers; all of them were well-written, but many were inaccurate. Examples of questions I asked included, “Who is Mitchell Wein at Aite-Novarica Group?” “When did Mitchell Wein graduate from college?” and “Where did Mitchell Wein work at the turn of the century?” The bot told me it could not read the LinkedIn link. When I pasted in the section of my resume related to the question I asked, the bot responded with a well-written answer summarizing the information I had just given it. Interestingly, the bot could not remember facts from prior sections of the conversation. When I asked the bot about the forgotten facts I had provided it just a few minutes earlier, it gave me this response:
“I apologize if I am not retaining the information you are providing correctly. As an AI language model, I do not have a memory and can only respond based on the information provided in the current interaction. Please let me know how I can assist you further.”
The bot also told me this: “As an AI language model, I do not have memory in the traditional sense like humans do. I have been trained on a large corpus of text, and I can generate answers based on that training. When answering a question, I do not store information about previous questions and answers in this conversation or any other conversations I have had. Instead, I process each question independently based on the information provided in that specific question and my training.”
The bot is purely text-based and cannot interact with pictures or graphics. This reminded me of early DOS prompt interfaces from the 1980s.
So, ChatGPT is still in its very early days. There is much more to do to make it a fully functional, general-use AI platform. Additionally, the underlying computing environment is still limited—especially regarding concurrent processing—even though it is in the cloud; quantum will solve that by 2030. When I was using the tool, it was slow at times and crashed at times. Also, it is not apparent how the data streaming through the ChatGPT platform will be protected from a security perspective. Has zero trust been deployed for users of this platform? I don’t think so! However, ChatGPT has massive potential.
Generally, firms are investing a lot of money in AI automation. Angus Loten’s article “Uncertain Economy Spurs Growth in AI-Powered Office Automation,” published in the February 2, 2023, Wall Street Journal, suggests that the pressure to do more with less has led a third of global corporate technology chiefs surveyed—more than 2,000 respondents—to anticipate that AI-powered automation software, including decision support and process/task automation, will account for the largest share of new IT funding. Microsoft is planning to integrate ChatGPT into its software and platforms.
Community banks need to start thinking about how they can leverage these new technologies as they continue to mature their IT ecosystems. Conversational AI platforms could be powerful tools in a community bank’s technology arsenal. That is, if the conversational AI platform can leverage everything available on the internet and vet the information against the insights needed (similarly to a search engine); interact with customers, regulators, and executive management to determine what they need; and has the APIs to connect to third-party data sources, fintechs, and back-end data and processes from the core systems and incorporate operational process documents. Possible use cases include loan processing and origination, customer onboarding, payments, settlements, cash management, generating regulatory responses and reporting, and cross-selling new products.
Of course, to meet regulatory muster, it will have to leverage all of the technology best practices, including zero trust, to ensure that communications and data moving between customers and the AI platform are not compromised or stolen. Change management will be another area that will need to be tightly governed to ensure that AI-based learnings are not corrupted through human error regarding change control and testing.
It is still very early days for this technology. However, community banking CIOs and CTOs must keep track of how AI evolves. They must also assess the impact of quantum on AI as it becomes available and what the core system and office productivity vendors are doing to incorporate AI platforms directly into their offerings. It is clear that the digital interactions that AI powers will be vastly different in 2030 than in 2023.
