The Rise of A.I.

July 19th, 2023
ChatGPT x Awareways: opportunities & challenges for our Data ‘N Analytics cluster
DNA or “Data ’N Analytics” is our cluster of data analysts. They gather and process the data from our phishing simulations, security awareness training and culture scans into comprehensive reports that enable behavioral change in information security.
Moreover, they look for ways to improve on all of our security awareness interventions by crunching the numbers. How can a program like ChatGPT – or AI in general – support that mission?
Entering the mainstream
“ChatGPT is very popular right now”, agrees Daniël Kok. “That’s first of all because there is a very low threshold for the masses, which is usually not the case with AI. It is very accessible and the fact that it is language-based makes it even more approachable. Using the language model can also help us find a solution faster than googling and going through all other platforms one by one, but we can’t entirely trust the results either. As such, it is – as of yet – unable to properly support our work in programming.”
“We do most of our analysis of statistical information in R, an interpreted programming language. When we try to use ChatGPT as a question bank for R, it comes up with functions of R that don’t exist. Because it builds knowledge based on information provided, and then processes that into language it is programmed to offer. The approach is very step-by-step: it gathers knowledge about programming [language], then it looks into information about R as a specific programming language, and then it tries to predict an outcome based on what it has learned. But that’s not really how programming languages work.

…

Personal experience
“Let me make the same point with a more personal example. I’m currently planning a hiking trip in Austria. We’d like to avoid the Klettersteig paths, which is a German word (also known as ‘via ferrata’) for climbing trails that follow a cable-protected route through steep rocky terrain. I’ve asked ChatGPT to come up with a route that avoids all that to have a more pleasant and less equipment-heavy journey.”
“The result is a very convincing, well-written text – but it is completely useless. Because the offered route doesn’t avoid Klettersteig at all. The AI predicts how you process a text, how you’d like to process the results and how to convince on a language-level, but it isn’t supported by factual substance at all.”
Omniscient
“It’s important to note here that we’re not talking about an omniscient, all-knowing computer”, adds Jan-Willem Bullée. “The ChatGPT model is trained within a certain framework. It uses information available on the internet to draft its written output – but only up to a point.” Right now, its models are being trained on datasets that include information till 2021, signifying that ChatGPT doesn’t provide data on topics after this time.
“Which is also where another of its main drawbacks come from. It has no real thoughts or insights of its own, as it processes a collection of knowledge we already have.”
“Put another way: if you think of it as a chatbot, you’ll get a response that is quite useful – and indeed much better than what we’re used from chatbots in the past. But if there’s something it doesn’t know, simply because it doesn’t have access to the right input, you’ll get a gap in output it will nonetheless try to fill – much like in Daniel’s example of his hiking trip. It’s the same when you have gaps in your memory and still try to tell a complete story.”


AI as a support tool
“I do expect much from ChatGPT down the line”, says Daniël. “Going from the initial version to the now-current 4.0 was already a huge leap. Even going from 3.0 to 4.0 was a big step. In a sense, it’s not unlike the steam engine; a huge leap at the time in for instance agriculture, performing mechanical work faster and more efficiently. Society as a whole was afraid of job losses then and we’re seeing that again now. But we should focus on the positives, find out where it can support your daily routine.”
For instance, there is a tool that works rather well in programming, called GitHub Copilot. It uses the OpenAI Codex to suggest code and entire functions on demand. We can tap into that when coming up with scripts for dashboards and new methods for data visualization.
On a more basic level, it is already hugely supportive in repetitive tasks, but also in increasingly complex tasks that require the same method or approach every time. A great example is the data analysis button in Excel. It allows follow-up questions as well. When algorithms are more equipped to handle and process context, it’s going to be really exciting to see what new possibilities will emerge.
Waiting for the future
Regarding the latter”, adds Jan-Willem, “that is exactly what we have to wait for. In the end, you can teach almost everyone statistics, including an A.I. program – but the important part is always in the interpretation. What does it mean? Without that translation of all those numbers and statistics in any given dataset, what’s the point? Is a chess computer very smart, or just very good at statistics? At calculating probabilities of a huge number of possibilities, which it then used to calculate the next step? Provided with all the knowledge ever written on both statistics and the game of chess, it will probably blow you of the water in a one-on-one match-up – but only within the framework of the input it was given.”
“Look at it like this: a world-renowned expert in psychology may have all the answers when it comes to diagnosing the mental health of a patient – but what if you ask her to write a bunch of mathematical equations, perform a medical procedure or even come up with an innovative new way to tackle challenges in cyber security…?”
“The very particular skill set – or database of knowledge – quickly becomes useless out of context. I do think we’ll witness something quite spectacular in our lifetime, but we’re not there yet. I completely agree though: it will be very exciting.”

Get in touch!
Want to know more about our vision or approach, or curious about other information? Then feel free to contact us!
