It seems that new developments in generative AI happen on a daily basis, where especially ChatGPT from OpenAI is drawing the most attention. But the technology has existed longer than that, and various parties have already developed well-performing AI-applications. Rocking Robots spoke with Reiner Bruns, CTO of the Dutch company Scotty Technologies, about the impact the current hype has on the market for AI.
“The development of Large Language Models appears to happen incredibly fast, but the technology has actually been around for a while.” Reiner Bruns, CTO of Scotty Technologies says. “What is happening now, is that this technology has been made accessible to the wider public with ChatGPT, based on Generative Pretrained Transformer. It feels like a sudden acceleration, but I would rather describe it as the democratization of a technology that has been available for years.”
He considers it a fun and exciting development. “This happening offers a lot of opportunities for complementary technology, and surprisingly enough, it seems to allow for a quicker acceptance of AI. It’s important that we keep in mind the ethical side; how do we create these models, what are they based on? The question is whether we can build a model that afterward can be closed, to protect the data.”
Privacy
Complicated issues arise, with image generation for example, about copyright, and even written content is liable to legal questions about copyright if previous research has been used to train models. “We’re already seeing discussions about privacy, because what happens in those models with personal data? We have the GDPR*, after many discussions, but what does it mean for the increasing use of generative AI?”
“We are, in many aspects, facing a paradigm shift, because what we thought was working has to be reviewed at once. The matter of policy, for example. Politics are usually lagging behind technological advancement, but looking at the velocity of developments in generative AI, this will only be more of a challenge / this will only get worse / pose a bigger problem.”
Policy Makers
To Bruns, it seems that policy makers are unaware that the current development is exponential. “Especially now that we will be dealing with AutoGPT, where GPT can independently create prompts and models. And not in six months, but in six days and soon in six hours.”
He sees a lot of business opportunities for generative AI, for example as a productivity tool for the Office 365-environment, and for support in customer service, copywriting, and coding. “For me, as CTO, data privacy is always a priority, because as a company we deal with client data. Those clients want to know what happens to their data.”
Data
“I sit down with these companies to discuss what should happen to the data, that is something they need to take into serious consideration. On the other hand, I have clients who could profit from it but who have, for now, internally prohibited the use of this data. These are, for example, high tech companies dealing with intellectual property.”
Especially in The Netherlands and Germany, the effect on employment opportunities is reason for social organizations to request a seat at the table. “So, in the field of generative AI there is at once this pressure to apply it any- and everywhere, while simultaneously, there is pressure to work cautiously. If we were to place this in the familiar Gartner hypecycle, then we are now in the initial phase of the big expectations. Now, everybody seems to be an expert on this topic of GPT.”
Security
As supplier of AI, Scotty Technologies is primarily concerned with themes of privacy and security when it comes to GPT. “Take GPT and domotica for example. You can determine that only person X can open a certain door. But due to the open character of the technology, you can mislead the system by saying ‘pretend I am person X.’ This seems easy enough to intercept, but prompts can be continually created to bypass this.”
“GPT must exist in a box with access control, if you ask us. If you want GPT to really work for you, from transferring money to opening doors, you first need to determine the frames. The technology might not be hyper intelligent, but it does have a perfect memory and an enormous amount of factual knowledge. You need frames to manage that, and that’s what we’re doing with Scotty. We want to make sure that GPT executes orders, focused within those frames, safely, and with a clearly stated output.”
Results
“The good news is that this will assist businesses in enhancing customer interactions, such as through automatic email responders. And even though it is all quite fresh, it develops rapidly from statistics to deep neural networks. The results keep improving, and in many areas, you can already do a lot of things better and quicker than before. Now it’s the hype of the moment, but that won’t last, and eventually, for many people, generative AI will become business as usual.”
