My view on artificial intelligence, a little over a year with ChatGPT
I have kept out of the AI discussions until now, but let me sum up how I see the era after ChatGPT.
I was quickly impressed by the results from ChatGPT when it launched, and I was early to try it out. Even though the results were good, I felt at the time that I could produce better results on my own. It could be useful to start from a suggestion from ChatGPT when writing a text, but I ended up prioritising getting into the flow zone myself and working with my plain, organic intelligence instead.
My background in AI
I have been fascinated by AI since I was introduced to it through popular culture in the nineties. When I started my master's degree in 2004, I went deep into the technology behind the artificial neural networks (ANNs) of the day. My thesis was about linking ANNs to the biological neural networks in our brains. Yes, our brain can be understood as several neural networks tied together, each with its own speciality.
Specifically, I looked at the possibility of using agent technology, autonomous computer systems, to simulate the three biological networks that are central during attention tasks.
The study itself was interesting enough, but there was too much uncertainty about how the biological networks actually work, and it was not easy to build an artificial model that resembled the biological one closely enough. Despite this, I got the paper published together with my supervisor at an ESM conference in Toulouse in 2006, so the work was not entirely without academic value!
One moment I remember particularly well was when I was working on a hand-in assignment where we had to implement a pattern recognition algorithm to identify specific gene sequences. I was in a bit of a hurry, so I had not got to grips with the problem very well. Still, I got the implementation to work. I remember thinking that for the first time I had made a computer program that was smarter than me: a program that delivered a result I had not foreseen, but that was correct all the same!
When I started my professional career there was little focus on AI and machine learning, but I have followed developments all the way. We have seen explosive growth in the reach of AI, and in access to it, with ChatGPT. Even so, there have been several bigger breakthroughs over the last decade, with object identification in images, more or less successful chatbots, speech synthesis, voice recognition and deepfakes.
Welcome, ChatGPT
So, late in 2022 we got an AI tool that everybody started talking about! As I mentioned at the start, I was impressed by the results, but I had a hard time finding useful areas of use. When I thought of a question to ask, the answer was often so incomplete, and often wrong, that you had to spend as much time fact-checking what came out as it took to find the answer yourself from various sources.
You could also ask it programming questions, but when I tried that, I always spent more time on asking the question and integrating the code into the solution. The answer from ChatGPT usually did not work directly, but it could be a good starting point for solving a problem.
I saw ChatGPT as a useful toy if you were, for example, starting to learn programming, or as an assistant that could take you from ignorant to amateur in a subject.
Constant development
Over the past year ChatGPT has developed rapidly and achieved better and better results. ChatGPT 4 in particular! It can search the internet and integrate results from other AI modules. There is API access, and you can also create your own GPTs, which are predefined contexts that tie the conversations to a given topic.
Over the past year there has been a lot of focus on communicating effectively with ChatGPT, so-called “prompting”, and a new role has appeared: “Prompt Engineer”. This field deals with the practical activity of setting a context in a way that makes ChatGPT answer in the direction you want, techniques for ruling out bias, and making the communication and the resources needed to find the answers more efficient.
ChatGPT today and going forward
Today I think ChatGPT is a super useful tool, a faithful companion when you are working on problem solving, analysis or creative work. If you manage to identify the good areas of use, and you have learned to ask questions effectively, you can free yourself from most trivial tasks and spend your energy on more valuable work.
Still, ChatGPT is just a scaled-up version of the models I used when I studied long ago. It will not manage to create anything truly revolutionary or new. It is limited to combining the sum of all the existing ideas it knows about. That is very useful, but you cannot expect to cure cancer, travel out of the solar system or get eternal, clean energy any time soon.
I expect ChatGPT to be able to handle larger contexts, get more “memory”, and contribute insight into more complex problems. I also think the need for “prompt engineer” skills is temporary, and that ChatGPT will learn to interpret more natural communication as it develops better intuition.
It will be exciting to follow along!