“Most of the time, when searching for information about anything on Internet, you get lots of results that are mostly not exactly what you were looking for. A search through these results is needed until you find what you really need, wasting time that could be dedicated to something else. The purpose of the project is to build a mechanism that applying Artificial Intelligence techniques, would create agents capable of showing only relevant information to the user.”
It is an excerpt from my final degree project, “Content Personalization and Subjectivity Capture using Case-Based Reasoning and Fuzzy Logic techniques”, presented in 2001 at the University of Girona.
The goal was to build a system able to customize content, based on user preferences, trying to catch their subjectivity. That means not only capturing what things a person may like but also in which degree.
Agents Inspired, a spin-off company founded by my project supervisor and other teachers, had the same goal. Having the opportunity to work on something really close to my degree project was something that I couldn’t pass up. So when my supervisor told me if I would like to join I had no doubt!
During this intense period, I learned an important thing about AI: it is as “simple” as applying maths to a set of data. A system with a well categorized set of data and a good mathematical model should be able to provide personalized content to the user.
One of the formulas to calculate the distance between profiles that was in my final degree project. Please, don’t make me explain it, I lost my memory about it!
But this is just one part of the story. The real challenge in any AI system is the processing power needed to be able to perform these calculations. This problem can be mitigated by applying better algorithms and fine-tuning the databases, as we tried in our old AI system, but at a large scale this is not enough.
The growing number of AI Data Centers around the world, which are causing a huge environmental impact, will probably set the limit of what we can achieve with AI. It doesn’t matter how good the mathematical model is or how good the quality of the dataset can be. AI power requirements will push it to the limit of scaling.
Dr. Sbaitso (1992), one of the ancestors of ChatGPT
Some years later, after leaving the company, I read about the RSS format and I immediatly saw the potential to apply AI to create an agent that, reading information from several feeds and learning from my interactions, could create some sort of a personalized “newspaper” by leveraging the categorization feature of RSS and analyzing the words from each entry.
So I built a Java Swing application that worked nicely. But only at the beginning… I quickly realize that the more things the system learned, the less it knew about me. This was due because, like anyone else, I have several interests and as the system widened its knowledge about me the distribution of probabilities became so similar that it couldn’t tell what was actually interesting.
This was another important lesson learned about AI and data: more data doesn’t necessarily mean better results. But there is another important issue, AI as a recommendation system is limited by what it knows about you. What about those things that are far from your taste profile but you may like? How can you broaden your musical taste, for example, if the system never recommends you music genres that you have never listened to?
Apart from this, during my professional career and also for personal interest, I took courses and trainings about AI, Machine Learning, Natural Language Processing, …
This personal experience doesn’t make me an AI expert, that is far beyond my capabilities, but I hope it provides me with enough judgement to avoid the hype around it. My genuine interest is because I truly believe that we can find use cases where this amazing tool can be useful but not in the way that is currently presented.
Current AI is just a “wordjoiner”, it can’t reason or understand anything, it just joins words that look good together. So that is why It is hard for me to understand why people blindly trust in this systems when the result is totally impredictible.
Cover Image is from the film War Games (1983) where and AI called Joshua could provoke a global thermonuclear war
Top comments (0)