Artificial intelligence (AI) appeared in my life in the early 1990’s. It was essentially limited to LISP because it was part of the Slackware Linux distribution. I confess that I did not fully grasp the concept and, since I did not see its usefulness, I simply ignored it.
AI made a second appearance in 2000 as of IBM's Via Voice demo version of which was included in the Suse Linux distribution. This application was among the first to allow me to dictate to my computer. It worked very well. It is not, strictly speaking, AI in the common sense of the term but it certainly gave the impression. I told my 6th grade French teacher that in the year two thousand I would not need to write well anymore because there would be a machine to do it for me. Here it is; I was right! This conversation had taken place because my handwriting is, let’s say, cryptic at best...
It was three years ago that AI had a serious impact on my life. In fact it was rather the understanding of its overall impact that hit hard. I went back to school. then realized that more than half of the students used ChatGPT both for homework and during exams.
Besides the feeling injustice -- I did not resort to it; by choice-- there was the question of the degree's real value. It no longer represented the graduate’s overall knowledge as much as his ability to use a single tool. I also understood that, as a result, all graduates will eventually be redundant and therefore unemployed. When the AI does 90% of the job, no one will want to pay for more than one pinch of graduates to lead. And even so.
We fear the unknown. The idea that a machine can give the impression of reason is already staggering but what worries me the most is that this machine learns to reason from human knowledge. Human are, very generally, not very intelligent. They are endowed with reason but is greatly influenced by feelings. It is also very prone to preconceptions, prejudices, advertising, etc.
These machines ingest a lot of data during their training. Data from scientific disciplines tend to be factual and follow rigid protocols and laws. Trained in this way, AI can be very valuable. When data includes social media posts, that’s where it’s most likely to become dangerous. The mythical collective intelligence is saturated with unqualified opinions, religious, moral, social, etc.
Then there is the problem of censorship. To be socially acceptable, AI must operate among the limits of human social codes. This restricts output to a given scope, if the answer goes against code, it will be omitted however true it may be.
But despite everything, I doubt t AI will go away. So I will continue to explore this phenomenon and deepen my knowledge.
I already have some notions about the implementation, the structure, and the use of some software such as Ollama. For now, I mainly interact with local LLMs on a P7 computer with 16G of RAM without a graphic accelerator. It is from 2011.
I also interact with online AI but use the Tor network to maintain anonymity.