Tech Talk: 2023 – The Year of Artificial Intelligence
 | Author:
Victor Sample
Vic Sample: MT43 News Treasurer |
2023 – The Year of Artificial Intelligence
Victor Sample
This has been the year of Artificial Intelligence. All the tech publications I see have AI articles in every issue. The US government has been researching laws on AI. Several countries, including the US, attended a conference on regulating AI.
I have seen political leaders and tech industry leaders (Bill Gates has been very active) touting Artificial Intelligence AND issuing warnings about the dangers of Artificial Intelligence.
I read one interesting article from a Wall Street reporter that AI might totally destroy the world stock markets. According to the author, most of the large Wall Street firms and hedge managers will start using Artificial Intelligence to make trades – and since they will likely be using the same AI engine, AI could destroy the stock markets worldwide.
Last January, I was in a meeting in Butte. The meeting was specifically to create a new Mission Statement and Vision Statement for Southwest Montana. After hours of struggling with ideas and wording, someone from a marketing firm brought up ChatGPT (an Artificial Intelligence program) and asked it to generate a Mission Statement for a tourism organization.
WOW! The mission statement the AI program generated was really good. It was pretty impressive. We did NOT use the generated mission statement; but, we did scavenge some of the ideas and wording from the generated mission.
Both the Microsoft Bing search programs and Google made Artificial Intelligence programs available for asking questions (searching).
Out of curiosity, I started using both the Bing AI program and the Google AI program every morning to see how they really performed. I was underwhelmed!
Where the Artificial Intelligence programs really shine is as Generative Artificial Intelligence – where they generate some sort of writing. When we asked for a mission statement we got a very good statement. Asking a Generative AI program to generate an essay is likely to work very well.
The AI programs are also actually pretty good at conversation. If you get into a conversation with a really good Generative AI program it is hard to tell that it is not a human being. They respond and carry on a conversation quite well.
Where they fail terribly is learning and giving correct answers. I have asked the AI engines technical questions and gotten terrible results. Sometimes they do answer correctly, but many times the answers are just plain wrong. When I pointed out that what they answered was wrong and what was wrong, they would say I was correct and that what they said was wrong and give me a new answer – quite often another wrong answer. Sometimes, after repeated iterations of wrong answers, they would come up with the right answer; sometimes they would just go into apologizing for the wrong answer and then repeat the wrong answer.
Quite often, after giving wrong answers repeatedly, the AI engines would just quit answering and say “I am just a Large Language Module and can’t answer questions” – which was actually more correct than the answers I got.
One question I asked the Microsoft AI program every morning was “What is the current newspaper in Townsend Montana”. Usually, the answer was the Broadwater Reporter OR the Townsend Star. Quite often the answer would mix the two: “The current newspaper is the Broadwater Reporter. The Broadwater Reporter started publishing in 1899.”
Every morning I would respond that the Townsend Star stopped publishing in 2013 and the Broadwater Reporter stopped publishing in 2022. It would acknowledge I was right and then tell me something nonsensical. Every morning I would ask “How about MT43 News?”. Then it would tell me the newspaper in Townsend is MT43 News and tell me it was dedicated to reporting the news in Townsend and Broadwater County and give me links to the Facebook page.
After months, it did start listing MT43 News along with the Broadwater Reporter (or sometimes the Townsend Star). I would again respond that whichever had stopped publishing and the date.
Every once in a while, it would actually just give me MT43 News.
When it regressed back again, I would respond that “Yesterday you knew about MT43 News and today you seem to have forgotten”. Generally, the response was “I’m just a poor Large Language Model and don’t know anything about that”.
Once, I asked it to list the 10 largest towns in Montana by population. The list flashed on my computer (I was able to read the first 2-3 entries) then it went away and I got “I am just a Large Language Model and can’t answer that question”. When I pointed out that it actually did answer the question it responded “Oh, you are correct. I apologize for the confusion.” then actually presented the list again.
If the AI danger to the world is that it might write essays for college students, then there is a clear and present danger.
For anything else, the only danger is people just believing that the information it spews out (when not stating it’s “just a Large Language Model”) is correct.
Even though Artificial Intelligence has been worked on for many decades (at least 50 years), it is still in its infancy. And the “learning” what the AI engines are doing seems to be based on information distilled from the internet – and we all know that just because it’s on the internet doesn’t make it true OR correct.
I do actually use a form of Artificial Intelligence that does work extremely well. Every week after we publish MT43 News, I run a program that “indexes” each article. The program analyzes the article and returns a list of words and what type of grammar they are. I look for just the nouns. In addition, it returns very useful information – for instance, if the article contains the name “Royal Smith”, I get the proper nouns “Royal” and “Smith” but it also recognizes that it is a name and returns “Royal Smith”.
The information that I get back about the article is saved in a database and can be used to search the articles for specific information.
For instance: I could ask for all articles that include the name “Royal Smith”. When I do that, I can distinguish between articles that contain the name “Royal Smith” and articles that just contain the word “royal” or “smith”. That allows me to rank the results and display the most relevant article that contains “Royal Smith” vs. the articles that just contain the words royal and/or smith.
As Artificial Intelligence continues to develop it might become a concern, but the state of the AI I now is pretty weak.
Just be aware and know that articles you read or conversations you have with “support people” might be generated by Artificial Intelligence – and beware of the information we get from that article or “support” person.
