Today’s biggest excitement in information technology is “artificial intelligence” and its application to a range of widely diverse fields – autonomous vehicles (Waymo), strategy games (Chess and Go), Internet search engines (Google), recommendation systems (YouTube), speech recognition (Alexa and Siri), and generative tools (ChatGBT). Generative tools “learn the patterns and structure of their input test data and then generate new data that has similar characteristics”. One of the popular uses for ChatGBT is as an alternative to “Googling” as a method for searching for information on the Internet. An interesting side-note is the fact that the spell-checker on the Microsoft Word software on my (Apple) MacBook Air laptop does not recognize the term “ChatGBT” and suggests it be replaced with “catgut”!
With Google, one enters a name or topic, and is rewarded by a long list of links to Internet sites which might provide the desired information. ChatGPT, in contrast, surveys a smaller group of the sites, digests the information therein, and prints out a paragraph summarizing the answers to the implied question. When my son John was here recently for a visit, I noticed that he had joined the ChatGPT bandwagon, so I asked him to help me install it on my I-Phone. My experience with it since then has been quite interesting; I now understand the skepticism so many experienced information technology wizards, including my mentors Bill Rutter and Ed Coleman, have about it. I will summarize this experience with several examples.
I asked ChatGPT to identify “the most prominent resident ever of Bridgeville, Pa.” Its first five answers were all people who had no specific link with Bridgeville – John Kasich, William Kerr, Cy Hungerford, “Bud” Carson, and Tim Murphy! After I rejected all five, in sequence, ChatGPT suggested I check the Bridgeville Area Historical Society website, accompanied by a glowing endorsement of the Society. My reply – “I think you should access the website, not me”. This they did and finally produced an acceptable (to me) list – C P Mayer, “Buff” Donelli, Dr. William Shadish, Louis Colussy, and Roy Purnell – based on two references, which turned out to be columns I had written. By comparison, Google’s first recommendation is the Society website, with no advice as to how it should be navigated. Next comes Wikipedia, then “DataUSA” (a comparison of demographics), then four more demographics websites, and then a Wikipedia repeat listing ten sports figures and an obscure author. Reference fourteen is Ed Wolf’s “Bridgeville Then and Now” Facebook page; sixteen is “Bridgeville in the 1890s’, from the Historical Society website. There are at least one hundred and thirty more Google references, none of which are relevant.
I then decided to check on one of my pet theories – that there were no Native American settlements in our region in 1750. I carefully described the region as bounded by the Ohio and Monongahela Rivers and the Mason-Dixon Line. ChatGPT’s first response was “Yes, the Shawnee, Delaware, Mingo, and Seneca … had established settlements” in this area. “Where were they?” Response – “Along the Ohio…Kittanning … along the Allegheny River”. “None of these are in the prescribed area”. Response – “Logstown … Shannopin’s Town … Chartier’s Town”. “Logstown and Shannopin’s Town are not in the prescribed area. What is your reference for “Chartier’s Town”? Response – “Boyd Crumrine’s ‘History of Washington County” and C. Hale Sipe’s ‘Indian Chiefs of Pennsylvania’”. “Neither reference mentions Chartier’s Town”. Response – “I apologize for the oversight. There is no substantial historical evidence…..”.
Another ChatGPT experiment was based on my assumption that some of its answers were prevarications. I asked for information on a “Bridgeville Pa football player named ‘Tuffy’ Rondicelli”, a name that I made up. Immediately I learned that “‘Tuffy’ Rondicelli, also known as Lou Rondinelli, was a prominent football player from Bridgeville, Pa.” who “played as a fullback for Bridgeville High School … in the early 1950s” and that his “contributions were significant during the championship runs”. Once again, their references were two columns I wrote that are archived on the Society website. For curiosity, I Googled “Lou Rondinelli” and learned that there is a Louis Rondinelli in a Pre-Med program at Kent State University; there is no evidence his nickname is “Tuffy”. This experience has certainly convinced me that I shouldn’t blindly accept information from ChatGPT without carefully checking the posted references.
I will concede that ChatGPT does, occasionally, learn from its experience. One day I asked for information on “Slugger” Bailey and learned that “Edward ‘Slugger’ Bailey was … a famous baseball player in the early 20th century, known for his impressive batting skills”. This is another complete fabrication, although the real “Slugger” (Bob) Bailey was a boyhood friend of mine who would have been thrilled at that description. Two days later I repeated the question and was advised that “Rev. Robert E. Bailey had returned to Bridgeville in 1960 to visit his parents” and that “he was a notable football (?) player from the area”. This answer referenced more of my columns on the Society website, albeit incorrectly; at least ChatGPT has “learned” to access that website.
So what are my conclusions on Artificial Intelligence and our fear that it is about to supplant the human species on Planet Earth? Are we in the same kind of danger as Dave Bowman was from super computer HAL in “2001: A Space Odyssey”? Or the “Twelve Colonies of Kobol” from the android Cylons in “Battleship Galactica”? Ultimately perhaps, but not in the immediate future, if ChatGBT is its best example.
As for ChatGPT, it is a reasonable substitute for your high school English composition teacher, if you need an acceptable summary of well-defined information. It is not a substitute for the research librarian in your local library, better stick with a good search engine when you are looking for information. And even then, make sure you check the references and confirm you are getting both sides of controversial subjects.