Spelunking CACM, vol. 9 (1966): Eliza (and now 2022's LaMDA)

 Eliza has arrived. Eliza, who is a few months older than I am, is inarguably one of the seminal events in computing history. It's just about the only program created in the 1960s that most of us can still spontaneously name. It helped to spur my own interest in computing.

My first computer was a Radio Shack TRS-80 Model I. I also got a book with programs in it, which meant copying the programs in by hand. (Good training for catching syntax errors!)  One of those programs was called Eliza, though I suspect the BASIC version in my book was pretty different, and probably much simpler, than the 1966 version created by Joseph Weizenbaum in SLIP, a language adding list processing features to FORTRAN.

Eliza works by parsing input text into tokens and finding an important phrase (where phrases are delimited by commas or periods) based on the key word in the phrase, using a list of important words that was created by hand. Then it replaces the primary verb or subject in the phrase with one of a set of stock phrases. If you type in "I love my dog," Eliza might respond, "Tell me why you love your dog." Does mean it has any idea at all what a dog is, it just found a simple pattern and followed it.

My TRS-80 was at home, but the high school also had one, and friends of mine also enjoyed playing with Eliza. And what do high school boys want to talk about? Sex and scatalogical things and bad language, of course. You could make Eliza say some pretty hilarious things that way, by the standards of high school boys.

And now, 56 years later, we have a Google engineer and AI ethicist arguing that an intellectual descendant of Eliza has become sentient.

Chat bots have evolved tremendously since then, and are generally connected to a backend system of some sort, so that they can provide airline assistance or what have you. Often, they are connected to a neural net-based AI both for parsing the language and creating the responses. Generally speaking, no one believes they are actually sentient, but the responses will sometimes appear so insightful that your hair stands up on the back of your neck.

My skeptic's nature and my own very limited experience with chat bots cause me to lean toward siding with the Googlers who said there is no evidence that it's sentient, and plenty of evidence against it. But we are now clearly entering the realm where it will get harder and harder to tell, and the stakes of the arguments will continue to grow. Research into AI ethics grows more crucial every day.