Just for yuks, I asked “What does “condemned to repeat it” mean?”
In 15 seconds, this is what came back:
“The phrase "condemned to repeat it" is often used to refer to the idea that history has a tendency to repeat itself, and that if we do not learn from the mistakes of the past, we are doomed to repeat them. The phrase is often attributed to the philosopher George Santayana, who wrote in his book "The Life of Reason" that "Those who cannot remember the past are condemned to repeat it." This means that if we do not learn from the events of the past, we will be doomed to make the same mistakes again in the future. The phrase is often used to encourage people to study and learn from history in order to avoid repeating the same mistakes.
It's not a scintillating answer, but it’s on point and plausible. Let me just say that I have seen worse answers from college freshmen.
There are still quite a few limitations and rough spots, but, like all machine learning; it will get better fast. Indeed, that is the whole point of ChatGPT: to train the machine to do better.
This public splash has engendered all sorts of chat about the coming age of AI (as to which, see my post from August, 2021 here). But, I’m not going to go down the road of bewailing the end of humanity etc, etc; nor go into detailed analyses of the current state of AI in general or ChatGPT in particular. It’s way too early to be parsing the answers it’s giving at this stage. We (society in general) need to pay attention and start thinking about how to handle this capability. There are a wide range of implications; not least: how to deal with the “It must be true, I read it on the internet!” in our truth-challenged discourse. But let’s not panic quite yet.
Even for those of us in the teaching (or, more precisely, the evaluating/assessing/grading) biz, the challenge is rapidly approaching, but not imminent. Simply put, if a student can grab the prompt from a take-home essay exam (which I usually use) and drop it into their friendly I responder, will I get back anything I can use to understand what the student actually absorbed out of the course? Do I really want to spend twenty minutes chasing down the source of any individual answer to an exam? Grading is already draining enough.
Most tests “back in the day” were “in class,” usually closed-book, and were as much about memory capabilities as about reasoning and understanding. I realized some time ago in my teaching that memorization wasn’t all that important at the college level, and turned to open-book, take-home tests where the student would have to marshal ideas and information to come up with an answer showing some insight about the material and issues covered in the class. Then, along came Wikipedia and other internet sources which provided all sorts of “facts.” So (it seemed to me) that there was not only little point in forcing students to memorize facts that they could easily look up, but that I didn’t care (again at the college level) about them remembering whether Napoleon was exiled to St. Helena in 1815 or 1816.
At least (so I figured) I can still pose interesting questions that require thought, reflection, integration of sources and ideas, etc. Students can pull their facts from Wikipedia and the course materials and show me that they “got it” when I talked about the impact of guilt on the treatment of perpetrators and collaborators in the aftermath of WWII in Europe.
Well, that seems to be going by the boards soon, too. ChatGPT provides some decent answers to questions like: “How has democracy changed from Ancient Athens to the present day?” or “Why don’t we have so many political revolutions anymore?” Clearly, I (and teachers everywhere) have some serious work ahead in reconfiguring our exams and other assignments. As a good friend of mine points out, we need to improve how we assess students already and now we have another incentive. More, we can use the defects in ChatGPT to point out the difference between an intelligent answer and one that is merely coherent. In any event, the academic “arms race” is escalating!
One of the tools already in many college teachers’ arsenal is a service called “Turnitin,” an antiplagiarism software program that takes students’ electronically submitted essays and compares them to all the material on the internet, including Wikipedia, scholarly articles, and papers from students at other universities across the country. It’s very helpful, but it is a ‘dumb’ tool, it just matches words. Now that ChatGPT and its progeny will start writing student essays (slightly different every time), it’s about to go the way of the bi-plane. Fortunately, there is a new site that promises to be able to detect when an answer is written by an AI like ChatGPT. It’s called Originality .
I haven’t tested it yet, but I am quite curious to see how it develops. Competition (both economic and techie) being what it is, I can see a serious escalation of software vs. software intelligences coming up.
All of which leads up to the title of today’s posting: Grand Turing Test. The original concept was developed by the brilliant British mathematician Alan Turing in the mid 20C, at the very dawn of the computer age. He posited that computer “intelligence” could only be determined by a human who would pose questions to an intelligence (human or machine) in another room and, if the questioner was unable to tell whether the answers came from a human or machine, then the machine was, in fact, intelligent. This is the goal towards which AI has striven for several decades.
I’m proposing that we will soon be facing a slightly different version, i.e. whether one computer/software program/AI (like Originality) can tell whether an essay is written by a human or by another AI (like ChatGPT). Both sets of programmers will be beavering away to instill even more intelligence-appearing (or intelligence-detecting). Is there an end in sight? Doubtful.
In the meantime, I hope to continue writing these essays with just enough quirkiness and insight that it will be some time until I can be replaced by an AI-blogger. Or, perhaps, I will just find a program, drop in a couple of prompting words and tell them to riff for 1100 words or so.
Or, maybe I did already.