Will ChatGPT Destroy Education?
Surely you have heard about ChatGPT. Right? OK, let’s assume that you know very little about this new artificial intelligence (or as I like to call it — a robot). ChatGPT is an online AI from OpenAI. What does the GPT stand for? Well, you could just ask ChatGPT and it will tell you that it stands for Generative Pre-training Transformer. In short, this is an online robot that you can ask questions and tell it to do stuff for you. Here are some things you could do:
- Write a 1000 word essay about the history of artificial intelligence.
- Create a python program using sympy to find to solve two equations (you pick the equations).
- Describe the ultraviolet catastrophe.
- Create a recipe from stuff that I have in my fridge (you have to give it the stuff).
- Write a sonnet about physics in the style of Shakespeare.
It’s seriously crazy awesome. I’m surprised how well the AI works. You really need to try it out.
Wait, ChatGPT can write my essays?
I guess it can. Let’s try this.
I was just hoping you could give me some insight into the evolution of the market economy in the southern colonies. My contention is that prior to the Revolutionary War, the economic modalities — especially in the southern colonies — could most aptly be characterized as agrarian pre-capital.
Ha. Yeah, that’s from Good Will Hunting. But if you put that into ChatGPT, you do get a nice long response. I guess it would make an acceptable essay for a class. Who knows.
The thing I’m wondering — let’s say I’m a professor and I ask students to reply to the Good Will Hunting quote. If all the students use ChatGPT, would they all have the same essay? I assume so. But still — it seems plausible that at least in the very near future, everyone could have their own personal AI to create their own personal essay.
Can ChatGPT solve physics problems?
Here’s a physics question:
A ball has an initial velocity of 4.2 m/s and is launched at an angle of 33 degrees above the horizontal. If the ball starts 1.1 meters above the floor, how far horizontally does it travel before hitting the floor?
If you give that to ChatGPT, it gives you the wrong answer. It’s mostly correct, it just makes an error when solving for the time using the quadratic equation.
There are two funny things about this AI answer. First, ChatGPT attempts to solve the problem AND produces python code to calculate the answer. I think that’s awesome since I didn’t even ask for python code (I guess the AI already knows what I like). Second, the AI made an error in the program and then tried to fix it (but still failed).
ChatGPT says the answer is 11.62 meters (which is really far for a ball starting at 4.2 m/s) but the correct answer is 2.7 meters (I calculated it myself). Here is the first error the AI made.
It correctly uses the quadratic equation to solve for the time of flight. Good job, robot. However, since the acceleration is in the negative direction, the discriminant should be:
discriminant = vy**2 + .5*g*(y0-0)However, I’m going to give partial credit for this solution because it included the possibility that the time would be imaginary and the ball wouldn’t hit the floor — but the ball will TOTALLY hit the floor.
For the second attempt, robot incorrectly calculates the time as:
# calculate time of flight
t = (2 * vy) / gThat’s just wrong and for some reason it still reports the final distance as 11.62 meters. Weird. Oh well.
I tried giving it another problem.
Wrong again. OK, I actually assumed it would get these questions correct. Let’s just assume that the robot will improve in the near future and be able to solve these kinds of problems. I think that is very likely.
How do we deal with AI in physics courses?
I don’t think this is going to be a huge problem in your introductory physics course. First (as you can see above), ChatGPT isn’t always correct. So, if you gave this as a homework problem it wouldn’t really help students solve the problem.
Second, if the course has a traditional in-class type of written test then ChatGPT doesn’t matter. Right? If they can’t use AI on the test then it’s fine. I mean, they can’t (or shouldn’t use one of these online solution sites that rhymes with the word plegg) — right?
Finally, and here is the most important point — even if the robot gives the best answer possible with all the steps and everything. Is that what you want? Do students solve problems to get the answer? No. That’s not why they solve problems. The goal of the physics problem is to practice and to learn.
Let me give an analogy. Should we have cars? Should we use cars to get around? Yes. Absolutely. But won’t cars destroy running? I mean, who would run somewhere when they could drive instead? Of course humans have had cars (automobiles) for more than 100 years and yet we still run. If the goal was just to get from one place to another then a car would usually be the answer. But that’s not why we run. We run because it’s fun (sometimes yes) and we run for exercise. The same is true for most physics problems — they are fun.
Oh, and what about this for a new physics assignment? Give the students a question (like my projectile motion example) and have them use the robot to get an answer. THEN have them analyze the response. Is it correct? Is the explanation useful? What would you change? That would be a fun activity.
- what is chatgpt
- it’s like a car. Just because I can use a car to get from A to B doesn’t mean I won’t still walk or run
- at the end — chatgpt, write a blog about AI and education
