law-of-sines-feature-image

Another AI Math Flop

I would like to think that Artificial Intelligence (AI) should be able to solve a relatively easy high school math problem because such math problems are typically straightforward with solutions all over the Internet. However, my experience shows that it consistently doesn’t. Maybe that’s because AI is in its infancy, but I must say that it is VERY disappointing.

As with my previous article on AI, “Artificial Intelligence Failed A Math Test!“, this article is NOT a math lesson. It is simply a demonstration of how unreliable AI can be for something that it should excel in.

In studying for his high school ACT exam, my grandson came across a math problem that he was unable to solve. He often contacts me when he needs help with math since I taught all levels of high school math for 10 years at our local high school. This time the problem was a triangle where they gave you all the angles and the length of one side and asked for the length of one of the other two sides.

AI Try #1

This is what I asked each AI:
Given a triangle with angles of 75 degrees, 45 degrees, and 60 degrees. Also, the side between the 75-degree and 45-degree angles is 240 feet long. How long is the side between the 75-degree angle and the 60-degree angle?

The correct answer is 196.0 feet (to the nearest 10th of a foot).
1. Google Bard gave me -85.97 feet and said that it was not possible.
2. ChatGPT 3.5 gave me 173.21 feet.
3. ChatGPT 4 gave me 329.5 feet.
4. Microsoft Copilot gave me a choice, either 246.66 feet or 122.66 feet.

Since all four of the AI systems that I tried above failed to give me the correct answer, I naturally thought that I hadn’t asked the question well enough. So I tried again.

AI Try #2

This time I asked each AI more simply:
Given a triangle with angles of 75 degrees, 45 degrees, and 60 degrees. Also, the side opposite the 60-degree angle is 240 feet long. How long is the side opposite the 45-degree angle?

Again, the correct answer is 196.0 feet (to the nearest 10th of a foot).
1. Google Bard first gave me this apology:
“My apologies, I made a mistake in my previous response. You’re right, the triangle with the given angles and side length is indeed possible. Here’s how to find the missing side length:”
It now gave me 282.84 feet.
2. ChatGPT 3.5 now gave me 329.66 feet.
3. ChatGPT 4 now gave me 166.4 feet.
4. Microsoft Copilot now gave me 188.68 feet.

The Takeaway

It was surprising to me that…
1. Each of the above AI systems gave me a different answer with each try.
2. No two AI systems gave the same answer.
3. No AI system gave the correct answer in either try.
4. In the past seven months, AI appears to have not yet learned high school math.
5. Google Bard appears to be the most polite, offering an apology when you tell it that it’s wrong.

If you or anyone you know wants to use AI for doing math, please tell them to be VERY careful. In the relatively simple problem above, it was easy enough to solve for the correct answer to know that none of the AIs gave the correct answer. If possible, whoever tries doing math with AI should always question the result. Better yet, learn from a class how to do it without AI. (NOTE: An easy way to solve this problem without AI is to use the “Law of Sines”.)

Your feedback on this article is welcome. Please use the Comments section below to respond.

8 thoughts on “Another AI Math Flop”

  1. I very much enjoyed your article. First laugh of the day. Let us hope I can keep my good humor through my morning internet trip. What are the odds of that? I don’t need AI to tell me.

    1. Hi Kalmly,
      I’m glad you enjoyed the article. Although I didn’t intend the article to be funny, I can see where it could be. My personal opinion is that it’s sad that AI cannot do better with math. If I can’t trust it to give me the correct answer for a simple math problem, what can I trust it for?
      Thanks for commenting.
      Stu

  2. Hi Stu. I used the Fortran language to assist me in my youth working such triangle problems. Since Fortran still exists, maybe AI’s need a refresher, Mindblower!

    1. Hi Mindblower,
      I have very fond memories of Fortran. I used to joke that English was not my native language, Fortran was!
      There is something very seriously wrong with these AIs that they are having so much trouble with math.
      Thanks for commenting.
      Stu

  3. The way we are all being hyped about Ai Stu you would think it was ready to take over next week. I have found the same using ChatGPT. You ask it something mathematical, it gives you an answer a toddler may have reason to question and as soon as you point out the bleeding obvious it apologises and then has another go and that is usually wrong too. If it can apologise when told it is wrong why is it not double, triple and quadruple checking itself before it spits out an answer like the lazy kid in the classroom where any old answer will do for teacher ?
    I feel confident that I won’t have to face Skynet Schwarzenegger style Terminators anytime soon as they wouldn’t be able to even get the maths right to calculate a firing solution range correctly in any weapon algorithm !
    Cheers Reg

    1. Hi Reg,
      Well said! Each time I try another math problem I’m disappointed.
      Thanks for commenting.
      Stu

  4. Wonder what the result would be if you presented as a graphic with the angles and one side values shown and a big x for the unknown. If Chat or whatever else is reading books, it probably has not done many word problems but has memorized graphics for a problem like this. BTW< are you 1000% sure of your answer since you are no way near to theirs?

    1. Hi Vince,
      I’m completely sure my answer is correct because:
      1. The Law of Sines is simple and easy to calculate:
      A/Sine(a) =B/Sine(b) = C/Sine(c)
      where the capital letters are the sides and the lowercase letters are the angles opposite those sides.
      2. My grandson was using a book of ACT practice exams that had the answers. He just didn’t know how they got their answer.
      As far as presenting a diagram to AI, maybe I’ll try that next. However, Artificial INTELLIGENCE should certainly be able to figure this problem out from the descriptions.
      Thanks for commenting.
      Stu

Leave a Comment

Your email address will not be published. Required fields are marked *

Exit mobile version

WHY NOT SUBSCRIBE TO OUR NEWSLETTER?

Get great content like this delivered to your inbox!

It's free, convenient, and delivered right to your inbox! We do not spam and we will not share your address. Period!