Seen in Apple News this amazing article about the weaknesses of artificial intelligence in mathematics. Artificial intelligence (AI) systems like ChatGPT still perform poorly in mathematics.
In short: AI systems like ChatGPT were developed based on large language models (GML) and not based on mathematical models. In other words, we must clearly distinguish that mathematics and computation or computer processing are far from identical. AI systems use their computing power to calculate probabilities and decide which words to select from their massive database to form a meaningful sentence. While mathematics is based on logical thinking and pure logic.
The “probability” that 2 + 2 = 4
So, based on probability, AI is a model that does not take into account the concepts of “true” or “false” as in mathematics.
So, to determine that 2 + 2 = 4, AI models rely on probability. To get to the number 4, the AI model will not reason like we humans do, but will search its mega database for the most likely answer to the series of numbers and characters in “2 + 2”, namely the number 4.
Illustration Adobe Stock
At best, AI succeeds in imitating the logical thinking of mathematics, while the deductive thinking of concepts such as numbers, sets and figures in mathematics still remains one of the last bastions that still resists “artificial intelligence”.
It is easy for us to design a triangle, but for an AI machine it is much more difficult.
In December 2022, upon its launch, many quickly discovered the limitations of ChatGPT, which had to solve a simple mathematical problem like this: “If five machines produce five objects in five minutes, how long will that take?” Does it take 100?” Machines to make 100 objects. ChatGPT had replied 100 minutes ago. Fake! The answer to this fundamental problem for a mathematician is five.
PS: To students who get their math homework done using online AI tools, you may be disappointed!