By: Lauren Wu

In the school year that ended recently, A.I. does well with summarizing books and writing poetry, but has struggles with math. Technology specialists say this was because the AI programs were never designed to do this.

There are two sides who disagree on how to deal with these chatbots that can’t do math well. Some people believe that the large language models that power A.I chatbots now are basically the singular path to constant progress and ultimately artificial general intelligence (A.G.I.), a chatbot that can do most things our brains can do. The difference is that while A.I. is only good at doing one specific thing, A.G.I., after gathering a lot of knowledge, can do many things, like a human.

On the other hand, there are skeptics who question whether adding more data and computing firepower to large language models is enough. For example, Yann LeCun, the chief A.I. scientist at Meta. He thinks the A.I.s are lacking in logic and common sense, and the models need to learn how the world works like humans, mostly by watching it, and a bit by acting in it.

Some of the world’s smartest software engineers have made A.I. chatbots that are better at the arts than mathematics. For example, Wojciech Zaremba and John Schulman, both research scientists at OpenAI have made GPT 4, which is not the best at math. This is very different from the past, where computers invented in the 1940’s were accurate and fast with numbers. Now, chatbots have become skilled at the arts and had a setback with math.

Some chatbots, like Open A.I.’s ChatGPT, can do many things, such as poetry and answering questions. However, when a math problem requires multiple steps, these chatbots don’t do very well. Kristian Hammond, a Computer Science professor and Artificial Intelligence researcher at Northwestern University said, “The A.I. chatbots have difficulty with math because they were never designed to do it.”

A few months ago, Khan Academy, a software to help you learn math, significantly altered their A.I. chatbot tutor, Khanmigo. Rather than having the A.I. solve math problems, Khanmigo sends the mathematical equation to a calculator software. This is helpful because a calculator program is designed to do math, while Khanmigo is not.“We’re actually using tools that are meant to do math,” said Dr. DiCerbo, a well-known educator and researcher in the field of educational technology and digital learning. He remains confident that conversational chatbots will play an important role in education.

For over a year, ChatGPT has also used this strategy for math problems. The improved version (GPT 4) has accomplished nearly 64 percent accuracy on a shared database with thousands of problems needing visual perception and numerical reasoning, which is 58 percent greater than the earlier version, which was 6 percent.

Sources: https://www.nytimes.com/2024/07/23/technology/ai-chatbots-chatgpt-math.html https://www.nytimes.com/2024/07/23/technology/ai-chatbots-chatgpt-math.html