Published Date: 28/07/2024
In recent years, artificially intelligent chatbots have been praised for their remarkable language skills, but they often struggle with math. Chatbots like OpenAI's ChatGPT can write poetry, summarize books, and answer questions with human-level fluency, but when it comes to math, they can be error-prone.
This is because these systems are fine-tuned for determining probabilities, not doing rules-based calculations. Likelihood is not accuracy, and language is more flexible and forgiving than math. According to Kristian Hammond, a computer science professor and AI researcher at Northwestern University, 'The AI chatbots have difficulty with math because they were never designed to do it.'
Historically, computing has been defined as 'math on steroids.' However, with the advent of neural networks, a different approach has emerged. Neural networks, loosely modeled on the human brain, generate language by predicting what word or phrase is most likely to come next. However, this approach has its limitations, particularly when it comes to math.
Math word problems that require multiple steps to reach a solution often stump AI chatbots. To address this, companies like Khan Academy have made significant changes to their AI-powered tutors. For instance, Khan Academy's AI chatbot tutor, Khanmigo, sends numerical problems to a calculator program instead of asking the AI to solve the math.
Similarly, ChatGPT has been using a workaround for some math problems, summoning help from a calculator program for tasks such as large-number division and multiplication. OpenAI has acknowledged that math is an 'important ongoing area of research,' and its scientists have made steady progress in this field.
The erratic performance of AI chatbots in math has sparked a debate in the AI community about the best way forward. Some believe that advanced neural networks, known as large language models, are the key to steady progress and eventually to artificial general intelligence (AGI). Others, like Yann LeCun, chief AI scientist at Meta, argue that a broader approach is needed, one that involves 'world modelling,' or systems that can learn how the world works much like humans do.
information
OpenAI is a research organization that aims to promote and develop friendly AI that benefits humanity.
Khan Academy is a non-profit education organization that provides free online education to anyone, anywhere.
Northwestern University is a private research university located in Evanston, Illinois
OpenAI is a leading research organization in the field of artificial intelligence. Its mission is to ensure that AI systems are safe, transparent, and beneficial to humanity.
Khan Academy is a non-profit education organization that aims to provide a free, world-class education to anyone, anywhere.
Northwestern University is a private research university located in Evanston, Illinois, known for its academic excellence and innovative research initiatives.
Q: Why do AI chatbots struggle with math?
A: AI chatbots struggle with math because they were never designed to do it. They are fine-tuned for determining probabilities, not doing rules-based calculations.
Q: What is the approach used by neural networks to generate language?
A: Neural networks generate language by predicting what word or phrase is most likely to come next, much like humans do.
Q: How has Khan Academy addressed the math limitations of its AI chatbot tutor?
A: Khan Academy has made significant changes to its AI-powered tutor, sending numerical problems to a calculator program instead of asking the AI to solve the math.
Q: What is the dominant view in the AI community regarding the best way forward?
A: The dominant view is that advanced neural networks, known as large language models, are the key to steady progress and eventually to artificial general intelligence (AGI).
Q: What is 'world modelling' as proposed by Yann LeCun?
A: World modelling refers to systems that can learn how the world works much like humans do, an approach that Yann LeCun believes is necessary to achieve true artificial intelligence.