Last Updated on
ChatGPT is the world’s favourite AI chatbot. The artificial intelligence system is capable enough of answering its user’s queries or requests in the way it believes is most appropriate. However, there are still a few limitations to its work. Can ChatGPT do math?
Can ChatGPT do math?
It might seem a ridiculous question – Can the world’s most advanced NLP artificial intelligence do something that calculators have been doing from the comfort of your parents’ pockets since the 70s?
The answer lies in how an LLM (Large Language Model) like ChatGPT works its magic. While a calculator uses pure arithmetic to process your requests, this numerical logic is not explicitly programmed into GPT 4. The GPT-4 model is only available with a paid subscription to ChatGPT Plus or ChatGPT Enterprise.
Essential AI Tools
Content Guardian – AI Content Checker – One-click, Eight Checks
Jasper AI
WordAI
Copy.ai
Writesonic
While a neural network based on extensive training data is excellent for many things where alternative information sources like calculators and search engines fall short, it has a big downside too.
Attempting to recreate a calculator through machine learning methods is inherently impractical. AI models like GPT 4 learn from examples – and you literally cannot provide an example of every calculation along with the correct answer. That list is, of course, infinite.
Natural language processing tools like ChatGPT have opted to avoid reinventing the wheel. Instead, the arithmetic processing of the humble calculator has been added to ChatGPT and other models like it, which interestingly opens up new ways to use calculators we’d never even thought of. For a start, you can now ‘talk to’ your calculator in natural human language. You can even ask it to perform operations on the operations if that’s faster than retyping it yourself.
Can AI do math?
Most AI systems are insufficient. The very thing that makes artificial intelligence so good at thinking and communicating like a human also makes it less objective and therefore inconsistently good at doing math. To be clear, AI chatbots don’t use pure arithmetic.
Google DeepMind explains that “AI systems often struggle with complex problems in geometry and mathematics due to a lack of reasoning skills and training data.” To solve this problem, “AlphaGeometry’s system combines the predictive power of a neural language model with a rule-bound deduction engine,” which works in parallel as and when a given methodology is appropriate. Google’s team has developed “a method to generate a vast pool of synthetic training data – 100 million unique examples – [and] can train AlphaGeometry without any human demonstrations, sidestepping the data bottleneck.”
Why can’t ChatGPT do math?
We all know calculations or solving math equations require human interaction and some real-time thinking. For example, it would need pondering over a math calculation from different angles and brainstorming to get the correct answer. Therefore, ChatGPT is an automated System like an AI chatbot, although it can collect data from various internet resources. However, when it comes to math, it cannot function as a pure arithmetic calculator.
The natural language processing of ChatGPT blends with the arithmetic, and when you simply want some objective numbers calculated, the latter is only a hindrance.
Notice how, in the image above, ChatGPT fails to replace every instance of the number 5 with the number 6. There are two “5”s, but the AI chatbot only replaced the first of them.
In addition to this oversight, the answer was wrong anyway.
513 * 71.378 = 36616.914
This then divided by 9.3567 equals 3913.44320113
Not 3915.181, as calculated by ChatGPT.
What this also serves to demonstrate is that ChatGPT is not only limited by decimal length in output, but also limited internally for efficient responses across its millions of users. As a result, the errors from decimal shortening carry over to create an incorrect answer.
Can ChatGPT do math? Our final thoughts
Chat GPT can do math, but it’s not as reliable as a calculator.
Natural language processing and arithmetic both have their place, but when combined, you do not get the best of both worlds. Even the word “compromise” is flattering. What you end up with is a calculator that takes opinions, works through a problem by memory rather than pure logic, and even forgets to perform some parts of an equation.
What you end up with is, amusingly, human.
