Researchers have successfully developed an AI system capable of completing mathematics problems at a grade school level, a new report asserts.
Traditionally, while AI models are proficient at manipulating language to formulate sentences, the multi-step reasoning required to solve math problems has been a step too far.
However, researchers at OpenAI (the company behind language model GPT-3) say they have trained a model to recognize its own mistakes, which means it can repeatedly reassess until it discovers a workable solution.
In testing, the AI system was able to solve almost as many problems as a sample of children between the ages of nine and twelve. The children scored 60% on a test drawn down from the OpenAI database, while the AI system scored 55%.
AI takes on mathematics
Although grade school mathematics problems are simple enough for most people to complete with ease, OpenAI says the arrival of AI models capable of solving even basic math challenges is a major step forward and will unlock a number of opportunities.
“One significant challenge in mathematical reasoning is the high sensitivity to individual mistakes,” explained the researchers. “Autoregressive models, which generate each solution token by token, have no mechanism to correct their errors. Solutions that veer off-course quickly become unrecoverable.”
OpenAI worked around this problem by training a set of verifiers, the role of which was to evaluate the answers produced by the AI model. These verifiers were given 100 potential solutions, all generated by the model, and were then tasked with determining whether any were correct.
“Providing correct arguments and recognizing incorrect ones are key challenges in developing more general AI,” OpenAI added. “[Grade school] problems are conceptually simple, yet one subtle mistake is enough to derail an entire solution. Identifying and avoiding such mistakes is a crucial skill for our models to develop.”
The company believes the verification system that allows its AI systems to solve simple math problems with relative accuracy will become increasingly important as AI is applied to more complex domains.
In combination with advances in the field of semiconductors, which will make possible AI models that are many times larger (and therefore more capable) than they are today, the ability to tweak the way in which AI approaches a problem could prove transformative.