As the world tries to keep up with the rapid advances of artificial intelligence across all industries, some experts are already setting their sights on the next version of the technology known as artificial general intelligence. Although there is no consensus on an exact definition of AGI, it generally refers to highly autonomous systems that match or surpass the capabilities of well-educated adults.
Compared with generative AI — AI models that are trained to produce text, images or code based on prompts — AGI implies a “target level of capability for general problem-solving, such as reasoning, planning, sensing and acting, across many domains. People often consider generative AI as an early and incomplete version of AGI,” says Sheng Li, Quantitative Foundation associate professor of data science at the University of Virginia.
Click the banner below to get a better read of the AI landscape.
How AGI Could Transform Education
If achieved, AGI could have powerful applications throughout K–12 and higher education, with the potential to entirely reshape teaching, learning and assessment. AGI agents could serve as the ideal teaching assistants and learning partners by adaptively learning from context and interactions, planning teaching and learning activities, and using feedback to improve over time. AGI might also redefine educational assessment, reshaping both testing standards and the design of assessment content.
“With potential AGI technologies, K–12 students could be paired with customized AGI agents to improve learning outcomes,” Li says. “Also, the curriculum would be adjusted to emphasize more higher-order skills, such as problem-framing, critical thinking and human-AI teaming.”
Of course, as with current AI models in the classroom, if AGI is applied to education, Li says, guidelines and standards for safety, ethics and governance will need to be developed.
SUBSCRIBE: Sign up to get the latest EdTech content delivered to your inbox weekly.
The Current State of AGI
There have been some advances in the development of AGI, but these have been open to debate. In April, researchers at the University of California San Diego found that OpenAI’s GPT-4.5 model passed the Turing test better than a human 73% of the time. While the Turing test is one way to demonstrate that AI can match or surpass human intelligence — essentially achieving AGI — the researchers concluded that the test results did not prove that the AI bots had reached human-level intelligence. Instead, they were evidence that the large language models could potentially substitute for humans in interactions of limited duration.
Current AI models still struggle with long-horizon planning, causal reasoning and close alignment with human values, says Li, but he points out that there has also been “some exciting progress toward AGI.” One example: Large language models and multimodal LLMs can now reason, code, use tools and even perform on par with gold medalists in the International Mathematical Olympiad.
While AGI may eventually become a reality, Li thinks the timing is uncertain.
“The realization of AGI would require algorithmic breakthroughs, better memory/context architectures, awareness of the physical world and possibly new computing paradigms,” he says.
