In recent years, artificial intelligence models have become capable of writing code, passing university exams, solving complex mathematical problems, and succeeding in scientific tests, in addition to handling images, videos, and texts.
Hot Topics
There is no declaration that enjoys scientific consensus that a particular system has reached artificial general intelligence, which is supposed to match human cognitive abilities in most or all tasks, because the scientific community does not have a single point of access or agreed-upon test.
The intelligent agent represents an important technological trend towards more autonomous systems, and an attempt to bridge the gap that separates current models from the desired goal, but the supposed finish line moves as models get closer to it.
The disagreement over the definition of artificial general intelligence is not just about the name, but about the capabilities that a system must possess in order to be described as general (Unsplash).
The disagreement over the definition of general artificial intelligence is not just about the name, but about the capabilities that the system should possess (Unsplash).
A definition that changes as we get closer to it
Currently, there is no universally agreed-upon scientific definition of artificial general intelligence, as definitions vary between companies and academic laboratories, and this variation directly affects how progress is assessed.
While OpenAI defines it in its charter as highly autonomous systems that outperform humans in most economically valuable tasks.
Instead of using the term artificial general intelligence, the CEO of Anthropic prefers to talk about advanced artificial intelligence, which Dario Amode describes as smarter than a Nobel laureate in most fields, and capable of performing tasks for hours, days and weeks independently.
The absence of a unified definition leads to the lack of a globally recognized test that officially announces its arrival, similar to what standardized tests do in other fields.
This absence opens the door for companies and laboratories to use different definitions and metrics when assessing how close they are to artificial general intelligence.
The Ark Prize team explains that whenever researchers create a task that is easy for humans to solve but difficult for machines, models quickly surpass it, forcing the tests to evolve again.
Google DeepMind researchers have proposed a 6-level hierarchical framework for dealing with artificial general intelligence as a step-by-step process.
Other researchers have proposed a framework that defines artificial general intelligence as a system that matches or surpasses the multiple and deep cognitive abilities of a well-educated adult.
In its current version of the framework, the GPT-4 model received a rating of 27%, while the GPT-5 model achieved a rating of 57%.
Tags:
technology
