What is the t-AGI framework?

1 sentence answer

A system receives the designation of "t-AGI" if it can surpass a human expert in most cognitive tasks within the timespan 't'. (Where “expert” means anyone who’s capable of doing said tasks.) A system gets identified as (t,n)-AGI if, for most cognitive tasks, it can outdo a group of 'n' human experts working collectively on for the duration 't'.

The types of tasks that are expected to be solved include things such as the set of tasks within the Abstraction and Reasoning Corpus (ARC benchmarks). However, there is no consensus as of yet on the specific set of cognitive tasks that would need to be performed.

example

As an example, if both the human expert and the AI are given 1 second to perform some tasks, then the system is 1-second AGI if it achieves those cognitive tasks better than the expert. Similarly we can have 1 minute, 1 month, etc… AGIs, if their outputs outperform what human experts could achieve within a minute, month, etc…

Following are some other predictions by Richard Ngo on what types of capabilities we can expect an AI to be better than humans at, at different ‘t’ thresholds.

  • 1-second AGI: recognizing objects in images, recognizing whether sentences are grammatical, trivia answers
  • 1-minute AGI: answering questions about short text passages or videos, common-sense reasoning (e.g. Yann LeCun's gears problems), simple computer tasks (e.g. use photoshop to blur an image), looking up facts, etc.
  • 1-hour AGI : doing problem sets/exams, writing short articles or blog posts, most tasks in white-collar jobs (e.g. diagnosing patients, giving legal opinions), therapy
  • 1-day AGI: writing insightful essays, negotiating business deals, developing new apps, running scientific experiments, reviewing scientific papers, summarizing books, etc.
  • 1-month AGI: coherently carrying out medium-term plans (e.g. founding a startup), supervising large projects, becoming proficient in new fields, writing large software applications (e.g. a new OS), making novel scientific discoveries, etc.
  • 1-year AGI: Would beat humans at basically everything, because most projects can be broken up into sub-tasks that can be achieved in shorter time frames.

more detail

The t-AGI framework can be extended into the (t,n)-AGI framework, where n is the number of humans. So a single (t,n)-AGI would beat a group of n human experts working together on some set of cognitive tasks for time t.

conclusion

Overall existing systems as of Q2 2025 are believed to be 1-minute AGIs, and to be close to 1-hour AGIs. According to a report by METR, the time required to move from a t-AGI to a 2t-AGI is 7 months, which implies we are a few years off from having 1-hour AGIs.

Within this framework, Richard Ngo expects superintelligence (ASI) to be something like a (1 year, 8 billion)-AGI, i.e. ASI is an AGI that takes 1 year to outperform all 8 billion humans coordinating on some task.



AISafety.info

We’re a global team of specialists and volunteers from various backgrounds who want to ensure that the effects of future AI are beneficial rather than catastrophic.

© AISafety.info, 2022—1970

Aisafety.info is an Ashgro Inc Project. Ashgro Inc (EIN: 88-4232889) is a 501(c)(3) Public Charity incorporated in Delaware.