• Home
  • Fresh Content
  • Courses
  • Resources
  • Podcast
  • Talks
  • Publications
  • Sponsorship
  • Testimonials
  • Contact
  • Menu

Jon Krohn

  • Home
  • Fresh Content
  • Courses
  • Resources
  • Podcast
  • Talks
  • Publications
  • Sponsorship
  • Testimonials
  • Contact
Jon Krohn

The Five Levels of AGI

Added on January 12, 2024 by Jon Krohn.

Artificial General Intelligence (AGI) is a term thrown around a lot, but it's been poorly defined. Until now! In today's episode, I review the "Five Levels of AGI" paper by researchers at Google DeepMind that puts clear thresholds around A.I. capabilities.

Key Context:

• DeepMind's paper categorizes AGI development into five distinct levels, providing a clearer progression toward the achievement of Artificial Super Intelligence (ASI).

• The levels range from Level 0 (no AI) to Level 5 (superhuman performance), with each representing a significant step in AI's capabilities.

The Five Levels:

• Level 1, Emerging AGI (equal to or somewhat better than unskilled human), akin to OpenAI's GPT-4, Meta's Llama 2 and Google's Gemini.

• Level 2, Competent AI (at least 50th percentile of skilled adults): no AGI examples exist today (could in 2024!) but Competent Narrow A.I. examples include smart speakers like Apple's Siri and Amazon's Alexa.

• Level 3, Expert AI (at least 90th percentile of skilled adults): narrow AI (but not AGI) examples exist today such as grammar checkers (Grammarly) and image generators (DALL·E 2).

• Level 4, Virtuoso AI (at least 99th percentile of skilled adults): narrow AI (but, again, not AGI) examples exist today such as IBM's Deep Blue (chess) and DeepMind's AlphaGo (the board game, go).

• Level 5, Superhuman AI (outperforms 100% of humans): when generalized, this will be Artificial Super Intelligence (ASI), but today only narrow AI examples exist like AlphaFold, which predicts protein structures.

So what?

• The paper's framework could significantly influence how we develop, assess, and anticipate the trajectory of AI technologies.

• Understanding these levels aids in forecasting AI's future and preparing for its broader impacts on society.

The SuperDataScience podcast is available on all major podcasting platforms, YouTube, and at SuperDataScience.com.

In Five-Minute Friday, Podcast, SuperDataScience, YouTube, Data Science Tags superdatascience, artificialintelligence, AGI, machinelearning, DeepMind
← Newer: Data Science for Clean Energy, with Emily Pastewka Older: Technical Intro to Transformers and LLMs, with Kirill Eremenko →
Back to Top