Levels of AGI & Autonomy

Deepmind published a paper discussing levels of AGI and autonomy.

Levels of AGI: Operationalizing Progress on the Path to AGI

The authors discuss capability levels of AGI w.r.t. performance (depth) and generality (width, i.e., narrow or general).

During the era of LLMs, it is interesting to note the following:

ChatGPT is at Emerging AGI Level 1 and AlphaZero is at Superhuman Narrow AI Level 5.

The current LLMs may support Autonomy Level 2: AI as a Consultant or  Autonomy Level 3: AI as a Collaborator.

Autonomy Level 5: AI as an Agent requires Virtuoso AGI Level 4 or Superhuman AGI Level 5. That is, the current LLMs are far from being competent enough to support AI as an Agent.

The following implications are straightforward:

Current LLMs are far from being Superhuman AGI.

AI as an Agent are far from being ready.

See the tables below or the paper for more details.