🙄 OpenAI’s leaked roadmap to AGI

On July 9th, OpenAI internally presented a five-step roadmap to track its progress towards achieving Artificial General Intelligence (AGI). 

It’s a structured framework for tracking progress in AI development - and it leaked via Bloomberg.

OpenAI currently considers itself at Level 1, and approaching Level 2, their execs said in an all-hands call.

Level 1 and Level 2 represent AI systems that can solve problems at a doctorate level of education.

These are the AGI levels that OpenAI has laid out:

A closer look

Level 1 - Chatbot/Conversational AI: 

AI systems excel at engaging in natural, human-like conversations, demonstrating a foundational understanding and the ability to respond to a wide range of prompts and questions (e.g., ChatGPT, Gemini).

We have clearly reached that level already.

Level 2 - AI Reasoners: 

AI systems can solve complex problems with the proficiency of human experts. These systems are expected to perform problem-solving tasks at a person with a Ph.D.-level education without access to external resources.

What could this mean? A thought or two.

• You could have an automated research arm at hand. Pun intended. E.g.: You have a novel product idea? Let the AI build a PoC.

• You want to know if your idea resonates? Let the AI conduct 25 user researches, and put its findings together in a report.

Level 3 - AI Agents: 

Professionalized AI Agents that are capable of operating autonomously for extended periods (days, weeks, months). It can act on a user's behalf, taking on complex tasks, making decisions, and adapting to changing circumstances without constant human oversight.

A thought or two.

• At this stage, existing teams will be enhanced with full AI co-workers. AI project managers, developers, or sales reps delivering end-to-end.

• We will see whole departments with 0 humans. A team of AI Agents running it and achieving set goals. The head of department: an AI.

Level 4 - AI Innovators: 

AI systems capable of developing groundbreaking ideas and solutions across various fields.

Thoughts:

• You will have scalable access to fundamental researchers that are AI systems. Autonomously they will come up with hypothesis and proof/ disproof them.

• At Level-4 things really take off. 24/7 elementary research, among other things. Scaling AI resources up can help solving difficult problems, not all.

Level 5 - AI Organizations: 

AI Organizations function as entire entities, possessing strategic thinking, operational efficiency, and adaptability to manage complex systems in a competitive environment and achieve organizational goals. 

Thoughts:

• First we see small startups, then medium-sized businesses, then AI-conglomerates. Competing against such an AI organization is pointless.

• And that is ok. They should work for us. Their profit should be highly taxed to provide basic income for people.

• I know I’m opening a whole pandora box here, so let’s park that discussion for now. (I’m happy to engage in a conversation in the comment section of this episode.)

• This is the peak of the AGI roadmap. However, a local one, as the next stop is ASI - Artificial Superintelligence. 

The roadmap towards AGI has been murky, and OpenAI’s system could help establish more concrete benchmarks. 

While some may be disappointed at only being at 1 or 2 out of 5, exponential acceleration means we may move up the ladder faster than we can imagine.

The difficulties are between the levels of AGI are not equidistant. They get rather easier.

It was tremendously hard to get to level 1, as we are standing on the shoulders of giants.

Level 2 is literally achieved 2 years later.

From level 2, to level 3, to level 4, to level 5 - the steps will be probably easier to achieve.

If you have whole AI departments (level 4), then stitching it together to have a whole company run by AI (level 5), is in reach.

Years are ages in AI.

Want to know more?

• 🤖All things AGI: What is it and how do we get there?

• My book - Generative AI: Navigating the Course to AGI.