In the early days of artificial intelligence, the field was defined by a single goal: to build a machine that could think and behave like a human. We call that AGI, or artificial general intelligence, and it’s humanity’s final tech frontier. In the 70+ years since the field’s inception, we have made tremendous progress. We can now build machines that can beat humans at specific tasks like playing chess or go, and we are starting to see machines that can learn to perform multiple tasks. But we are still far from creating another universal intelligence.
One of the challenges in the field is that the definition of AGI keeps changing as our understanding of intelligence evolves. For example, early AI researchers believed that building a machine that could beat humans at chess would require AGI. But in 1997, Deep Blue beat the world chess champion, and we now know that playing chess well does not require AGI; it is a specific task that can be solved with task-specific algorithms.
Similar to what Aubrey de Grey’s list of the 7 causes of death did for longevity studies, or what the Millennium Prize did for mathematics, maybe we can collaboratively build a program that motivates young researchers to tackle AGI one jump at a time. A clear roadmap is much more useful than the nebulous challenge of ‘solving intelligence’.
The principles I’ve chosen to create, maintain and organize the list are:
- Simple: the list should not try to be exhaustive, but aim to provide inspirational challenges.
- Task-oriented: the list should focus on specific capabilities and not research or implementation details.
- Motivational over well-defined: it’s more important that milestones are fun and motivate research, than super clearly defined.
- Increasing complexity: milestones should be listed in (estimated) ascending order of difficulty and/or task complexity.
The Roadmap, v0.1
Here goes the first version of the roadmap, divided it into thematic groups of progressive complexity. It gets kind of weird and cringe-worthy towards the end, but remember I’m trying to be as “general” as possible. Whatever we value in other humans, AGIs should be able to do too.
Please bear in mind that this is just a conversation starter. This roadmap is meant to be a living document that will be updated over time as we learn more about AGI. If you dislike a particular milestone or ordering decision, feel free to open a pull request.
Summary and Final Words
I’ve seen several of these milestones beaten, and hope to see many more in my lifetime. I really dream of being alive when we achieve AGI.
Have improvement ideas for the roadmap? Suggest a change on GitHub!
Liked what you read? Subscribe!