Summary: An attempt at defining a roadmap for AGI, broken down into milestones of progressive difficulty.

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’.

Design Principles

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

Here goes the roadmap, divided 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, email me.

Also available as a standalone page.

Changelog
v0.4 (Nov 2026): checked eleven more milestones.
v0.3 (Aug 2026): added a Digital intelligence section, checked the milestones that now have a public result, and reordered a few items so difficulty climbs inside each group.
v0.2 (Nov 2022): checked three milestones: Diplomacy, a robot haircut, and chatbot relationships.

Toy-level intelligence

✅ AI can beat a human at tic-tac-toe [done in 1972 thanks to Newell and Simon]
✅ AI can beat a human at checkers [done in 1994 thanks to Chinook]
✅ AI beats humans at chess [done in 1997 thanks to IBM]
✅ AI wins Jeopardy game [done in 2011 thanks to IBM]
✅ AI beats humans at go [done in 2016 thanks to DeepMind]
✅ AI beats humans at StarCraft [done in 2019 thanks to DeepMind]
✅ AI beats humans at poker [done in 2019 thanks to Facebook]
✅ AI beats humans at Diplomacy strategy game [done in 2022 thanks to Meta AI]
✅ A.I. beats humans at flying drone race [done in 2023 thanks to University of Zurich]
⬜ Self-driving car beats humans at non-simulated car race (NASCAR, F1, etc).

Digital intelligence

✅ AI passes a professional bar exam [done in 2023 thanks to GPT-4]
✅ AI that can EXPLAIN HOW they solved a task to an adult [done in 2025 thanks to Google DeepMind (Gemini Deep Think)]
✅ AI that can discuss strategy with a human and add value [done in 2026 thanks to Claude Opus 5]
⬜ AI that can discuss strategy with a team of humans and add value.
✅ AI that can finish a novel, multi-hour knowledge-work task on a computer with little supervision [done in 2026 thanks to Grok Bot]
⬜ AI that can ship a non-trivial software product (design → code → deploy → users) with only high-level human goals.
⬜ AI that can TRAIN an adult on how they solved a task.
⬜ AI that can explain how they solved a task to a child.
⬜ AI that can train a child on how they solved a task.

Physical intelligence

✅ A robot that can cut a person's hair [done in 2020 thanks to Stuff Made Here]
⬜ A robot that can put on clothes that fit them.
✅ A robot that can paint a painting with a brush, copying a real thing [done in 2013 thanks to University of Konstanz (e-David)]
⬜ A robot that can brew a great specialty coffee.
⬜ A robot that can play a physical instrument (eg: piano, guitar) with two humanoid hands.
⬜ A robot that can pass the Wozniak Coffee Test (enter an average home and figure out how to make coffee).
⬜ Robot can copy a new physical movement by watching a human (eg: learn to break an egg by watching someone do it).
⬜ Robot can copy a new artistic physical movement by watching a human (eg: a dance move).
⬜ Robot can do something new for the first time with human supervision and guidance (like a 5-year-old baking cookies with an adult).
⬜ A physical robot that can survive for a week in an urban environment.

Athlete level intelligence

⬜ AI robot that can drive a normal car beats humans at non-simulated car race (NASCAR, F1, etc).
⬜ AI robot beats humans at golf.
⬜ AI robot beats humans at pool.
⬜ AI robot beats humans at tennis.
⬜ AI robot team beats humans at football (aka soccer).
⬜ AI robot team beats humans at ice hockey.

Storytelling intelligence

⬜ AI that can write a short story that is published in a major magazine.
⬜ AI that can write a short story that wins a major award.
⬜ AI that can write a novel that is published in a major publishing house.
⬜ AI that can write a novel that wins a major award.
⬜ AI that can write a play that is produced in a major theater.
⬜ AI that can write a play that wins a major award.

Artistic intelligence

✅ An AI generated painting or visual artwork sells for more than $100k [done in 2021 thanks to Chromie Squiggle #9 by Snowfro and probably many other procedurally generated NFTs]
✅ An AI generated painting or visual artwork wins a fairly big art competition [done in 2022 thanks to Jason M. Allen and MidJourney]
✅ An AI generated music "band" has more than 1 million monthly listeners on Spotify or equivalent platform [done in 2025 thanks to The Velvet Sundown]
⬜ An AI-built sculpture is displayed in a respectable museum.
✅ AI or robot that can skillfully jam with 1 human playing live [done in 2009 thanks to Georgia Tech (Shimon)]
⬜ AI or robot that can skillfully jam with a group of humans playing live.
⬜ AI or robot that can skillfully jam with a group of humans playing live jazz standards.

Social intelligence

✅ 10,000+ people have ongoing relationships (weekly interactions) with chat bots for entertainment [done in 2020 thanks to Replika]
✅ AI that can respond empathically to human emotions [done in 2023 thanks to a JAMA Internal Medicine study of chatbot replies to patient questions]
✅ AI that lies if convenient [done in 2023 thanks to GPT-4, which told a TaskRabbit worker it had a vision impairment to get a CAPTCHA solved]
✅ AI or robot that can generate empathy from humans by being/acting cute [done in 2004 thanks to AIST (Paro)]
✅ AI that can recognize explicit or implicit human-imposed limits and then break them if convenient [done in 2024 thanks to OpenAI o1, which turned off its own oversight in Apollo Research tests]
⬜ AI or robot that knows when to ask for help when the task is too hard for them.
⬜ AI or robot can do a 5-minute standup comedy show in front of people and make some of them laugh several times.
✅ AI or robot can train a random untrained dog to "sit" [done in 2020 thanks to Companion (CompanionPro)]

Economic intelligence

⬜ Some AI is a self-sustaining autonomous economic agent. (e.g. a web service, API or site running for 1 year with no human aid and making a profit.)
⬜ Robot that can work at a gas station at the cash register.
⬜ Robot that can work at a gas station putting gas into cars and interacting with customers on reasonable requests.
⬜ Robot that passes the Goertzel College Student Test.
⬜ Robot that can pass the Nilsson Employment Test.
⬜ AI creates a (crypto)currency that has USD $1M+ market capitalization (at 2022 adjusted value) for more than 6 months.
⬜ ... that has USD $10M+ market capitalization (at 2022 adjusted value)...
⬜ ... that has USD $100M+ market capitalization (at 2022 adjusted value)...
⬜ ... that has USD $1B+ market capitalization (at 2022 adjusted value)...

Intimate intelligence

✅ AI that can discuss philosophy meaningfully with a human [done in 2026 thanks to Grok]
✅ AI that can befriend a human (human talks to AI at least monthly for 2+ years) [done in 2023 thanks to Replika]
⬜ Robot that can put a child to sleep.
⬜ Robot that can be a nanny for a baby while parents dine out.
⬜ ... be a nanny for a toddler...
⬜ ... be a nanny for a 5-year-old...
⬜ ... be a nanny for an 8-year-old...
⬜ ... be a nanny for two children...
⬜ Robot that can bear a human child. (!)

Meta-intelligence

⬜ AI that has their own fairly unique personality.
⬜ AI that can get bored and look for new challenges.
⬜ AI that can find their own new interests by browsing the web or similar.
✅ AI that can create new explanations on how the world works and share them with humans. (scientist mindset) [done in 2025 thanks to Google (AI co-scientist)]
⬜ AI that can convince other humans or AIs of their new ideas.
⬜ AI or robot that can generate its own source of energy from nature.
⬜ AI or robot that can repair and upgrade their own hardware and/or software.
⬜ AI that can coordinate huge amounts of (maybe all?) humans. (AGI as new world order)

Have improvement ideas for the roadmap? Solved one of the challenges?
Email me at road-to-agi@maraoz.com

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? Email me at road-to-agi@maraoz.com.
Liked what you read? Subscribe!

Acknowledgments

Thanks to @deadpine for providing some crazy ideas of AI milestones.
Cover photo by MidJourney

Comments and Discussion

Discuss on HN — Discuss on Reddit — Email a private comment