For any task in ARC, it's possible for a human to write a reasonably short computer program that can handle new pairs. Everything is straightforwardly computable -- unlike, say, classifying MNIST digits. There's no human advantage in any single task.
Yet, there's currently no program that can solve a *new* ARC task. Such a program would be an incredible breakthrough in artificial intelligence.
The ability to look at a new ARC task and come up with a solution program is, as of yet, uniquely human.
A task-specific solution program has no intelligence of its own. It's merely the output artifact of the process of intelligence.
"Intelligence" is what happens in the mind of the 5-year-old kid who, looking at a *new* task, understands it and develops a solution protocol.
The single biggest misconception in AI is the confusion b/t unintelligent output artifacts (skill programs) & the process that generates them. "Artificial intelligence" has been focusing purely on task-specific automation, entirely disregarding intelligence.
Time to change that.
You can follow @fchollet.
Tip: mention @threader_app on a Twitter thread with the keyword “compile” to get a link to it.
Enjoy Threader? Sign up.
Threader is an independent project created by only two developers. The site gets 500,000+ visits a month and our iOS Twitter client was featured as an App of the Day by Apple. Running this space is expensive and time consuming. If you find Threader useful, please consider supporting us to make it a sustainable project.