Creativity Gain

Creativity gain refers to the benefit that arises when AI produces ideas that are stranger and more unexpected than what humans typically come up with. Humans tend to think within familiar frameworks, but AI's strength lies in its willingness to boldly suggest statistically rare combinations or paths no one has tried before.

Think of it like casting a wide net to catch fish. The wider you cast the net, the higher the chance of catching something rare and unusual. The problem is that a lot of what ends up in the net might be inedible junk fish or common species everyone already knows about. A large creativity gain just means the net is wide—it doesn't mean everything caught is useful.

That's why creativity gain always comes with a verification cost attached. The stranger and more surprising a result is, the more it needs to be filtered by human judgment to determine if it's actually meaningful. This is why you can't judge AI-generated ideas based on quantity and novelty alone.

The article introduces AutoDiscovery, a scientific discovery tool from the Allen Institute for AI, which is good at generating statistically surprising hypotheses—but many of them turn out to be biologically nonsensical. This is a case that shows a large creativity gain doesn't itself guarantee reliability. People often mistakenly assume 'AI generating lots of new ideas = AI doing well,' but what the article emphasizes is that without expert judgment (scientific intuition) to filter that novelty, creativity gain can end up as mere noise.

Give a chatbot a familiar problem and ask it to 'suggest 5 solutions that differ from conventional approaches, ones probably no one has tried before.' Then go through each answer one by one and judge for yourself whether it 'actually makes sense, or is just strange.' This will let you feel firsthand how creativity gain and verification cost play out against each other.