How many historical data do I need? If I currently have 20 projects but no historical data, then training would take a long time, wouldn't it?

No, because Can Do uses the method of imitation learning for AI resource alternatives.

Neural networks can start untrained in an approach known as 'learning by observation,' 'imitation learning,' or 'learning by demonstration,' and then be trained by observing users. This type of learning falls under the category of supervised learning or a specific form of reinforcement learning, depending on the exact implementation.

Imitation Learning

Imitation learning is a process where a model learns to perform tasks by observing human users or experts performing these tasks. The neural network then attempts to imitate the observed actions, which is often seen as a more efficient way to learn complex behaviors compared to learning from scratch through trial-and-error (as is often the case with reinforcement learning).

In Can Do, this is essentially represented by the 'Replace' and 'Discard' buttons. The user decides how good the AI's suggestion was, and thus the AI learns to make better suggestions over time.