1. At it's core, it's a Proof-of-Work captcha, so compute load depends on the usage pattern. AI (LLMs) does get better at image/audio recognition, but compute task is the same anywhere, does not matter who "clicks" the checkbox.
2. Bots usually have such access patterns that cause compute task difficulty to become prohibitively high, requiring exponentially more CPU resources. This renders bots economically nonviable for submitting forms or scraping website.
