But once OCR is improved to the point where it can read poorer scans, won't spammers be able to use that new technology to eventually defeat CAPTCHA?
If you look at CAPTCHAs as a method for improving artificial intelligence, then the issue you're raising gets turned into a benefit of ever-improving Turing tests.
- Find a problem that AI cannot readily solve, yet humans can (obfuscated word recognition).
- Develop a CAPTCHA out of the problem (reCAPTCHA).
- Either the problem remains unsolved and the CAPTCHA must be solved by humans...
- ... or we develop stronger AI capable of solving the original problem.
In either outcome, we get something useful, either better Turing tests or more robust AI. The hardest part of this loop is developing novel CAPTCHAs, but at least the cycle results in useful outcomes every time.