The theory of the Web of Trust is that participants act as local certification authorities, verifying the correspondence between public keys and identities. Guarantees can be obtained from chains of signatures on keys based on the following heuristic: presumably, an attacker could corrupt or deceive some users, but not all. So if you can build many chains of signatures, from a key you trust a priori (your own key) down to the key you want to validate, and all these chains go through distinct intermediate keys, then all the involved presumed manual verification somehow add up: if you have 12 such chains, then, the attacker (who tries to impersonate a user) would have had to bribe or fool at least 12 distinct people to achieve such a picture.
Whether this heuristic validation is sufficient is open to debate (personally I don't find it anywhere near decent, as far as security is concerned). However, the core principle is that any act of key signature contributes to the overall security only insofar as the corresponding identity verification is robust against corruption.
In the case of your robot, the robot would merely verify control of an email address. Unfortunately, emails are not a secure system; in fact, PGP has been invented precisely because emails are not secure. When you use PGP, you assume that possible attackers can read and fake emails at will; that's the very reason why you use PGP. In that context, basing the safety of the key-to-id signatures on emails seems unwise. Or, said otherwise: the contribution of the signatures produced by your robot to the overall "validation guarantee" ought to be considered to be very low. It does not hurt per se, but it won't help a lot.
(Or, rather, it may hurt if people begin to trust such signatures for more than what the actually provide.)