Great idea! Wasting spammers time is one of the very few ways of making them effectively lose money.
Still the project needs some pseudo randomization in order to fool basic research. Some of the messages I read contained the sentence "I am a bit busy now, but I am definitely interested. When can we talk?". If one searches for that sentence in quotes Google returns four pages of references to Spamnesty. A slightly more clever spammer would sense the trap in minutes.
I'm not aware of any program capable of add some entropy to a sentence maintaining its semantics and readability (save for insertion of random errors), but one could use a translator in different languages. Say English->German->Russian->Spanish->Polish->Danish->English, then send the resulting string as a reply. The above string for example becomes: "Now I'm a bit busy, but I'm definitely interested. When can we speak?" which (almost) fools the search. Almost because the 1st page still contains a couple references to Spamnesty due to the above sentences strong similarity, but they're well buried together with other completely unrelated stuff.
BTW, pairing this with a personal assistant stripped of any access to personal information/property/devices (such as an hypothetical open source cloudless one) and instructed to ask for details on every possible part of the offer, one could make the perfect weapon against phone telemarketers as well.
I guess spammers make most money from the lowest IQ percentiles on the internet. Therefore, we only have to develop AI that is as smart (stupid) as those people, and spammers won't be able to tell the difference.
there is no end to that argument, if a spammer is good enough to be able to foresee anti-spam software then they are likely beyond the target audience. BUT the average and by large general spam crew are not technical experts, most of them have bought some software to advertise to a large audience by mass email / crawl / filter, and are just looking at monetizing some product (or get your infos / CCs).
By wasting time for the 95% of the spammers (ironically with techniques that are very similar to them) you can still weed out a big chunk of that sector and have a big impact.
BTW, pairing this with a personal assistant stripped of any access to personal information/property/devices (such as an hypothetical open source cloudless one) and instructed to ask for details on every possible part of the offer, one could make the perfect weapon against phone telemarketers as well.