[mattlondon]

The trouble is, most ad serving systems already work out the context/theme of pages and target ads appropriately but even then its not perfect.

E.g. you are an airline and you want to advertise your flights. You might want to have your ad appear on pages that have contexts of "travel, flying, vacations" etc since you believe that those pages are likely good pages to reach people interested in your services. Sounds reasonable. An advertiser would obviously NOT want to appear advertising their cheap flights to Tenerife next to this sort of terrible, tragic news.

As others have mentioned its perceived as bad for the brand be be seen next to bad news, but more importantly (and more likely IMHO) people are human and understand the gravity of the situation so apply common sense and tact.

So it gets into a silly situation where you have advertisers targeting contexts "travel, flying, vacations AND NOT disaster, crash" - whilst you are there you may as well start excluding some other negative things too, like "thailand, children, ladyboy" etc etc until you have these huge lists of tens of thousands of exclusions getting into the ridiculous depths of human depravity, whilst you just want to show on "good" pages about just 3 simple contexts: travel, flights or vacations.

And even then ads sometimes still slip through and appear on "bad" pages because someone forgot to put "erotic trepanation" or whatever in the exclusion list.

[smoe]

I agree and think it's impossible to foresee every possible "bad" correlation. Furthermore the problem is not limited to advertisement but every type of automated content. E.G. a widget with a generated list of "related stories" or an autocomplete search box.

With more and more of this style of content being used, it is in my opinion important to better inform the not so tech savvy public, that computer generated relations can be really useful but will yield some results that are wrong or might seem distasteful for the human mind.