This makes no sense especially given that even a human observer could not tell instantly just by looking at a piece of content whether the content is properly licensed. As others have said, there no reasonable way to 100% correctly detect and filter restricted or copyrighted content based purely on the content itself.

Just look at the mess that is the patent system... even "experts" in a field or art are making huge mistakes when it comes to judging obviousness and originality. Why do people think a machine will do as well, let alone better when trying to judge legality of a piece of content?

I think what it comes down to is the content industries wanting to force technology companies to fight piracy for them. Technology companies know that trying to fight something like this (past a certain point) just leads to an arms race with huge amounts of collateral damage on either side. Media companies are throwing in the towel, they just want companies like Google to fight for them, and in doing so bear all of the responsibility and cost (while gaining zero benefit).