It is scary, because right now, most of these emotion detection algorithms are being developed in competitive closed source / black box form, for one purpose; to maximise profits. They will not care whether or not mis-diagnoses are made, if by that means, profits are maximised.

The ambition to make money via profit for these developments goes all the way back through all of the Universities, who run multiple projects researching similar subjects in closed silos, isolated from one another, operating in competition with one another, sponsored by different competing profit seeking companies. Ironically, the net output of such developments could be so much better and more powerful towards good ends if they collaborated with one another, but sadly, they never will, as long as profit is the driver.