Let's say a company is trying to convince buyers of media localization services that machines can understand films, plots, human emotions and relationships. And this is the solution for translating content at scale, because it allows them to cut costs to 50%, obviously by replacing the expensive human who used to do the understanding part.

It's hard to believe that buyers will actually believe machines understand. But they might be ready to go along, because who doesn't like half-price?

Those of us who know, because we're seeing it in our everyday work, that the simulation of understanding is not understanding; and that machines can't be held accountable for a drama turning to comedy or the other way round - we need to be louder about this. We need to find ways to show that giving the machine more context or letting it extract metadata from the content does not mean understanding.