I am concerned by the fact that some major companies are proposing that in on-line meetings we should all appear to each other as computer graphic avatars. It’s not just a philosophical issue — it’s also about the devil in the details.
One of the key elements of such an initiative is solving the latency problem. If I am going to appear to you as a synthetic avatar, then your computer needs to know what to do if there is any delay in network transmission.
For video-based interactions (like what you see over Zoom), this is not such a big issue. We understand that networks can be unreliable and that sometimes a video feed can stutter.
But you can’t quite do that with synthetic avatars. Instead you need to fill in the gaps with machine learning. You need to fake it.
In particular, you need to synthesize body movements and facial micro-expressions to fill in the gaps. And those movements and expressions don’t necessarily fully correspond to reality.
So that direction sends us down a slippery slope to a make believe reality in which the subtleties of people’s emotions are represented not faithfully, but rather as third party constructs.
That might not be such a good idea.
I may have mentioned this before, but after Avatar came out, there were behind-the-scenes clips on YouTube showing the actor’s performance side-by-side with the CGI rendering.
Watching the movie, the CGI characters looked pretty good; it was spare-no-expense movie tech a decade ago.
But when you watched the CGI next to the actual actors you saw how much of the expression and emotion was drained in the reproduction. Human brains are highly tuned to read human faces, and even very subtle details matter.