RE: AI: Job Replacement Is Coming

The next big improvements in chat AI might come from their introduction into RPGs. On top of regular chat bot you need:

• knowledge filters - if you start talking about quantum physics to a bartender in Skyrim, he should treat you like you were a madman
• personality traits - each NPC should be different
• emotions

Text-to-speech is already top notch. The accent, the tempo, pronunciation, general flow of speech is pretty much natural. While for some people seeing, let's say, Marlon Brando suddenly speaking Polish might trigger uncanny valley effect, it might happen mostly for older generation (and no amount of software improvement can fix a problem that lies in the listener's head). Voices of less known actors won't be a problem and for games and animations AI speech can already replace humans.

Finally images/videos. In my opinion current technology is in a bit of a blind alley. The AI "doesn't know" what's on the images it creates, it just matches text to images it have seen during training (with a healthy dose of mixing). Everything would instantly change if generation of image started from actual objects - a 3D model (with physics model for video) - and continued with adding specific filters on top of that. We would no longer see any of current blatant giveaways like people with seven fingers (we could still specifically request that, but in such case first step would add extra fingers to the 3D model first and also by default in place where you could expect them, unless explicitly told otherwise, and then continued from there). The problem with such approach is that there is limited amount of 3D and 4D data for training, especially compared to abundance of flat images and videos. Maybe full self driving will be a trigger in right direction as it requires AI to recognize and track objects in 4D.