In the quest to make robots more human-like, the holy grail has always been a humanoid machine that is as indistinguishable from a real person as possible. It’s a high bar to clear, and we’re obviously not there yet, but a Samsung-backed startup called Neon is aiming to impress with a cast of unreal virtual humans that it brought along to CES 2020.
In still images, the faux humans look reasonably legit, but as more and more people get a chance to interact with the two-dimensional digital guys and gals, it’s becoming increasingly clear that these are not the digital humanoids anyone is looking for.
Eyes-on demos from the BBC, the Verge, and several others reveal the virtual humans to be, well, pretty lame.
The promise here is that the virtual people can interact with a person in the same way that a real person would. That means responding to various inputs, including speech and motion. The reality, at least at this stage of Neon’s development, is that the life-sized avatars act more like collections of short video clips that are triggered based on whatever algorithm is running the show.
So, for example, asking the virtual person to pose for a selfie might get the 2D avatar to strike a pose, but that’s where that particular interaction ends. The avatar returns to its default state and waits for another input it recognizes so that it can carry out whatever action it thinks is appropriate.
This clip from CNET at CES is a really great example of how far the tech falls short of its ultimate promise:
This is a significant letdown, particularly when Neon has positioned these avatars as artificial humans, when in reality they’re little more than oversized video clips stitched together. Could we someday get to a point where a 2D human on a display could act and react in a way that a real person might? Sure, it’s feasible, but this most certainly isn’t it.