Does anybody else find "Sophia" concerning?

[rhyolight]

I found some interesting details about Sophia recently posted on the opencog mailing list:

The robot has a super-crude vision system. It’s nearly blind. It can see faces, locate them in 3D space. It has an Intel realsense thingy that returns human body and hand position data, but its … crude and inadequate. The realsense is not hooked up to the atomspace at the moment. Perhaps we could steal some code or ideas from CogSketch or QSRlib. (google those)

Sound: we have speech-to-text, via google. It sort-of-ish works sort-of if you have a good microphone, and no background noise. Otherwise, its not so good. It cannot tell if you are whispering, shouting, angry, serene. It cannot tell if there is an audience that is clapping, booing. It cannot tell if its in a crowded room or an empty room.

Turns out sensory input is a really hard problem.

And from Ben Goetzel:

Over the next few months here in HK we are making a renewed push at
getting OpenCog to usefully control the Sophia robot (we did this sort
of work a year ago and then paused it to focus on various internal
improvements)

And my favorite:

The robot is as dumb as a rock. Its not a whole lot more than a glorified chatbot, at this time. (I actually counted – the robot has somewhere between 5 and 8 bits of knowledge about the external world. That’s “bits”, not bytes.)

So I don’t think this thing is anywhere near a functioning intelligence at the moment.

---

On another note, I just spent some time reading through their mailing lists… and this place is a lot nicer