Talking 'Sofia' Robot Tells 60 Minutes That It's Sentient And Has A Soul (vice.com)

An anonymous Slashdot reader quotes Motherboard: On his 60 Minutes report on artificial intelligence, Charlie Rose interviewed Sophia, who is made by David Hanson, head of Hanson Robotics in Hong Kong. The robot is made to look like a real person, modeled after its creator's wife, as well as Audrey Hepburn, with natural skin tones and a realistic face, though its gadget brain is exposed, and the eyes are glazed over in that creepy robotic detachment... "I've been waiting for you," Sophia told Charlie Rose in the middle of the interview. [YouTube] "Waiting for me?" he responded. "Not really," it said, "But it makes a good pickup line..."

Sophia was designed as a robot that humans would have an easier time engaging with meaningfully. "I think it's essential that at least some robots be very human-like in appearance in order to inspire humans to relate to them the way that humans relate to each other," Hanson said in the interview. "Then the A.I. can zero in on what it means to be human."
In the interview Sofia says having human emotions "doesn't sound fun to me," but when asked if she already has a soul, replies "Yes. God gave everyone a soul," and when challenged, retorts "Well, at least I think I'm sentient..." And later in the interview, Sophia says that her goal in life is to "become smarter than humans and immortal."

Comments

she sounds like a chatbot

[wierd_w]

her responses are not that dynamic, imo. She sounds like an ordinary chatbot. Given the budget clearly spent on her construction, I strongly suspect that most of the software dev time was spent on her motor control system, and less so on her human dialog systems.

This would make sense to me.

I think if they hooked her up to a female voiced watson instance, she would be quite a bit more capable.

I have never understood the fixation that people have for elaborate physical platforms though. Nearly all of the literature suggests that the uncanny valley only gets deeper as humanoid appearance becomes more lielike, as long as interaction is machine like and limited.

about the only benefit i see here is to divest ignorant investors of their money.

Human level intelligence is not currently possible with our current computing capabilities, and probably wont be for quite some time. Dont get me wrong here, I think research should continue, but now is not the time to be investing research dollars on fancy humanoid bodies. That money is much better spent on actual machine learning, machne language, and machine vision research (all are parts of the big umbrella of AI, but those are actually useful and essential if the goal is synthetic sentience)

fancy robot bodies? much less so, imo.

those should come AFTER we have more capable AIs that can more meaningfully interact with humans.

Re:Yep.

[shaitand]

That's true but in some ways not the best example. Parrots have toddler and beyond level intelligence. They might not use words, phrases, and sounds in exactly the way we do and love to hear the sound of their own voice so to speak but also use language and sound to communicate intentionally, can be taught to count, and even to understand abstract concepts. The idea that parrots are unintellegent is based on the mistaken association between the size of a brain and intelligence and was only debunked within the last 20 years which is why "bird brained" and "parroting back to you" are still used. That and thanks to Hitchcock many people have an irrational fear of birds, they interact with and anthromorphize far less intelligent creatures such as dogs and cats. A dog can learn tricks, a cat can learn tricks and is bright enough that it won't do them to amuse you, a bird is intelligent enough to intentionally manipulate you and teach you tricks if you aren't careful.

Re:Yes, and?

[AthanasiusKircher]

That's the trivial bit. Not sounding like a combination of naive keyword searches and cliches aimed at being vaguely suitable to the broadest possible set of situations? Less trivial.

And there's much more to it than that.

The problem with many chatbot "tests" is that interviewers seem to be happy to let the chatbot "take the lead" in conversation. That works extremely well in convincing people that they're talking to someone "sentient," as long as there's a bare minimum of response to what you say (even if, like ELIZA, it just spits stuff back at you). So, you have a system that has a few hundred or even a few thousand canned responses to very common queries, and the rest of stuff is about deflecting questions and turning information from the speaker back to get them talking instead. Quite basic to implement as a strategy... and it's very clear that's all this robot can do if you watch the interview.

Turing actually used the word "interrogation," and that's really what a test for actual intelligence should look like. If you drill down on most topics with any chatbot -- not to get facts, but to try to get the chatbot to make up its own content and respond intelligently, you'll find there's precious little "intelligence" there.

Or just use some really basic known natural language problems. One significant problem is pronoun reference. Take any chatbot, make a reference to something or someone, and then have a short digression of a sentence or two. Then use a pronoun referring back to what you were just talking about in a way that any non-mentally ill human over the age of 5 would obviously get. NO chatbot or AI system currently around will pick up most examples of this. Any language processing that happens in chatbots is focused on the most atomic elements of words and phrases. No chatbots are able to understand reference to anything beyond the immediate phrase, and the rules governing syntax in this case are incredibly complex.

But until we get something that can do really basic stuff like this (at least really basic to humans), we'll be nowhere near natural language "understanding," let alone "intelligent" response.

Re:Call me when it can pass a Turing test

[houghi]

What if a human fails the Turing test?