SoylentNews

Amazon's Alexa Needs a Body to Understand the World

posted by martyb on Thursday March 28 2019, @06:35AM   

takyon writes:

Alexa's chief scientist thinks the assistant needs a robot body to understand the world

Amazon's Rohit Prasad, head scientist and an instrumental member of the Alexa division, says the company's personal software assistant would be far smarter if it had a robot body and cameras to move around in the real world. Prasad, speaking at MIT Technology Review's EmTech Digital AI conference in San Francisco yesterday, said, "The only way to make smart assistants really smart is to give it eyes and let it explore the world."

Some Alexa-enabled smart devices already have cameras. But a robot body would be new. Prasad's comments suggest that work could be in service of one day giving Alexa a body — although he wouldn't confirm this directly. Prasad works on natural language processing and other machine learning capabilities for Alexa, so it's likely if he wanted to test these features out, he'd be one of the few Amazon employees who could easily go ahead and try it.

Someday, we can truly have sex with Alexa.

Related: Amazon Plans to Add Alexa Voice Support to Microwaves, Amplifiers, Subwoofers, and "In-Car Gadgets"

---

Original Submission

• Re:Not required (Score: 2) by takyon on Friday March 29 2019, @12:31AM

  by takyon (881) <takyonNO@SPAMsoylentnews.org> on Friday March 29 2019, @12:31AM (#821593) Journal

  If this hardware can be read out (obvious example is a neural net — training/learning is hard, but duplication is not), then only the first one needs to do that.

  I considered adding a couple of sentences addressing this in my first comment but decided to ignore it.

  We can just copy a successful "AI", assuming there isn't something crazily irreproducible happening in its hardware (like a neuromorphic architecture that undergoes tiny physical changes that are hard to measure and/or recreate). Barring that, "AI" will enjoy the benefits of mind uploading/downloading long before us meatbags can.

  First we have to get to "I", then we can talk about "strong", or not. Right now, all we have is "A", and "strong/weak AI" is just marketing nonsense designed to confuse people the same way the term "3D TV" was.

  I like to use the phrase "strong AI" as a shorthand for something like "real [signspecialist.com], non-biological intelligence, similar to or surpassing human intelligence, created by humans or another intelligence".

  Is it the best phrase ever? No. But "strong AI" invokes a certain threshold of intelligence whereas just plain "AI" as many articles discuss is simply muddled, and now a catch-all-phrase for machine learning algorithms.

  Yeah, they should definitely do this first — for all kinds of good reasons — but such a thing doesn't need to be AI. LDNLS will serve very well for this, and it handily obviates the ethics issues (both real and imaginary) of an intelligence being designed to do this kind of work, as opposed to just being capable of it.

  Yeah, rolling all the functions into one "literally intelligent" bot isn't strictly necessary. I just think that will be cheaper, more convenient, and more desirable. A robust bipedal robot could navigate existing human environments. One robot performing all of the functions gives the impression of a live-in maid/butler (with benefits). Maybe the robot will delegate vacuuming to a Roomba so you can fuck more.

  I'm giving a hard pass to ethics concerns. The ethicists are trying to squash innovation before it happens. I see it as inevitable, so I'm happy to throw the switch that may or may not doom humanity or lead to cries of bot slavery. Maybe I'll give my sex maid bot a pension and days off.

  --
  [SIG] 10/28/2017: Soylent Upgrade v14 [soylentnews.org]

  Parent

  Starting Score:	1		point
  Karma-Bonus Modifier		+1
  Total Score:		2