Hacker News Reader

Comment by Taikonerd | Hacker News Reader

Voice interfaces. We have Siri and Alexa, yes, but they're very limited -- they just have a list of pre-programmed tasks they know how to do.

I realize that a fully-general voice interface is an "AI-hard" problem, but surely we can improve over what's on the market now?

View on HN

I think ironically your criticism of voice interfaces actually betrays perhaps their biggest weakness over visual interfaces - it's not immediately obvious what their capabilities are, what they will respond to or understand, or what your options in a situation are.

Super users and developers of these systems are likely able to harness them for much more powerful usecases than the average person, but how do you make those options and usecases visible to the end user?

It would seem like the answer is either 1) a robocall / callcenter bot style solution (press 1 to ____, press 2 to ____, press 3 to ____) where you must listen to lists of things but can more expressively verbally respond rather than press 1, or as you said solve the Hard AI problem of the bots contextually understanding your current situation and suggesting options based on it.

I think you're underestimating what Google and Amazon are doing behind the scenes (there are a lot of people working on this currently).

G&A are going to be so concerned with "loosing out on data" that to their detriment all of their voice-driven interfaces will be cloud dependent. Hopefully, someone with enough wits will dare to lose out on that data and instead create devices with local voice models that can still learn and become more useful as you use them. I want this model to become embedded into my operating system.

I want an OS that gets smarter by the day but without phoning home. I want my OS to become better than yours, simply because I trained it better that you did yours. I want this OS to be _my OS_. Not your's. Not Apple's. Not Microsoft's.

Did you see the movie "AI" (written by Kubric, directed by Spielberg)? In the movie there's an AI driven teddy bear that is awfully close to being an AGI. I want my OS to be like that.

One of the challenges with voice assistants is that voice is a very low bandwidth medium for communicating information. If I ask Siri to tell be about Bill Gates, it would take her a minute to read to me what I could skim in seconds from a Google Search result.

There's nothing inherent to voice assistants that requires that they speak back to you. You can ask them using voice, and get the results on a screen.

Same with phone messaging (like whatsapp etc) where it's much faster to receive a text message vs the same message in spoken audio.

Voice recognition general is half baked. Voice recognition has been pushed on consumers for 10 years. Siri barely knows what I am saying. The voice recognition on customer service is always broken.

The technology is in its infancy. Yet, being built into more and more devices and services daily.

Not to mention, I don’t want to talk to my television remote. I want to use muscle memory to press a button.

Voice recognition is a mess.

Reply on news.ycombinator.com