Amazon Proposes a Home Robot that Asks You Questions When It's Confused (venturebeat.com) 30
An anonymous reader shares a report: AI models invariably encounter ambiguous situations that they struggle to respond to with instructions alone. That's problematic for autonomous agents tasked with, say, navigating an apartment, because they run the risk of becoming stuck when presented with several paths. To solve this, researchers at Amazon's Alexa AI division developed a framework that endows agents with the ability to ask for help in certain situations. Using what's called a model-confusion-based method, the agents ask questions based on their level of confusion as determined by a predefined confidence threshold, which the researchers claim boosts the agents' success by at least 15%.
"Consider the situation in which you want a robot assistant to get your wallet on the bed ... with two doors in the scene and an instruction that only tells it to walk through the doorway," wrote the team in a preprint paper describing their work. "In this situation, it is clearly difficult for the robot to know exactly through which door to enter. If, however, the robot is able to discuss the situation with the user, the situational ambiguity can be resolved." The team's framework employs two agent models: Model Confusion, which mimics human user behavior under confusion, and Action Space Augmentation, a more sophisticated algorithm that automatically learns to ask only necessary questions at the right time during navigation. Human interaction data is used to fine-tune the second model further so that it becomes familiar with the environment.
"Consider the situation in which you want a robot assistant to get your wallet on the bed ... with two doors in the scene and an instruction that only tells it to walk through the doorway," wrote the team in a preprint paper describing their work. "In this situation, it is clearly difficult for the robot to know exactly through which door to enter. If, however, the robot is able to discuss the situation with the user, the situational ambiguity can be resolved." The team's framework employs two agent models: Model Confusion, which mimics human user behavior under confusion, and Action Space Augmentation, a more sophisticated algorithm that automatically learns to ask only necessary questions at the right time during navigation. Human interaction data is used to fine-tune the second model further so that it becomes familiar with the environment.
Re: (Score:2)
Come on, it's the 21st century (Score:1)
Let's not be cavepeople here, the robot can be any gender it wishes to identify as. As long as it stays in the kitchen where it belongs.
Re: (Score:2)
I was actually thinking kids. Wives don't get confused they are always right and if you disagree you know where the couch is.
Re: (Score:2)
What Amazon needs: (Score:2)
One of the MANY articles about poor management of Amazon:
Amazon dodged workplace safety regulators for years, investigation shows [investrecords.com] (Nov. 26, 2019)
Glass Door comment. Another article. (Score:2)
"The ignorance among managers is unbelievable." [glassdoor.com] (Jan. 24, 2019)
Another article:
Suicide attempts and mental breakdowns: 911 calls from Amazon warehouses reveal that some workers are struggling [vox.com] (Mar. 11, 2019)
Quote: "Amazon has long been accused of poor working conditions."
Re: (Score:2)
If anyone was going to click that first Glass Door link, don't bother. It literally has no other information than what is quoted in the above comment.
Full text: (Score:2)
I worked at Amazon for more than 5 years
Pros
Great benefits and coworkers.
Cons
The ignorance among managers is unbelievable.
Advice to Management
Educate your managers. 90% of employees quit their management, not their jobs.
Microsoft Tried This... (Score:2)
with Windows Vista, it was a royal PITA. Constant questioning over mundane things quickly becomes tiring.
(see also: looking after young children)
Re: (Score:2)
(see also: looking after young children)
(That's ironic. There's a reason providers treat users like children today; because they've become that ignorant about these idiot-proof boxes we used to call computers.)
Re: (Score:2)
User: Cortana, what time is it?
Cortana: This program is not responding. Would you like to restart?
User: Yes.
(1 hour later)
Cortana: This program is not responding. Would you like to restart?
Re:Microsoft Tried This... (Score:5, Interesting)
They tuned it down in Windows 7 so that it only activates when a program is requesting to run with elevated privileges. That's good, the lower frequency makes it more meaningful, but most regular users still don't understand what it means because most users never understood the Windows permissions model to begin with. This is why most users still just click through it without even thinking, which makes it meaningless other than to people like us who actually do understand the permissions model.
To bring this all back around, this is distinct from this system because if you ask an assistant to do something and it asks for clarification, the function of that request is generally going to be obvious and meaningful to the user. The example used in TFS illustrates this adequately.
Re: (Score:2)
but most regular users still don't understand what it means because most users never understood the Windows permissions model to begin with.
Yup, and this was something that MS should have explained and enforced early on. But security was never really a big concern in Redmond, to put it mildly.
Most Windows users started off with an unlimited admin account that could do almost anything, and many of those things were bad, like happily allowing malware to install itself.
Confused robots (Score:2)
"Consider the situation in which you want a robot assistant to get your wallet on the bed ... with two doors in the scene and an instruction that only tells it to walk through the doorway," wrote the team in a preprint paper describing their work. "In this situation, it is clearly difficult for the robot to know exactly through which door to enter. If, however, the robot is able to discuss the situation with the user, the situational ambiguity can be resolved." The team's framework employs two agent models: Model Confusion, which mimics human user behavior under confusion, and Action Space Augmentation, a more sophisticated algorithm that automatically learns to ask only necessary questions at the right time during navigation. Human interaction data is used to fine-tune the second model further so that it becomes familiar with the environment.
If it's going to be too confusing for a robot to simply go get a wallet for me, then why bother having one? It'd be easier and faster for me to go get it myself.
Re: (Score:2)
If it's going to be too confusing for a robot to simply go get a wallet for me, then why bother having one? It'd be easier and faster for me to go get it myself.
Right, but the point is that after a couple of times of teaching it what you mean, it'll be able to go and do it on it's own from then on. Just like with a child, but with better repeatability. :)
Re: (Score:2)
Yeah, I was thinking I already have kids and they are past the stupid question stage.
Re: (Score:2)
A robot that questions you. Hmm, in line with Amazon ring, the front door bell that spies on you for the state at a profit, do they really mean the robot will ask you questions or will it interogate you. Questions, like should you be doing that, what did you mean when you complained about the state, did you know that was a forbidden word, why did you talk positively about that political party, have you bought enough today, have you voiced your support for the government today, do you need to report on famil
I don't have any of (Score:2)
Just my 2 cents
Kill all humans yes / no ? (Score:2)
Kill all humans yes / no ?
And then ... (Score:2)
User: Alexa, what's your confidence level?
Alexa: I'm not sure.
No, just NO. (Score:2)
This system was clearly not designed by anyone who has ever had children. As anyone with children knows, they ask at least one question for every command you give them; often, it's quicker just to do whatever yourself rather than explain how to do it.
Whatever physical energy might be saved by the robot will be more than made up for by the increased mental and emotional energy required to answer the seemingly random, odd, and excessive questions posed by the robot.
Re: (Score:2)
Yes! Why would I want a robot that is going to annoy me with questions. If you have kids you know.
Sounds familiar (Score:1)
Eliza: Can you elaborate on that ?
In other words... (Score:2)
The robot is now "self-aware" that it is about to make a mistake? Ok. Now just supposing here, that the robot decides that the "going to the store for you" in the first place WAS that mistake, and not the "which door it should use" to get there problem. What question will it ask then?
Could go so wrong so fast... (Score:2)
"Mary, why are you on top of Derek. Usually it's the other way around. According to information provided, it's not your birthday."
Hmmm (Score:2)
Human which door should I use? (Score:1)
Robot: Human which door should I use? ...
Me: It doesn't matter.
Robot:
Overly Simplistic (Score:2)
I really don't need a robot to go grab something in another room. I need it to do dishes, laundry, and scrub toilets. Picking up dog poop in the back yard would be a bonus too. How about that, Amazon? :)
- Necron69