Microsoft Proposes AI That Improves When You Smile (venturebeat.com) 39
Positive affectivity, or the characteristic that describes how people experience affects (e.g., sensations, emotions, and sentiments) and interact with others as a result, has been linked to increased interest and curiosity as well as satisfaction in learning. Inspired by this, a team of Microsoft researchers propose imbuing reinforcement learning, an AI training technique that employs rewards to spur systems toward goals, with positive affect, which they assert might drive exploration useful in gathering experiences critical to learning. From a report: As the researchers explain, reinforcement learning is commonly implemented via policy-specific rewards designed for a predefined goal. Problematically, these extrinsic rewards are narrow in scope and can be difficult to define, as opposed to intrinsic rewards that are task-independent and quickly indicate success or failure. In pursuit of an intrinsic policy, the researchers developed a framework comprising mechanisms motivated by human affect -- one that motivates agents by drives like delight. Using a computer vision system that models the reward and another system that uses data to solve multiple tasks, it measures human smiles as positive affect. The framework encourages agents to explore virtual or real-world environments without getting into perilous situations, and it has the advantage of being agnostic to any specific machine intelligence application. A positive intrinsic reward mechanism predicts human smile responses as the exploration evolves, while a sequential decision-making framework learns a generalizable policy. As for the positive intrinsic affect model, it changes the action selection such that it biases actions providing better intrinsic rewards, and a final component uses data collected during the agent's exploration to build representations for visual recognition and understanding tasks.
smile, or smirk? (Score:3)
so.. (Score:2)
But (Score:2)
First they have to figure iut how to improve the AI
Re: (Score:2)
I think the movie version is going to be a sequel to The Beaver [wikipedia.org].
Re: But (Score:2)
improve the AI
Easy, pull the plug and cut the cord. I don't really trust anyone with a stupidphone that has an open microphone and an unblinking camera lens bonded to their person.
This is stupid (Score:1)
AI does not need to experience happiness, sadness, or even satisfaction with its work. Why on Earth would you want to give it such capability?
Why do I want a pair of scissors that can be happy or sad? Should we make scissors that can get bored when you don't use them?
AI programmers are idiots who want to anthropomorphize every fucking thing. Fuck your Clippy, idiots.
Re:This is stupid (Score:5, Funny)
Make the scissors happy, Johnny. You remember what happened last time, when the scissors were sad.
Re: (Score:2)
This is how the Butlerian Jihad
Re: (Score:2)
"AI does not need to experience happiness, sadness, or even satisfaction with its work. Why on Earth would you want to give it such capability?"
They can't build an AI scientist yet, but for a standup comedian it's enough.
Re: This is stupid (Score:2)
F--- your clippy, idiots
We have just found our winner, gents: nothing says Welcome your AI overlords, and the horse they rode in on quite like our articulate poster above.
Re:This is stupid (Score:4, Interesting)
The AI in the book decides to tack a large number of extra digits onto a janitor's paycheck as a joke. A technician is called in, he discovers the source of the joke, and then offers to teach jokes to the AI to correct the problem. This results in the AI learning that some thing are funny once, some things are not funny at all, and yet other things are decreasing amounts of funny by repetition. Shortly after, the AI begins additional training with another character in the book and learns that the set is not identical for all users.
It's quite valuable to give real emotional feedback to train an AI that A) some things have diminishing returns, B) some results are user specific, and C) there's no way to guess these correctly in advance.
A pair of scissors that can smile isn't particularly valuable, but a pair of scissors that could be fine tuned to work better when smiled at would be potentially useful.
If a scowl is sufficient to get rid of Clippy, they've added value to the product.
Re: (Score:2)
Technically most reasonable people, want an emotionless, clean, invisible delivery of service ie the plate of properly prepared food arrives no fanfare. Egotistic freaks want the servants so fuck them, why serve them at all. So an emotionless automated kitchen, your provide the raw ingrediants and it prepares the cooked food and you pick it up at the counter when it is ready or have a simple remote simple bot on wheels run it out to you and you pick it up off it's top and drop the dirty dishes there when fi
Changing someones emotional state (Score:2)
Is the goal of the personality disordered. It maybe unwise to teach AI how to create personality disorders.
relax, and let the hooks do their work (Score:2)
Can't wait for the first application that requires you to smile or it won't let you log in.
Re: (Score:2)
"Please drink a verification can to continue"
http://i.imgur.com/dgGvgKF.png [imgur.com]
I, for one... (Score:2)
...welcome our new pizza-bringing sexbot overlords!
Re: (Score:2)
That made me smile. Oh fuck - there's a knock at the door.
Russia (Score:3)
This definitely won’t work in Russia.
Re: (Score:2)
Re: Russia (Score:2)
in Russia AI smiles at you
All I can see is a poster of the Joker with the heading, "Big Brother is Watching" and at the bottom, "Why so Serious?".
Russian coworker story--smiles are for crazy folks (Score:2)
In Russia, they think you're mentally ill if you're not smiling for some clear reason...seems to be correct from the 100s or so I ha
Submit to your AI overlords with a smile (Score:3)
Re: (Score:2)
This is America. It already is too late. (Score:2)
Smile or die [youtu.be]. (RSA Animate)
Signs of stroke (Score:2)
This could work well ... (Score:2)
Note: Not when you are happy! When you smile! (Score:3)
So be a good psychopath, and don the most realistic smile, and you'll be accepted by The Society Of Psycopaths. Then you can stay being a dick all day, because this society is fake fakety fake as fuckin fake as it can be.
Don't ever try to have a real(istic) mood that fits your dire situation. Besides, it is't /dire/. It is minushappy! Doubleminushappy! So happy! So what do you complain about, negative nanny? Back to work!
Re: (Score:1)
Re: (Score:2)
Re: (Score:2)
Future looks good for Serial Killing (Score:2)
To put it more explicitly... (Score:2)
...many of the responses to this post have alluded to one simple fact about people's facial expressions: They don't tell anyone how we're actually feeling in any consistent or reliable way. Human feelings are frustratingly complex & evasive to study & to make matters worse, too much of research in psychology & the social sciences relies heavily on self-reporting, i.e. subjects telling researchers how they are or were feeling.
We don't have a valid, reliable conceptual model of how human feelings
When you're happy and you know it, fool AI (Score:2)
AI controlled robots ( and they are coming ), will misinterpret facial expressions because not everyone is happy all the time.
Not good for us autistic folks (Score:2)
Re: (Score:1)
Thought is the opposite of progress (Score:1)