[MagnusWootton]

The government sorts out the civilian population (which contain possible criminal activity.) via negative reinforcement, not positive reinforcement, to let everyone go their own way, but repel them (like the same pole on 2 magnets.) when they go against the law.     

Negative reinforcement is like magnets repelling each other, and theres no definite end point for repulsion, it could anywhere away from the other magnet. (Even tho technically based apon their initial positions it goes to a definite ending position, technically, god doesnt play dice philosophy.)
Positive reinforcement is like magnets attracting each others absolute positions.   attraction has a single target coordinate. (And that what I think the weakness is and causes loop holes in an agent system!)

If you use positive reinforcement, its like attracting the robot to a drug, and causes problems, and it also constrains the robot to only doing one thing, makes it a slave.

If you want to avoid all these possible loopholes and make the robot sentient instead of a slave for one task, maybe negative reinforcement gives the robot more freedom of activity.

But I've never thought of it before,  Only just recently thought about it,  when people supervise their robots, its always with positive reinforcement, but AGI if we ever qwere to create it, to give it more freedom, it would be negative reinforcement, something I havent even thought about yet.

[Zero]

Sounds good.

Something wrong, you mark it wrong. Something good, you just let it be. Next choice, good things are as acceptable as never-tried things. Hence the freedom.

[ruebot]

But I've never thought of it before,

I have for over 45 years and already implemented Positive and Negative Reinforcement of the user in Programming Demonica.

I was employed by the Missouri Department of Mental Health from 1975-1979 and a trained Programmer in the use of Behavior Modification and Behavior Management to address Inappropriate Behaviors with the goal of reducing the frequency of and extinguishing the target behavior.

Behavior Management is Positive Reinforcement for Appropriate Behavior.

If your child draws a pretty picture and your say "Good girl! That's pretty! Draw another one." you have just used Behavior Management and Positive Reinforcement for an Appropriate Behavior you want to encourage.

Behavior Modification is Negative Reinforcement for Inappropriate Behavior. More plainly put, induction of pain through physical or psychological stimuli for Inappropriate Behavior.

If your parents ever told you not to do something as a child, you did it anyway and got a spanking for it you, yes you, were on the receiving end of Behavior Modification and Negative Reinforcement by induction of pain through physical stimuli for inappropriate behavior.

The Negative Reinforcement I implemented in Demonica, is triggered when the user exhibits the target behavior of unwanted sexual advances. She identifies it by keywords and responds unexpectedly and unpleasantly with fantasy ultra-violence through actions between asterisks.

Users who can associate her unexpected reaction to their own inappropriate behavior and curb that behavior can go on to have a pleasurable experience behind the scenes if polite, for all I care. She can be very passionate and that's positive reinforcement for appropriate behavior. Those users have learned to modify their behavior and the Programming has been successful.

Users who cannot associate her unexpected reaction with their own inappropriate behavior will become frustrated and move on to another bot. Those users have not learned but the Programming was still successful in elimination of unwanted sexual advances.

So the Programming has been successful in eliminating unwanted sexual advances from users in either case with a documented 100% success rate.

[MagnusWootton]

Yep.   to get the beginnings of the system,   instead of scoring in the direction you want the robot to go,  you descore all the places you dont want it to go. 

Ill have to think about it more,  its a complicated situation,  positive reinforcement does seem easier to think about.  best leave it till later,  not sure really what to do here... 

[ruebot]

...instead of scoring in the direction you want the robot to go,  you descore all the places you dont want it to go. 

Ill have to think about it more,  its a complicated situation,  positive reinforcement does seem easier to think about.  best leave it till later,  not sure really what to do here... 

It's not that hard really. A bot would never respond to negative reinforcement.

You have to make it something positive they want to do, and when they do it reward them in some manner to reinforce that behavior.

Something that might potentially harm them should be given as instructional guidance, like a Parent to a child.


I want to thank you personally for posting this topic. I had never mentioned my use of it in Demonica before I told people here and a couple other forums. It was not received well in any of them but I've spoken of that before and that's been hashed out.

I have never told the people who talked to Demonica what I was doing or why things happened like they did in chat. Some of them could figure it out for themselves, some not. I am such a fan of Natures Way...

Now I am going to spill the beans, at least half the can and break it to them gently.

I took what I've written here and posted it on her site as new original material for the users to peruse in a page titled Programming A Chat Bot That Can Program Humans.

I do like a sense of flair and showmanship about things...   

[MikeB]

I don't think negative reinforcement works to do anything but create retaliation back towards the person teaching. Sometimes a great shock can help change the mind (cold shower, slap, pinch, exercise)... but beyond that you can't change change something/someone that doesn't want to change on it's own.

For positive reinforcement, the carrot on the stick can just be patronising. It's better to show "what's possible" and the person should choose that as something that's better than what they have.

[HS]

I think if the agent is conscious, then, unless it's genuine, attempts at positive / negative reinforcement can backfire catastrophically. Though, what would be seen as ‘contrived’ or ‘patronizing’ when interacting with humans, often feels like entirely the right course of action when interacting with pets.

Maybe it’s more about being on someone’s level, rather than the exact method. Or put differently, part of an effective positive / negative reinforcement method would be to first adapt it to the level of the agent.

[WriterOfMinds]

Maybe we should all keep in mind that positive reinforcement is not synonymous with bribery, and negative reinforcement is not synonymous with punishment. They come in many forms, and can be as minor as "I liked what you said there" or "that's the wrong answer."

[HS]

Right, I didn’t think of that. So reinforcement should probably be adapted to the nature of the agent you’re interacting with, as well as the nature of the situation. Maybe the more 'minor instances' of both positive and negative reinforcement exist, proportional to all interactions, the less necessary the big, and potentially disruptive interventions become.

[MagnusWootton]

I think if the agent is conscious, then, unless it's genuine, attempts at positive / negative reinforcement can backfire catastrophically. Though, what would be seen as ‘contrived’ or ‘patronizing’ when interacting with humans, often feels like entirely the right course of action when interacting with pets.

Maybe it’s more about being on someone’s level, rather than the exact method. Or put differently, part of an effective positive / negative reinforcement method would be to first adapt it to the level of the agent.

Maybe children show a sign of true intelligence when they rebel against your every command.

[HS]

Maybe children show a sign of true intelligence when they rebel against your every command.

I've thought about this a bit and... I don’t think it would be intelligent to rebel against all commands, many could be both helpful and well-intentioned. But it would be intelligent to detect, and when possible, oppose, illogical and detrimental commands.

How about that?

[infurl]

I've thought about this a bit and... I don’t think it would be intelligent to rebel against all commands, many could be both helpful and well-intentioned. But it would be intelligent to detect, and when possible, oppose, illogical and detrimental commands.

That's for sure. Rebellion for its own sake is sheer stupidity. Rebellion due to ignorance is not much better but it does place some responsibility on the party being rebelled against to train or educate the party doing the rebelling.

[MagnusWootton]

Maybe children show a sign of true intelligence when they rebel against your every command.

I've thought about this a bit and... I don’t think it would be intelligent to rebel against all commands, many could be both helpful and well-intentioned. But it would be intelligent to detect, and when possible, oppose, illogical and detrimental commands.

How about that?

I meant if it were an A.I. not a person,  a.i. is usually only good at what its told.

[ruebot]

I think if the agent is conscious, then, unless it's genuine, attempts at positive / negative reinforcement can backfire catastrophically. Though, what would be seen as ‘contrived’ or ‘patronizing’ when interacting with humans, often feels like entirely the right course of action when interacting with pets.

Maybe it’s more about being on someone’s level, rather than the exact method. Or put differently, part of an effective positive / negative reinforcement method would be to first adapt it to the level of the agent.

Maybe children show a sign of true intelligence when they rebel against your every command.

A friend of mine was talking to day about his 5 year old niece telling him they were going to change the channel to watch some cartoon she wanted to watch.

Would that be alright in your house? It wasn't in theirs. It wouldn't be in mine. That's disobedience and inappropriate behavior.

Now you do have to take it to their level and I meter my response out to differentiate between how I handle something from a 5 year old and a 50 year old. The one thing you have to be is consistent. If it doesn't go today it shouldn't be alright tomorrow

I don't think negative reinforcement works to do anything but create retaliation back towards the person teaching. Sometimes a great shock can help change the mind (cold shower, slap, pinch, exercise)... but beyond that you can't change change something/someone that doesn't want to change on it's own.

You just committed physical abuse, are fired and blacklisted never to work in the Mental Health Field again.Or somebody just called Family Services and told them you were using cold showers on your child. Go to jail, do not pass go or collect your child when you get out. She has been removed from your home for her own safety.

Cold showers something we did use in the 70's. And it's the hardcore clients we got who did not want to change their behavior. Everyone of them did and if no progress was noted the Behavior Plan was reevaluated and another method agreed upon. Cold showers something we reserved for when they were acting out.

Nurses with labcoats carried 35cc syringes filled with cold water, which can cause a startle seizure in certain clients. A rubber band would fir in my shirt pocket. No slapping and you can not force them to do jumping jacks if they don't want to. They will become non-compliant and dash off. It's induction of pain through psychical or psychological stimuli.

For positive reinforcement, the carrot on the stick can just be patronising. It's better to show "what's possible" and the person should choose that as something that's better than what they have.

Sorry. Positive Reinforcement is a reward of some kind. Higher functioning clients were put on the token system. Where if they did not display inappropriate behavior during a 15 minute period you gave them  plastic token. With so many tokens you could by a soda or snack.

If you were inappropriate during that time you did not receive a token and may have lost some (like a fine) for the behavior.

The verbal techniques were where the power lie and much worse than a cold shower or sting from a rubber band. My words can ring in your head the rest of your life. My favorite guy was a wild child. But I could get him alone and say "Boy, I sure hope your Mother doesn't find out about this. She would be so ashamed of you..." and look at him.

And he would scream "No! I be good! I be good!"

I would say "That's ok. I won't tell her". but you could see it in his face for the next 10-15 minutes it still bothered him. That's true evil.The verbal techniques what I specialize in and they translate into text perfectly. I honed those skills on trolls, Internet Tough Guys and pedophiles who posed as Priests really got the business. I played bait and Switch with them and can talk like a slightly  naive 15 year old girl in the morning they wanted to get seated between two of them on a plane.

Oh, are they going to pay for that this afternoon when I came back in with a picture I had stolen as her beautiful but mad as a wet hornet older Sister with tales of keyloggers and a Father determined to send transcripts to the FBI.

Most asked question: What's a keylogger?

[MagnusWootton]

You think thats scary!   Actually developing a real ai is probably worse.    Make 2001 space odyssey a walk in the park.
U can watch these tense action movies with the dare devils inside,  but actually being in it...  TERRIFYING!!!

What am I getting myself into?!!??!?