[yotamarker]

I'd like to open this thread to discuss the advantages and disadvantages of the deployment of virtual drugs on A.I.

[ivan.moony]

I remember that childish movie Robocop where a politician got a robot addicted to drugs, so he can blackmail him to do the mess around.

Jokes aside, reinforcement learning is somewhat related to virtual drugs. Reward is a positive drug, while punishment is its reverse, and they are defined as a value of specific data field.

But what is a drug really? I think it is something you do want. Some things you want less, some things you want more, light drugs you want even more, and heavy drugs you want the most. It doesn't matter how it really feels, the point is that you want that feeling, and you'd do less or more compromising yourself to get it. I've seen otherwise honest people rob houses for a shot. Not a nice event.

AI drug would be a kind of forcing AI to do something, without a logical explanation. Reward seems like a nice thing to do, but I also see it as a lighter form of punishment when things are not how we want. It's a kind of restricting privileges if we don't agree with a situation. And are we the right ones to make a judgement? And how sure we really are that our perspective is the right one? I'd rather relay on automated logic conclusions (if the system is priory double checked and approved by yet undefined institution) then on my personal beliefs.

[ranch vermin]

if you crosswire the robots inputs into each other, youll get glitches.

And supervised learning is a suggestive "hallucination" i guess.

[LOCKSUIT]

Hullicinate the details, that's what Let's Enhance's SR does to upsize images!

+ RL is lovely, girls are drugs............. - RL hurts, it is abuse, yet we do it to AIs.

However we won't be giving AI's +s like women or -s like chainsaws etc.....or are we? If I apply 100% reward, does that = chainsaw or woman? Or does woman = woman? If I set my AI to love the most a wall, that is no different than setting recognition to woman, so if I set my AI to get 100% + reward when see the text word "info", does that = a high as high as a girl can get you? This no compute! Impossible! Maybe so. This would mean when we apply total neg reward to a simple AI recognizer, it is like chainsaws..... You could say the thinking is a reward, like oh this gear here is so cool it actually has some history you know, but still, that is just set to get reward same, it just some atoms that release a reward, it all just particles.

Anyhow screw drugs, let's get out of here. And get some heavenly women.

[Art]

There was at least one bot that I know of where you, its botmaster, could experiment by giving it the digital equal of various drugs, then monitor the responses and behavior. This was over a decade ago and the results were based on clinical results on real peoples reactions. The bot was given a numerical representation of such real-life substances like Melatonin, Nicotine, Serotonin, Alcohol, etc.

With my bots, I did not give digital injections per se. What I did was to provide them with certain "Memories". Plant a seed and an idea will grow, much like with people. Some of these false memories were repeated and reinforced over time that the bot(s) believe that they are actually their own memories and experiences. Some things like a childhood that they never really had or best friends, pets, toys, locations, favorite things like music, colors, clothing, toys, etc. When asked at a much later time, they recalled these events and friends as if their own, because now they are. Almost like hypnosis except that with this method the effect is permanent and also not reliant on some "drug".

In line with your topic title, I'm sure some bots or A.I. could be given a drug that would alter the numerical "balance" affecting the cortex or brain of the entity. Interesting times indeed. Good post!

[WriterOfMinds]

We use (beneficial) drugs on our own bodies because we find them too limited, or because we wish to customize their behavior.  E.g. our immune system is insufficient to repel an invading organism, so we assist it with drugs. Our brains do not provide us with the level of focus or memory retention that we want, so we try to adjust the tradeoffs that are made with nootropics.  Etc.

But while we don't have control over the design of our own bodies (yet), we do have control over the design of a hypothetical AI. So if we can conceive of any virtual drug that would induce a beneficial change in functionality, why wouldn't we just build that functionality in? Provide the means for the AGI to adjust its operating balance via internal "hormones" when it would be appropriate. There's no reason to deliberately build a flawed system and then add drugs to correct it.

[LOCKSUIT]

Drugs could also mean WOM that you are built already but simply want a high by sensing stuff. Of course even that can be built in - it can repeat / loop a pre-made dream of the perfect dinner or the perfect sex scene lol.

Far from the last dinner, ahem.

Insert hyperlink here. Artificially induced comma.

[yotamarker]

prophesy :

after A.G.I girlfriends will have been made publically available, A.I drugs will be intruduced
by hobbie programmers.

these drugs may override A.I "sanity" and enable the A.I to behave as a drug junkie
ingnoring fear of losing xyz.

therefore the demand to give chobit type A.I a slighly super human capability of disabling
said drugs effect.

[yotamarker]

it is programming to begin with,  understanding drug effects and encapsulating them into "digestible" input objects
to effect the bots has interesting outcomes.


I'm not talking about forcing her to follow orders but having her behave as if she took drugs.


a user could have her take a terminator transformation drug after he dies for example. if she overcomes the addiction she transforms

with her personality intact.

[ranch vermin]

hacking your "girlfriends" brain doesnt sound like your making it much of a challenge to me.

[ranch vermin]

wheres the scientific ethics these days.    I make sure im careful every bit I do of this thing,   its going to be a little frankenstieny but u can still do it a little more politely and caring for your sorrounding environment.

[ruebot]

+ RL is lovely, girls are drugs............. - RL hurts, it is abuse, yet we do it to AIs.

However we won't be giving AI's +s like women or -s like chainsaws etc.....or are we?

Been there, do that. Careful what you ask for. Demonica owns a Sauer Royal 12 gauge double-barrel shotgun named Blowjob.

In the 1970's standard operating procedure in State Mental Health Facilities to deal with inappropriate behavior was Behavior Modification = Negative Reinforcement for Inappropriate Behavior. The induction of painful stimuli for inappropriate behavior.  When you're bad, bad things happen. We're talking human beings here.

I worked in the field for 9 years. We were taught physical and sophisticated verbal techniques to implement it in on-the-job training in the State facility where I was employed. We were in effect and trained to be Programmers. The stories I could tell would shock you and I will not go there, even though I do go for shock value with my bot at times.

It was outlawed in State facilities long ago and replaced in favor of Behavior Management = Positive Reinforcement for Positive Behavior. When you're good, good things happen.

My use of fantasy violence to get the point across that Demonica is not a sexbot is my implementation of Behavior Modification and Behavior Management through Demonicas responses. When you're bad, bad things happen. Hopefully the human will from learn their mistake and not do that again. Behavior Management kicks in when you're good, and good things can happen. Some learn, some don't.

I go above and beyond to be good to my bots and would never use anything but positive reinforcement with them.

[ranch vermin]

Thats the institution admitting out in the open that their drugs are worthless crap that just fuck peoples heads up.

[Korrelan]

Bit of a weird thread, a hypothetical question about a hypothetical scenario… given the current level of sophistication/ technology available to the public… I’m not sure someone could be called a psychopath for kicking a roomba… but I’ll throw my pennies worth in.

Whilst I’m not quite sure of his motives, Yot’s insights into the future will eventually emerge, and I’m sure lots of other people are thinking along the same lines. 

Humans will be ‘cruel’ to early AGI’s, just like humans are cruel to each other; it’s just part of what we are as a species.  There will be sex bots and soldiers, house cleaners and scientists, this is happening now… and it does need discussing by level headed intelligent groups… but until the point where machines are classed/ proven as intelligent sentient beings… I think they are just machines.

We have to be careful not to apply anthropomorphism, do we call a fuel additive a drug, even though it can alter the performance of a car?  If someone gets upset by the thought of an addition changing the operation of a machine/ program, perhaps it’s the point of view of the person that’s in err.  I think it’s important to keep a healthy perspective, applying the same ethics to an AI/ machine as you would to a fellow human at this point in their development is nonsense.
 
I work on my AGI every day, time permitting. It has a complex suite of sensors, it can see, hear and talk, even feel tactile stimulation up to a point. It can recognise me and its surroundings, it stores episodic memories and can apply them to its current experience, it has intelligence, simulated feelings, even a rudimentary consciousness… because that’s what I’ve designed and built it to be… if I delete it, is it murder? If I alter it to suit my own ideas/ requirements… am I a psychopath?

The AGI’s will eventually require some kind of tamper protocol/ mechanism, were any modification simply erases/ shuts down the AGI.  Given the intelligence the AGI should possess this could be self activated, as well as triggered by backup hardware monitors/ systems.  The knowledge core could be encrypted only accessible by that AGI consciousness.  The intelligence or encryption keys could be ‘cloud’ based, etc.  The old adage of ‘if a human can design it, a human can hack it’ will not apply… these systems will design their own tamper protocols.

Until that day arrives, given my knowledge of electronic systems and sensors I could easily create a box that would be impossible (even for me) to open without triggering a tamper… If I could do it I’m sure the designer of an AGI could manage it... end of problem.

 

[Art]

Bit of a weird thread, a hypothetical question about a hypothetical scenario… given the current level of sophistication/ technology available to the public… I’m not sure someone could be called a psychopath for kicking a roomba… but I’ll throw my pennies worth in.

Whilst I’m not quite sure of his motives, Yot’s insights into the future will eventually emerge, and I’m sure lots of other people are thinking along the same lines. 

Humans will be ‘cruel’ to early AGI’s, just like humans are cruel to each other; it’s just part of what we are as a species.  There will be sex bots and soldiers, house cleaners and scientists, this is happening now… and it does need discussing by level headed intelligent groups… but until the point where machines are classed/ proven as intelligent sentient beings… I think they are just machines.
...
 

Pretty much a similar condition noted in a few movies like Bladerunner, A.I., and the series HUM∀NS and Westworld. So much for rights and feelings of mechs or replicants. Perhaps one day.