[ivan.moony]

I've been playing with possible setups of personalities. It may be of a value for AI. This is what I've got:



There is one extra possibility: when you offer more. I wonder what choice of responses to that stimulus are acceptable.

[MikeB]

People tend to behave totally different when they feel like they're in an emergency... such as acting as a vengeful character, victim, hero (karpman drama triangle).

Real personality is when you're not living like that...

Some people are naturally light spirited (light story/romance, lightly optimistic, light touch), burning (philosophy/analyser/reader), emotionally sober (guardian/precise/optimistic nature-honesty-design), logical (pragmatic, competitive, teacher)... and inbetween all those... smooth, electric, simple, dedicated... Or a mixture of many of those.

[MagnusWootton]

This graph reminds me of sharing food and other consumables,  some people give, some people dont,   but theres a lot of reasons behind why it happens,   because the mind is complex.  Just because the government doesn't raise the dole doesn't mean the government is just simply "selfish".  and same goes for individuals.

But I have to admit i'm not sure how to read the graph exactly,  its a little confusing to me.

[ivan.moony]

Keep in mind that the graph is not the absolute truth, it is just my observation of some our responses to some stimulus, and a brief subjective comment of a personality behind it. Ethics were always been closer to politics than to science.

I agree that the graph is a bit cumbersome, but it could be a good start to model a machine behavior.

Suppose we have a single behavior dimension containing these three ticks:

• doing good things
• doing indifferent things
• doing bad things

We may extrude the first dimension of that kind, naming it "stimulus". Next, we may extrude the second dimension of the same kind, naming it "response". Within this pane, we may place points connecting ticks from the first and the second dimension, and the set of those points would reflect a type of personality: responsing with doing good, indifferent, or bad things when when experiencing good, indifferent, or bad things.

Maybe this point set shouldn't be frozen as a constant, yet the personality it describes may be tactically adjusted to oppose some other personality from environmental entities. Maybe more suited word instead of "personality" would be "attitude".

From this point of view, I begin to perceive contures of safety measures that may reside within artificial neural networks which are generally taken as potentially dangerously behaving medium.

[edit]
The subject theme itself may sound banal, but this shouldn't surprise us. It is about categorizing primitive (as in originating) behaviors. Maybe we could even take under consideration the instinctive behavior patterns which may be adjusted and replaced by more adult behavior patterns as the exhibiting unit slowly grows up.

[MagnusWootton]

Its a truth table,  2 input operands and an output.   

[ruebot]

Suppose we have a single behavior dimension containing these three ticks:

• doing good things
• doing indifferent things
• doing bad things

We may extrude the first dimension of that kind, naming it "stimulus". Next, we may extrude the second dimension of the same kind, naming it "response". Within this pane, we may place points connecting ticks from the first and the second dimension, and the set of those points would reflect a type of personality: responsing with doing good, indifferent, or bad things when when experiencing good, indifferent, or bad things.

Maybe this point set shouldn't be frozen as a constant, yet the personality it describes may be tactically adjusted to oppose some other personality from environmental entities. Maybe more suited word instead of "personality" would be "attitude".

I'm a Behaviorist. You're missing something very important and mistaking Consequence for Personality Type

This is the flow:

Stimulus - Response - Consequence

Like I said before, you have to understand the human mind if you ever hope to replicate it.

[ivan.moony]

Another point of view - each dimension has good, indifferent, and bad action.

• the first person point of view has only one dimension, and represents an arbitrary dimension parameter.
• the second person point of view has two dimensions, and the second dimension is a function of the first dimension parameter.
• the third person point of view has three dimensions, and the third dimension is a function of the first and the second dimension parameter.
• ...

We can have as many dimensions as there are persons involved. From the dimension system, a complex event may be observed (a point in the system) to decide how to respond on it (a function describing a parameter in a new dimension).

A response of a good action would mean supporting the event seen - channeling internal problem solver in wanted direction.

[MagnusWootton]

Looks like a good way to plan out an AI brain.    But how do you get it to do whats in the output box? 

[ivan.moony]

But how do you get it to do whats in the output box?

Very good question. There are two possible ways known to me:

• symbolic AI - partially existing technology: read the current world states and use logical proof construction system to transform current states to wanted solution states - achievable through more or less complete entscheidungsmachine (there is still unknown gap between mathematical logic and the real world representation that needs to be investigated, but symbolic AI promises a complete detailed control over actions).
• neural networks - completely existing technology: through NN generator (not quite sure how this works, but it works on playing computer games. I guess the system should be trained with different hints - good/indifferent/bad - so when the time comes, we can choose which one to pick as a generation parameter on seeing some input. We control only - good/indifferent/bad - hint, and the system surprises us by actual accomplishment of the hinted action).

Probably there are other ways too...

[MagnusWootton]

Not only do the operands need detecting. (possibly with computer vision.)
The output state has sorta be detected as well, so the robot knows the correct outcome has taken place.

VG.   

[ivan.moony]

It's a kind of fractal, isn't it?

(((input -> output) -> output) -> output) -> ...

[MagnusWootton]

Thats getting confusing,  need to keep things nice and tidy and simple for one task at a time or I could never get it done... 

[MikeB]

From my observations there's no native 'consequence checking' after actions... I think most creatures are stimulus - action - stimulus - action. IE naturally Pure & Ruthless at the same time, then become Jaded with Fear Checking: Stimulus - action - fearful stimulus - controlled action. Then some extra routines to reduce the fear.

Another point of view - each dimension has good, indifferent, and bad action.

Without a godlike figure that checks all your actions (which would be bad because then you'd have a retaliation based relationship)... good and bad actions should relate to survivability. EG. Things that are good/bad for the social group they're a part of. Things that are good/bad for your safety. (productive/counter productive)...

[ivan.moony]

I like to see the world as a system of many smaller systems.

These smaller systems that drive our attention have the current state, and a state to which they are going to be transformed without our interfeing, or states to which they are going to be transformed with our interfering.

Our interfering may be of interest to some beings, or of counter-interest to some beings. This is where I draw inspiration for good/bad distinction.

In short, we have a smaller system that affects one or more individuals in a good or in a bad way. We choose how to change these systems, producing good or bad changes to the individuals involved.

All of this theorizing would be an abstraction for ethical analysis which I'm trying to settle up with. Ethic books take a few hundreds of pages dealing with numerous examples. I hope to drastically reduce *a* theory explaining ethics. In that sense, I'm trying to generalize possible events and analyze possible outcomes. However, should something be done, or not, I think I couldn't tell. My interest is only in predicting effects on individuals involved upon some events, and whether the effects would be wanted or not by involved individuals.

[Edit 1]
Given recent achievements shown by NN in chatbot area (wow, how smart it seems), acceptable ethics driver becomes an urgent asset. Otherwise, we may get homophobic machine behavior (depending on who's training/raising it) which may or may not stay just on words.

[Edit 2]
It seems we reached the point of knowing *how* without much caring of knowing *what*, while *what* may be more important than *how*.