What about AI free will?

Probably our most direct experience with AI has been with video games. Characters in these scenarios have a list of code options which run behaviors and animations based on the environment, player activity and A to B scripted scenes. These have all been carefully orchestrated by the programmer to give the illusion of greater complexity and versatility of AI behavior than really exists.

Lets say that when a player character enters a room an enemy Ai has a number of options for what to do next and their probability weighting depends on what is known about the player.

If the player enters the room with low health, the Ai enemy might attack. If the player has high health and a powerful weapon, the AI enemy might flee or activate an alarm to summon re-enforcements.

Currently these behaviors give the illusion of some complexity, but really amount to one of perhaps six options the programming could execute.

But think of a future AI with a human-like body and complex programming. This AI is only tasked with exploring an environment, rather than assessing the combat ability of an opponent. It’s walking along and comes to a fallen tree which blocks it’s path. If the AI is sufficiently complex you won’t know exactly what it will do to overcome this obstacle. Nor will the programmer know, specifically, what the AI will do. They could tell you that the robot has a number of options for getting past the obstacle, but not exactly which choice it will go with.

There are still a number of choices it could execute and those would be probable in accordance to pre-arranged assessment codes designed before hand, but the resultant choices would be in the thousands, or hundred of thousands.

No different, really, than what a person would do in that situation. We evaluation the obstacle, evaluate our own abilities and then choose the option we think has the highest chance of success. Try to batter our way through the branches? Go further down the tree to where there are fewer branches and climb over there? Detour around the tree completely? All are possible and the choice between them depends on our previous moments. Whether we are energetic or lazy, confident or cautious.

An AI would be put with the same problem and the same options for dealing with it. The AI’s responses are determined by it’s programming and it’s hardware, which are “written in stone” so to speak, but the individual challenges it faces will not be predictable, and so the individual choices it makes to navigate those problems will also be unpredictable.

We also have what amounts to “written in stone” behaviors. If you touch something hot you are going to jerk your hand away. Theoretically you COULD leave your hand here and let it be destroyed, but it just isn’t going to happen. If you are functioning normally, you will jerk your hand back. Even though our nervous system is built up of simple cause and effect reactions of this kind, the huge number of them combined with the constantly changing and unpredictable environment we find ourselves in leads to the phenomena that we call free will, or self determination.

How does that differ in quality from what happens with an artificial intelligence faced with the same environment? We are quantitatively different, not qualitatively. The kinds of things that happen in our brains are more complex because they emerge from a broader range of simple interactions in our brain than what we’ve programmed into any AI, not because they are fundamentally different.

So a sufficiently complex thinking machine could be predicted to try to overcome that obstacle in general terms, but not specifically. Just like a human faced with that obstacle could be predicted to try to overcome it generally, but not specifically, and to the same degree, I think.

The only way to predict the activity of the AI with high accuracy is to precisely control the environment and all variables. But the same could be said of a human, couldn’t it?

What do you think would be real grounds to distinguish ourselves as being self determinate, and the AI as "merely following a program"?

This was written by Jeremy Bentham. I buy it!

[...]are reasons equally insufficient for abandoning a sensitive being to the same fate. What else is it that should trace the insuperable line? Is it the faculty of reason or perhaps the faculty of discourse? But a full-grown horse or dog, is beyond comparison a more rational, as well as a more conversable animal, than an infant of a day or a week or even a month, old. But suppose the case were otherwise, what would it avail? The question is not, Can they reason? nor, Can they talk? but, Can they suffer?

Artificial imagination.

http://googleresearch.blogspot.co.uk/20 ... eural.html


These researchers showed an image recognition software noise patterns and asked it what it saw. Then they had the software improve the quality of the image it saw.

It painted pictures.

"What do you think would be real grounds to distinguish ourselves as being self determinate, and the AI as "merely following a program"?"

In what context? In the context that youve set up to box intelligence (whether artificial or natural) into a set of choices to overcome an obstacle like a tree blocking your favorit path in the forest?
Of course our choices are determined by what could only be physically possible in your scenario.

Prior to that walk in the forest the cloudy weather made me FEEL depressed. As a result, i cancelled my daily stroll out in the woods until some sunshine appeared to make me FEEL better emotionally.

So you are saying emotion is what distinguishes us from an AI?

One page 2 i talked about emotion and how an AI with sufficient capabilities shouldn't have their emotional reaction ignored simply because they don't have cheeks, or veins to fill those cheeks with a blush.

What do you think?

johnson1010 wrote:So you are saying emotion is what distinguishes us from an AI?

I suppose that would be a start.

What is the relationship between our intelliegence and emotions and how do both interrelate to help navigate our environment and the complexity of interractions within it?
Our emotions are always projected out into the world and influence our decisions in the process.
we just dont feel anger. We are angry at something.
We just dont feel compassion. We are compassionate intentionally and selectively.
We dont just feel love. We love certain things and certain people in different ways.
Emotions are intelligent.

Can you program that into AI so it cant be distinguished from what it is to be human?

johnson1010 wrote:So you are saying emotion is what distinguishes us from an AI?

One page 2 i talked about emotion and how an AI with sufficient capabilities shouldn't have their emotional reaction ignored simply because they don't have cheeks, or veins to fill those cheeks with a blush.

What do you think?

Saw this after my response.
Ill take a look a little later. I have to go get ready to get drunk at happy hour.
You bring up some great things to think about here.

Here is a quote from the article:

If it's wrong to kill an intelligent robot, is it also wrong to reprogram it? You could say that the machine should decide, and if it says no, the answer is no. You could also say that no matter what, the machine has a personality or thoughts that you programmed into it. A person is shaped by their society, yes, but not every function is pre-ordained when they were made. Doesn't that separate them from an AI? If you made the first personality, why don't you get to make the second? Especially if the second makes the robot happier. At what point does it become a silly, undefined fear of 'playing god

Johnson seems to be asking what are the moral implications after our species has designed AI.

I havent read each post inthis threas yet but a few of my own questions and thoughts are as follows:

Let's forget about the assumption that there will come a time when we've manufactured AI:
Why SHOULD we? Just because we can? Does that line of reasoning immediately follow our scientific prowess - we can so therefore we ought to. ?

If we program an AI to commit acts we find morally objectionable and they do those acts willingly and are "happy" doing those deeds, have we committed a moral wrong against them?

If Johnson's AI creation experiences a program malfunction and kills someone, who is morally responsible?
The AI is not because it was a program malfunction.
Johnson is not because there was no intent.
Are human criminals with bbrain chemical "malfunctions" that cant be helped responsible for murder?

I think the question I have trouble with first is the why should we play Dr. Frankenstein?
There seems to be an elemnt of blasphemous hubris to it all.

You guys probably have all the answere and can think of a reason why we should create something that is by definition unnatural.

Ant wrote:
I think the question I have trouble with first is the why should we play Dr. Frankenstein?
There seems to be an elemnt of blasphemous hubris to it all.

You guys probably have all the answere and can think of a reason why we should create something that is by definition unnatural.

If we were to consider very deep space travel, then AI would solve the issue of how to survive the time involved with extreme distance's . Pick a destination, calculate time and distance, load the craft with fuel, test and communication's equipment, program the AI unit to wake when the craft arrives at the predetermined destination and let it go to work. No food stuffs required. Its a more likely scenario than human cryogenic sleep.
Its an idea implied in PKD's Androids.
The space vessel is the simultaneous object to design.

@ant; was your happy hour all that it could be?

ant wrote:If Johnson's AI creation experiences a program malfunction and kills someone, who is morally responsible?
The AI is not because it was a program malfunction.

Any AI advanced enough wouldn't just have a single program malfunction. It would be something systemic, like a virus. Like a dog contracting rabies.

I could see it happening when real life screws with the AI's moral coding. What happens when "loved ones" die? Could that corrupt the AI like it does a human? When a human with a broken heart kills in revenge, who is morally responsible? What if an AI does the same thing? The person had something of a "program malfunction".

Or what of the people who have "intermittent explosive disorder". It's very similar to a program malfunction. So are people with this disorder not morally responsible when they kill someone?

I think we will end up holding AI's morally responsible. Sounds silly, right? But how do you know how we'll act when the AI passes our own personal turing tests? Have any of you seen Ex Machina yet? This gives a good example of what I mean. I don't think we'll hold the AI to human morality, but it will be a very close approximation. Some moral system we develop specifically for AI's.

ant wrote:Why SHOULD we? Just because we can?

I'm not sure. How do we know the development process won't include many flawed prototypes that feel genuine, constant agony? I don't know if we should, at least until we have a much better understanding. But the way things work, nothing will stop it. As soon as the technology is available, someone on Earth will do it.