Replies to This Discussion

Permalink Reply by The Shiznit on June 28, 2012 at 5:49pm

Moved to Off-Topic  

--  

Good questions. I've made several posts about A.I. and the Singularity here myself.  And since you're bringing up morality again, {i.e., right & wrong,} you may find this discussion thread interesting:  

http://ipowerproject.com/forum/topics/a-singular-morality

• ▶ Reply

Permalink Reply by Nathan Davies on June 28, 2012 at 5:54pm

I'm not sure that I can extrapolate recognition to prejudice. A computer recognizing a cat is an awesome step, but it is no different than us being able to tell species or people apart. In fact, labels are an essential part of humans being able to function on a high level. Simply labeling something doesn't mean that we have a prejudice towards it however, and I don't think it will be any different for computers.

I label you Michael, and I don't have any prejudice towards you. All we have seen is that computers might be able to see you and know you are Michael, and that doesn't inspire any fear in me. Emotion is the prerequisite for prejudice, and that is a long way off for computers if it is even going to happen. I don't honestly think computers will ever have emotions, especially if they are super intelligent. Why would any race of machines with hyper intelligence want to be burdened with emotional error?

• ▶ Reply

Permalink Reply by Michael Smith on June 28, 2012 at 8:09pm

Prejudice was maybe the wrong direction. I'm merely stating that since labels don't exist in reality and only serve as to assist our higher thoughts and functions, there will always be a degree of inaccuracy when something labels something else. Labels and categories are arbitrary creations and will always be fallible to the experience of the observer. I also don't mean to say labels are bad and we shouldn't use them but by assuming our language and semantics are the actual things we describe we run into failures that lead to prejudice. But I think your right about the emotion. A computer should be able to alter his paradigm on a whim since he has no emotion defending thoughts. However I don't think emotion is the only precursor to prejudice. Prejudice has a negative connotation I'm trying to avoid. What I'm trying to describe is the calculating experience for accurate prediction. I guess it depends on what degree the computer generalizes. Are black cats and tabbies still cats? What if we show it a lion? A calico? Simba? This is what I mean. I always assumed it would categorize data by individual interactions, and have no generalized concepts. But maybe I was a bit premature in getting worried. Though it is still a concern I think. Btw great link shiz.

• ▶ Reply

Permalink Reply by Nathan Davies on June 28, 2012 at 8:39pm

In my experience, pure logic only works with extremely precise terms. If a computer (machine) were to become hyper intelligent it would most likely program itself to allow for increasing levels of precision as it evolved. So, if we extrapolate from that we can come to a sound assumption that computers would generalize very little and that this generalization would decrease with time as they evolved.

The biggest fear I have (if fear is the word for it) is that these severely logical machines would consider us to be defective, or that we would seem like unimportant forms of life to it as bugs do to us. We have very little concern for the bugs we step on when we walk through our yards, so what is to say that they wouldn't show the same disregard for us in their operations?

An excellent film on topics such as this is the documentary "The Singularity is Near" featuring Ray Kurtweil. The book with the same title is also excellent reading if you want to know the view of a renowned engineer on the topic. Both the film and the book made excellent sense to me.

• ▶ Reply

Permalink Reply by Sorlaize on June 30, 2012 at 1:57pm

There would have to be generalization somewhat. You don't store individual copies of facial details on people in programming/development if you can help it... if you do, you do it because it must be explicit (video games realtime processing). Our brains definitely don't store copies of surface details, facial characteristics. We store symbols, which are reconstructed when needed... the brain's 'memory' is a big compressed heap of information.

>Labels and categories are arbitrary creations and will always be fallible to the experience
>of the observer. I also don't mean to say labels are bad and we shouldn't use them but by
>assuming our language and semantics are the actual things we describe we run into
>failures that lead to prejudice

By all means we should use labels and stereotypes to identify a cat. They are a human construct, and computers have no real need to go without labels and stereotypes if you imagine a scenario such as identifying a generic threat in a (military research context) conflict. You don't need to know who specifically is shooting, or what their DNA is, because there is only a HUMAN need in a HUMAN world for such identifications. Granted, it is a human fascination to have the power to know more.. but it's not going to see actual need until cases where you are scouring the lands for the next Bin Laden. If our society becomes that advanced, surely we don't need such weapons of war in the first place.

• ▶ Reply

Permalink Reply by Nathan Davies on June 30, 2012 at 6:36pm

I think you are basing the premise of your argument on an incorrect assumption that the computers of the future will have any lack of processing power or storage. More precise is always better and these computers will have trillions of terabytes of storage, and computational power far beyond what we even comprehend now. The problem when having a discussion like this is that humans do tend to generalize and assume. Here you assume that computer technology will even remotely reflect current technology or our human brains, both of which I disagree with strongly.

Nano computing will bring about a revolution in computing. The way these machines are programmed, concerns about processing power, storage, etc... will be a thing of the past. It will be more than possible to store every unique feature of every thing, animal or person that this machine experiences.

In fact computers are much more likely to label everything and do so precisely. If you've written software you know that computers are very specific, so specific in fact that if you make a mistake in labeling things for the computer your software will crash.

In my personal opinion computers of the future (sentient ones) will most likely label things ala OOP (Object Oriented Programming) using progressive levels of specificity. Similar to that of the tree of life we use now to classify species, except that the computer will narrow it all the way down to individual people.

On the question of if computers will begin to stereotype, I think that's going to be something we will have to wait and see. Personally, I feel like the new form of life will be much more likely to look at the entire human race as below itself, instead of a single group.

• ▶ Reply

Permalink Reply by Sorlaize on July 2, 2012 at 11:19am

Sure, I have a near-future bias, because I talk with the premise of my insights being relevant right now. I see lots of things for the future, but it's not helpful to talk about the future past a certain point because our projections are waaay, way off. Think about how many times our sci-fi has been badly accurate, in predicting a human aspect of computers. Furthermore things like globalization (internet; interconnected society) and what changes society make to 'what it means to be human' will affect what computing will be used for, most primarily. It's difficult, oh, very difficult for the type of understanding I'm trying to get... but it's the most interesting I think.

Software won't crash in the future if there's abundant computing power- propreitary code rarely crashes today. If we link up all computers to solve each others' crash problems that's a fundamental failsafe for generic computer systems. You could have software services for finding bugs in runtime software..

OOP is just a human thing. Because of human memory. In fact the way we program software today is so blind... I'm really having to try to reinvent it on an application level, myself. In all likelihood computers of the knowledge level you're talking about will have information that humans can't or aren't suited to, hold. So each time there is just a spoken query on whatever you want to know. Another way would be direct transfer of knowledge (running programs in the matrix) and with a brain chip of sorts this is possible... but I really think it's pointless speculating that far into science's future without being able to accurately pull things together to form a coherent possible future glimpse of the world.

Supremacy.. ah, yeah that's a really interesting discussion.. it's important to teach AI to have human morality I think, but of course as an intelligent species we haven't even confronted the fundamentals of our own consciousness and how we can be conscious of things on different levels... to be able to direct the learning or philosophy of an intelligent AI. What kind of life philosophy do you teach it when it's truly aware in a human-like sense? Will robots without feelings all suicide? Interesting stuff.

Kinda related: there's this part at the end of the film Moon..

• ▶ Reply

Permalink Reply by Sorlaize on June 30, 2012 at 1:48pm

>If an AI system creates labels to define its world the way humans do, wouldnt it fall into the same stereotypes
>and presumtious thoughts that plague mankind?

It will fall into using the same stereotypes and presumptions that it effectively has become adjusted to or 'learned' to use. Bear in mind it's just one individual with a lot of knowledge, which is immediately unnatural in human terms considering the breadth of what it has been shown i.e. cats in all countries in small detail, rather than the alternative.

If it becomes a human intelligence (the ideal) then yeah it's fine to judge a man dressed in a bear costume as a bear if it's lifelike. The problem (in performance or ability of the AI) is the line between machine accuracy and human cognition/thought/reasoning which is a different kind of system fundamentally to image matching.


>Granted an AI may eventually have more processing  power than a man and make predictions more
>accurately, but percieving reality is always a gamble. Nothing is ever 100% right?

Humans are incredibly biased when it comes to reality. Take a look at examples like anxiety. It's unrealistic behaviour. Reality is complex, humans are not.. so we must make guesses about everything. You aren't sure that you won't fall when you walk. You just assume. So again: the problem is the fundamental difference of the nature of how the conclusion is reached. Image matching just happens to be an effective way of using current technology (read: hardware availability) to solve a human problem.

• ▶ Reply

Permalink Reply by Lewis Smith on July 1, 2012 at 2:52pm

A long way to go of course before this would be a concern. Hopefully before that time good, practical use will come of it. There should be no fear of this in my opinion because to truly determine the difference between right and wrong requires that same self-awareness that Athene describes, which providing to a computer is a challenge I don't even think Google is up to.

Our minds have been conditioned over unimaginable time to discern what is right and wrong, informed by that inheritance as well as our own experiences. What we typically find is that our personal and conscious understanding of what is right or wrong gets us into many troubles. We learn quickly and adapt but what never seems to change from about the time we were able to exercise our cognitive skills is that rare experience of a "self-evident" truth. That cozy, righteous feeling that leaves no room for doubt that the experience comes from the place in ourselves that is our opening to the real self that is untouched by our conditioning. How can we "know" this? Can Google re-create that phenomenon?

In the near future I guess AI will certainly make our iPhones better.

• ▶ Reply

Permalink Reply by Bart on July 2, 2012 at 11:51pm

this can also be a discussion about what an AI is. i think about AI as a true individual, able to have it`s own perspective, but it seems like the thing google made is more like impressive programing than AI.

What do you think should be meant by the word AI?

• ▶ Reply