/g/ and AI

still early days

>>60062937
>What does /g/ think of the AI revolution? That is, not a martial revolution, but the advancement of AI in general?
Still pretty early desu. I mean, with deep learning AI has gotten an interest boost, but it's still fairly limited to pattern recognition or highly narrow domain specific tasks (like driving a car / car safety features).

>om we wuz a K5 civilization to Skynet, and a lot of different estimates as to when it'll happen.
A general-intelligent AI (and a successive super-intelligent) is nowhere near happening, and I personally don't believe it will ever happen. AIs are particularly good at solving domain tasks, but they will never be able to understand context.

For example: An AI algorithm may be extremely well-suited for telling dogs and cats apart. But a human can see a picture of a dog looking animal in a tree and infer that that cannot possibly be a dog (because dogs don't climb), whereas an AI would simply look at features and have no understanding of context.

>>60063257
>whereas an AI would simply look at features and have no understanding of context.
i see no reason why an AI can't be taught to recognize both a dog and a tree, the fact the dog is "in" in the tree, and find relevant problems with the fact that the dog is in the tree, given information about dogs and trees
current AI's are limited in scope, but don't assume what we current have is as good as it'll get

>>60063284
It was a simplified argument, my point here is an understanding of context. AIs are good at making decisions based on FEATURES (aka pattern matching voodoo), which is why AI algorithms excel at tasks that involve classification, face recognition etc, language interpretation etc. But context is not something you can translate into a feature. Sure, you can make an AI understand "the animal is in the tree" and therefore teach it that it cannot be a dog, and I guess you can provide enough images of dog-looking animals in trees that would falsely be classified as a dog and then train your AI to understand that a dog cannot be in a tree, but that's still a feature (animal in tree) and not really a context.

>>60063313
>AIs are good at making decisions based on FEATURES
current AI's only do that, but in the future, this could be just one component of a more general AI

the publics understanding of AI is so far from reality and everyones acting like a sensationalist retard just because some google bot can play go

>>60063313
>AI excels at face recognition
it's nowhere close to a human

>>60063313
from what i understand, human thinking can be boiled down to little more than;
- pattern recognition
- memory storage and recollection
- "instincts", not really sure how to define these, but given their persistence through vastly different lives, it would seem they are something besides memories

AI's don't currently have the wealth of memories an adult human has about the world, i believe it's this that is their primary limiting factor

File: 68a76848e6b0eeb7bd8a2774125c5d16.jpg (834KB, 2060x1236px)

834KB, 2060x1236px

>>60062937
I doubt we will ever see a sentient AI in the form of "cogito ergo sum" appear, but rather a highly efficient imitative form of general intelligence that is composed of multiple specific "intelligences". An AI with inserted motivations and drives might imitate a sentient being but would still be following procedural instructions (albeit perhaps hihgly complex). If that is considered being sentient, then we need to re-evaluate our existence.

With the power of computing, this will take pattern recognition, data sorting and other menial tasks (repetitive tasks) out of the hands of humans, leading to some social unrest. I personally think that this will gradually put people in a hierachy in the same way of former pharaos, but in a modern, global situation. Perhaps this could leads to some neo-monarchies with "god-given" powers of the elite, which will be re-written by the elite to brainwash their human cattle, but that is just my speculation.

Essentially, we will probably in our lifetimes see amazing improvement of life through data manipulation and decision-making processes, implemented in daily technology (especially promising to me is the development in production). Human-based intuition and manipulation of the world will still be predominant, since emotions -> drives -> needs are a human thing.

If someone, however, got the idea of implementing a high-level general AI with the programmed capability to have a need for something or unstable emotions, then we might see a new chapter in the evolutionary history of lifeforms on this planet.

>>60063515
Human thinking is way more complicated than you're giving it credit for. That sort of behaviourist idea has been dead since the 60s. I mean it can barely explain language.

>>60063231
>>60063257
Fair enough.