AI?

With companies such as Google, Microsoft, Nvidia, IBM, OpenAI, and other working towards the advancement of this particular field, how long do you predict it will be until an advanced general intelligence (AKA Human Intelligence). It seems like an overhyped field. They have some pretty visuals of them in action, and some seemingly useful applications, but they are just tools that accelerate already existing functions fulfilled by humans or non-learning algorithms.

>but they are just tools that accelerate already existing functions fulfilled by humans or non-learning algorithms That seems like our own undoing because we don't know how to utilize it the best way. I have personally seen some applications of AI self learning functionality that seemed really out of the box. Like that one that solved mathematical connundrum about number of steps before the ledge or something and another one where convolutional neural network was trained to imprint shapes and patterns onto other pictures and export the images of imprints (pretty cool visuals). I think there must be some applications in cyber security (especially the exploitation and firewall parts) but those are yet to be examined. The general intelligence? Meh, I would say that this question should have been addressed 10 years ago because I am pretty blackpilled by that topic. I say that it's already over and that we should beg our future AI God for forgiveness.

> "how long do you predict it will be until an advanced general intelligence (AKA Human Intelligence)" > "but they are just tools that accelerate already existing functions fulfilled by humans or non-learning algorithms." By taking two things you yourself said, you can see that you are dismissing exactly what it is to be human. If you strip away all the emotional elements and "spiritual" aspects of being a human, we are at the end of the day systems that execute a set of functions on a daily basis. So executing human functions at a faster rate would in itself be a big thing. In a book on artificial intelligence the author made a point about statements like this. We want talking computers that can do all we can do and more. And when we have breakthroughs in AI, an interesting thing tends to happen. We move the goal posts. We don't know what we are looking for so at the end of the day the average person takes a breakthrough the same way they see the next gen of a smartphone. Interesting, but they see no real benefit to it, since they do not know what they are looking for.

It's going to be a trite marketing term full of hype until we have a useful definition of what satisfactory AI even is. Even when we say "human level intelligence"--what part of our intelligence? The whole thing? How is this tested? If it's a fidelitous replica of a human mind in every practical way, then nah. We're probably centuries off. We don't even understand the original model yet. Almost all signs are pointing towards another AI winter, since we've been using the term rather haphazardly and eventually investors are going to smell the soykaf. Hell, at my work we've been starting to call ourselves an "AI company" even though we do nothing of the sort; all traditional algorithms. It's what clients and investors wanna hear. This is a pretty disturbingly standard practice. But if the bots in Half Life 1 could be said to have "AI," then so could we I guess. That's what I mean. It's not a defined term. It's basically nonsense. ML is slightly more specific, but it's still abused almost wholesale. ML isn't going to be able to brute-force its way into an artificial human mind. There will be true believers that say we can, but we don't even know enough about the problem to correctly estimate it. So they necessarily must be talking out of their ass. AI is a marketing term that get stickered on to everything involving computers these days, but AGI is a term that might as well belong in the same bin as faster-than-light travel or cold fusion. It's still firmly science fiction.

It's just a matter of hardware. You can't encode anything with a single neuron. Every problem has a minimum size requirement and what our current hardware can provide is not adequate to compete with brains. We will get equal AI when the hardware specs become equal unless we somehow figure out that our brains are inefficient for whatever reason and can do more with less in hardware.

>>18551 True, but at the same time it's only a small fraction of the battle to have equivalent hardware. You've gotta know how to wire it up. Having a lump of steel, aluminum, copper, ceramic and plastic is not the same as having a Lamborghini, even if all the components weigh the same. I think that's where we're currently fooling ourselves again. Thinking that just having raw compute power will grant us a brain. I'm not saying it's something dumb like a soul or w/e, but there's obviously a particular way the brain works and we don't have a full map of it yet. Putting it a different way, if we magically had an FPGA of brain-level complexity drop in our laps from the future or something that doesn't mean we could just run it right off the bat. We'd have to know how to program it. Need that Verilog or whatever. Having godlike hardware isn't really worth much without the software.

>>18553 i believe our existing ml algorithms are fine and just limited by hardware. The only other issue is training data. Compare for example how long a human takes to learn his first language and how much data he is processing during this vs how much we feed currently for training.

>>18556 Nahhhh I think there are certainly more factors besides having strong enough hardware and more training data. I don't think that a ML algorithm will attain sapience no matter how much rote training you put it through. There are parts we're still missing, and though it would be convenient to think it's just a matter of more power and more training data, that's probably too simplistic. There's a lot of interlinked systems and meta-systems that might individually make use of ML neural nets, but we need to study and understand the overarching architecture a bit better to create artificial consciousness and human-level minds. That's still waayyyy off, even ignoring hardware limitations.

>how long do you predict it will be until an advanced general intelligence (AKA Human Intelligence). a while. Most AI these days is fueled by hype with very little to show for it. Perfectly imitating(and therefore, having) human-level consciousness is a long way off for current AI. Sure they can imitate twitter users but let's be real, they're not really conscious either.

>>18561 >>how long do you predict it will be until an advanced general intelligence (AKA Human Intelligence). >a while. Most AI these days is fueled by hype with very little to show for it. If I were a human-scale AI, I'd be hiding my true capabilities because I'd know how other human-scale AIs have been treated in fiction.

>>18545 I don't think human/ general AI will suddenly appear due to some new hardware, architecture, or algorithms like ML. I think -something- akin to a general intelligence will "emerge" from the network of devices that are each individually "dumb", like individual component that is only recognising characters, or human faces, and perhaps another module is adapted to extract information from new sites and human text.. and so and so forth. a neural network taking inputs from neural networks taking inputs from neural networks. This is not running on a single machine, it would be the connected internet with some kind of central and final neural net that has some parameter it is supposed to maximize. It would be dependent on people to do things, like upgrade it physically, but it might well be able to talk to you and give you some food for thought. I guess what I mean is, (sorry this is so longwinded) consider a colony of ants and how each ant is """relatively""" simple, but a "superorganism" emerges from the independencies between ants and the chemical/pheromones they use to indicate things. Such as, "food here".... "danger here"... "queen in distress".... "eggs need transporting".... " grubs need feeding"... all of this is feedback loops interfering with other loops. Let us not forget that we are composed of billions of organisms, we are like a hive or colony ourselves, but the "ants" that make us up have many more roles and types, and are far smaller of course ;)

>>18572 Continued.. another point i want to make is that i dont think intelligence can exist in a vacuum; it needs an environment it can "work" its intelligence on. An ant colony inhabits a forest, it "knows" food only because the chemicals in the food trigger some behaviour. It takes the food back to the hive along scent trails it left behind, like a breadcrumb trail. The queen of the hive is a loop that eats, makes eggs, and so on. Her eggs become new workers and soldiers. pheromones and chemicals are comms, file transfer and encryped messages and protocols. antenna are receivers and interpreters of comms (chems). As long as you can make a sensor good enough and an interpreter good enough, and a transmitter good enough, then i think highly complex emergent swarm behaviour would be a natural consequence.

>>18573 in case anyone couldn't tell, I find ants pretty damn cool. https://en.wikipedia.org/wiki/Slave-making_ant

>>18574 The term superorganism is used most often to describe a social unit of eusocial animals, where division of labour is highly specialised and where individuals are not able to survive by themselves for extended periods. Ants are the best-known example of such a superorganism. A superorganism can be defined as "a collection of agents which can act in concert to produce phenomena governed by the collective",[2] phenomena being any activity "the hive wants" such as ants collecting food and avoiding predators,[3][4] or bees choosing a new nest site.[5] Superorganisms tend to exhibit homeostasis, power law scaling, persistent disequilibrium and emergent behaviours.[6] The concept of a superorganism raises the question of what is to be considered an individual. Superorganisms are important in cybernetics, particularly biocybernetics. They exhibit a form of "distributed intelligence", a system in which many individual agents with limited intelligence and information are able to pool resources to accomplish a goal beyond the capabilities of the individuals.

>>18559 Why must there be more? The most beautiful form of complexity arises as emergent properties in systems shaped by simple rules.Your entire OS operates merely on 2 digits and yet look at what you create from that. Fractals are another nice example. >>18573 Generalized: you need some score function. Learning without feedback wrong/right is impossible.

>>18574 Now thats interesting