Artificial Intelligence

I agree, it's a refreshing perspective from the all-too common "super intelligent AI must be terrifying monster-gods" that I see when the topic comes up.

That was deeply fascinating, and an ecxellent example of how to give extremely detailed critical feedback in a thoughtful and professional manner.

@Imca: If we're talking about a zombie that is in all physical respects identical to a non-zombie save for the lack of consciousness, I agree that's more or less worthless. However, in my opinion, the question of how much we can accurately equate consciousness and intelligence remains open, and, in fact, a large body of neuroscience research supporting the notion that, while some degree of intelligence is essential to consciousness, many aspects of intelligence do not in turn require consciousness, which only enters the picture after the decision making process has already concluded in some cases, and in many others, not at all.

edited 9th Feb '18 7:56:53 PM by CaptainCapsase

It really begs the question of what self-awareness is for.

A lot of angst and navel-gazing over things like "What is my purpose in life?".

Its not for any thing, its an accidental byproduct.

Kinda doubt it. Our bodies invest a lot of caloric resources into conscious cognitive processes for something that produces no direct benefit of its own. No theory has been proposed to explain how a sense of self awareness could arise as a byproduct of anything else, including complexity. An accidental byproduct should show variation-that is, if it isnt strictly necessary itself, some people should be able to function perfectly well with less of it, or even occassionally none at all. Yet that seems not to happen. A loss of consciousness, other than sleep, is called a coma and is considered a serious medical condition.

I think self awareness is our brains method of coping with its own complexity (as opposed to arising directly from it). It provides a short cut for problem solving and decision making by creating a self image that is a kind of simplified version of the whole mind (or at least that part of it residing in the cerebral cortex). By working on ones self image, which in turn influences but does not directly control the rest of the mind, the brain can learn and adapt to external changes in the environment more quickly and easily. Without it, we would overwhelm ourselves. My two cents anyway.

edited 11th Feb '18 7:09:08 AM by DeMarquis

My thought is that it evolved as a kind of holistic thermostat that allows us to react and adapt to changes in the environment with more flexibility by sublimating our other forms of internal management. Once you reach a certain level of complexity, a top-down, holistic sense and recognition of self probably allows for greater self-preservation in the long run than non-hierarchal juggling of the various parameters responsible for keeping one alive. From there also evolves the ability to perform altruistic actions on other individuals judged by the self-sense program as advantageous to preserve for whatever reasons.

The problem is evidence that conscious thought does not control other cognitive processes, for example the famous experiment documenting that your arm receives neural impulses to move before you are consciously aware of the decision to begin moving. It's as if some sub-conscious process actually made the decision for you, and then you experienced yourself making the decision.

@Imca: Conscious (or at least behavioral correlates of consciousness) appears to have independently evolved at least twice, in vertebrates and in cephalopods. That definitely suggests there was an evolutionary advantage to consciousness, and from what I can tell, there's not really a very firm consensus in neuroscience or evolutionary biology about this topic; the perspective being offered by computer science is certainly valuable, but is not the final authoritative answer on this topic.

I assumed that we were talking the level of conceptual self-awareness that is present in humans. If we mean simple "self as independent object" levels of self-awareness, then that has obvious reproductive implications in any social species.

edited 12th Feb '18 5:54:58 PM by DeMarquis

x3 Again I wouldn't take that as evidence that the conscious isn't involved because I can't think of any model of consciousness that would give a remarkably different result.

There are a lot of neurological processes that don't involve the consciousness at all though. Moreover, there's instances of people with damage to the CNS who have lost conscious control of parts of their body, the most well known example of this occurring in hands in Alien Hand Syndrome. In such cases there are known instances of the "Alien Hand" performing complex motor functions autonomously. In fact there are some forms of aphasia in which speech and language can become decoupled in part from the conscious mind.

Things like this strongly suggest that consciousness isn't an integral part of these individual processes, but rather serves as a feedback mechanism which in some way allows the whole system to work together. I'm not convinced AI reearch is particularly close to replicating that function of the brain, nor does it seem to be necessary or even desirable to do so to replicate many aspects of intelligence.

edited 13th Feb '18 4:24:58 AM by CaptainCapsase

I imagine the main purpose of consciousness is actually to give the brain a centralized command-which is necessary for a lot of operations in your mind.

Good example actually would be alien hand syndrome-it gets in the way of doing a lot of things you need to do. Your body needs structure so that your left hand and right hand don't get in each other's ways.

Except that there is no real evidence that the consciousness serves that function, and some evidence that it doesnt-incl the alien hand syndrome (because the people suffering from it are fully self conscious).

This new game that's coming out, Detroit Become Human, is making me paranoid about the future of artificial intelligence. Not the part where they revolt, but the one where they replace our jobs. I mean, if they can be nurses/teachers, what's stopping them from progressing to other things?

And I can't or maybe don't want to imagine a future where androids make up almost the entirety of the workforce, while humans are just watchers and enjoy all the freetime.

Automation of jobs has been going on for 300 years, to the point that there are entire economic theories built around it. Traditionally, the way automation fits into economic growth and continued employment is well understood: employers automate old jobs to save money, then take that savings and invest it into research and development, which results in new goods and services that can be offered to the public, which in turn creates new jobs whose skill set is relatively new, and therefore can't easily be automated. Wait until those new jobs become old jobs, then repeat the cycle.

The problem now is that employers are not acting that way. They have been sitting on record profits since the aftermath of the 2008 recession, but they won't spend the money on R&D, or expanding production. Instead the tend to invest their savings in financial securities. There is more profit to be made buying each others' stocks and bonds than in selling new stuff.

Thus, the real problem these days isn't artificial intelligence, which is really nothing more (at the shop floor level at least) than the latest iteration of textile carding technology, but financial speculation as a percent of gross national product.

edited 22nd May '18 2:55:05 PM by DeMarquis

Well automation is doing it now.

BTW, I just watched the teaser trailer, and that's a relatively accurate depiction of downtown Detroit, without all the futuristic technology, of course.

Besides even if robots take over all the jobs, and create a post scarcity utopia, which as stated isn't the most likely outcome its not like you can't work.

People are fickle creatures, and there will still be a market for hand crafted artisan goods, on the simple fact that there artisan.

It becomes a matter of "do you want to" rather then "you have too" and the former is always better.

I found this fascinating, sharing it from where I saw on Reddit: an AI that randomly generates human faces with an incredible degree of realism. You can get some pretty good-looking results, but occasionally a nightmarish ghoul.

The hair is esp amazing, but the eyes are sometimes a little off...

That is pretty good. Especially for at a glance. It is still producing some weird artifacts in a number of the examples though. Like some sort of blur or blending effect.

Ah, here it is. I was looking for an existing topic on artificial intelligence and had to go back quite a few pages. As the rules say, better to bump a topic than make a new one in OTC.

Tesla had an "AI Day" presentation yesterday discussing the latest developments in their artificial intelligence technology. Teslarati has a live blog if you want to read a summary rather than watching a three-hour video. The primary purpose of the event was recruiting, but some fairly amazing things were shown off.

Aside from the big thing that I'll get to in a moment, the intended takeaway seems to be that, in solving self-driving using machine vision, Tesla is "solving the real world". In other words, it is building a neural net that can use vision to build a complete 3D map of the world around it, integrate that map over time, and make behavioral predictions — not just in physics, but in conceptual terms.

We aren't just talking about, "That object is a pedestrian walking west at 3 mph," but, "The pedestrian was walking towards the street when it was occluded by a car. I predict that it might emerge into my field of view at this time in this place, and there's a chance I may have to avoid hitting it." The neural nets will memorize parts of their surroundings in both space and time so that they can keep track of what's going on, something fundamentally necessary to solving driving.

I've previously talked about, in the Self-Driving Cars topic, the Dojo supercomputer that Tesla is building to train its NN model. We got to see the actual chips and the "training unit" that they are integrated into. Each of these tiles contains 25 processors, has a total of 9 petaflops of computing power, and 36 terabytes/sec of "off-tile" bandwidth. These are integrated into stacks to create the entire system, with a single "server" operating at one exaflop. Tesla showed off working hardware at the presentation.

On to that big thing: if this system can create a complete solution to understanding the human world, then why not move beyond cars? So Tesla is taking the next step: robotics. Elon Musk announced the development of a humanoid robot using machine vision that will be capable of understanding natural language orders and performing general labor tasks. To reassure everyone, it will not have human-level intelligence. He envisions this as way to replace humans doing dangerous, repetitive, and boring jobs and says that a prototype will be available in 2022.

Something Musk said during the presentation resonated. When asked "why are you doing this", he answered that someone is going to and it might as well be Tesla because then he can guarantee that it will be safe around humans. Essentially, he wants to get ahead of the curve and make sure the technology is used responsibly and ethically. For example, these robots will be built to a specification wherein a typical human can "overpower" and "outrun" them.

When asked why a humanoid formfactor, he said that our world is designed for humanoids with natural vision, so our robots should be as well. This is an interesting take. It's the clearest answer I've ever seen to the question. A lot of people question the value of humanoid robots when more specialized ones would be better a specific tasks, but what if you want a robot that can be used for any task?

Edited by Fighteer on Aug 20th 2021 at 8:18:05 AM

"Elon Musk announced the development of a humanoid robot using machine vision that will be capable of understanding natural language orders and performing general labor tasks."

That depends on what he means, of course. He surely means that he can envision a computer that can understand a "naturalistic" sounding set of commands when the human operator is willing to communicate within the computer's limitations. That's pretty reasonable, and a natural extension of what we have already been building. But if he means a computer than can understand the latest in-group buzz words, nuanced emotional cues and is robust against deliberate attempts to verbally hack it, that's a long way off.

"A lot of people question the value of humanoid robots when more specialized ones would be better a specific tasks, but what if you want a robot that can be used for any task?"

I guess the question would be why you would want to spend the extra money on a machine that could do "any task"? Who needs the capacity to do any task at all? I do not envision myself requiring anything like that in my life, nor any employer either. Maybe NASA as a substitute for human explorers, but that's a pretty niche market (admittedly, one that Musk already services).