https://en.wikipedia.org/wiki/Understand_(story)

what about something like this

(http://www.autoadmit.com/thread.php?thread_id=4149157&forum_id=2#37351721)

Most AI these days is glorified curve fitting

(http://www.autoadmit.com/thread.php?thread_id=4149157&forum_id=2#37351723)

It is as dumb to predict 10 years as it is 50 years or 100 years. There is hardly any basis other than some generic "Moore's law" that's not even valid anymore.

"We don't know". But this is apparently not an acceptable answer.

(http://www.autoadmit.com/thread.php?thread_id=4149157&forum_id=2#37354442)

Davidson’s a nice derangement of epitaphs might be relevant. Ai might ultimately lack true spontaneity. But we might be able to engineer spontaneity so well we can’t tell the difference.

You might be right that this makes ai asymptotic against true generality though

(http://www.autoadmit.com/thread.php?thread_id=4149157&forum_id=2#37352006)

https://philpapers.org/surveys/results.pl

You, stroking your neckbeard: "I'm a non-physicalist because I'm not an idiot"

(http://www.autoadmit.com/thread.php?thread_id=4149157&forum_id=2#37351926)

At a certain point, we will be able to map neural activity at a very fine level of detail. If you can find out what neurons are firing to cause people to talk about "subjective experience" the problem is basically solved. The hard problem isn't real unless you believe something magical occurs in the brain. Purely computational approaches like attention schema theory can account for consciousness and people's confusions when thinking about this topic.

(http://www.autoadmit.com/thread.php?thread_id=4149157&forum_id=2#37352458)

Attach 100 Einsteins (we'll have them thanks to genetic engineering) to a neural net connected to a supercomputer, and they''ll come up with an AI soon enough.

(http://www.autoadmit.com/thread.php?thread_id=4149157&forum_id=2#37351952)

Bottom line is I think we're about 25 years away from technology turning society into something the likes of which we've never seen before and into something fundamentally different.

(http://www.autoadmit.com/thread.php?thread_id=4149157&forum_id=2#37358437)

Development of high-resolution cellular brain-mapping technology is maybe a decade away. Developers already recognize that the era of hand-tailored scripting alone is drawing to a close.

(http://www.autoadmit.com/thread.php?thread_id=4149157&forum_id=2#37353868)

Eventually, the world economy will look like the Antebellum South, with robots instead of slavery; a mega-rich capitalist ownership caste; and religious impulse driving charity to the proles.

(http://www.autoadmit.com/thread.php?thread_id=4149157&forum_id=2#37353954)

The best example I've seen of AI bypassing the brain's more circuitous route was in Google Deepmind's video game play. In its first attempt at various classic games the AI looked like a retard (much worse than a human on his first try). But the AI "brute-forced" its way to becoming the best gamer ever, often employing techniques and strategies not seen before. Even if the human brain has the "intuition" to initially figure out the controls and learn and adapt, the AI brain could play the game 10 billion times in its "head" to figure out the optimal strategy.

(http://www.autoadmit.com/thread.php?thread_id=4149157&forum_id=2#37354601)

It was not an easy task to create AI that becomes a video game champion when the AI does not have knowledge of the code in the game and isn't exploiting it. It had no unfair advantage. It was given the task of maximizing score and only "told" what each function in the game does (i.e. this makes the player go right, this makes the player shoot a gun). Like I said, the AI actually sucked ass when it initially played the games (much worse than your typical human) but eventually bested all humans.

Our brains are computers. They are weird and oddly "engineered" computers that have some "mystical" components to them that current AI is incapable of, but in other ways they are actually pretty shitty relative to current AI. Far too often people use the brain and our thought processes as the benchmark for determining whether AI is inferior. I think that's pointless.

(http://www.autoadmit.com/thread.php?thread_id=4149157&forum_id=2#37356182)

They'll need some serious breakthroughs in AI/deep learning to make any progress.

(http://www.autoadmit.com/thread.php?thread_id=4149157&forum_id=2#37359029)

The full Starcraft 2 is just a different story. They don't even know where to start.

(http://www.autoadmit.com/thread.php?thread_id=4149157&forum_id=2#37364166)

**picks up wooden doorstop**

"As you can clearly see, this Intelligent Inclined Plane Machine is the *real genius* here; not humans! It has simply chosen to focus on becoming excellent at the narrow task of keeping doors open, while the wasteful human brain-- [(cough, cough, chortle)]-- runs in circles spending energy on things like consciousness, leisure, meaning, and all those other inefficiencies you call 'general intelligence.'"

(http://www.autoadmit.com/thread.php?thread_id=4149157&forum_id=2#37356154)

also, AI hasn't been scripting and pathfinding for a long time. the entire field is about statistical learning these days. of course, you are right that full human ability and behavior may be centuries away, but generally you have no idea what you're talking about.

(http://www.autoadmit.com/thread.php?thread_id=4149157&forum_id=2#37354622)

AI is a marketing term to build hype / sell products for the foreseeable future.

"Machine Learning" is just applied statistics.

Any talk of AGI is insanely premature - advances in ML applications gives us e.g. better facial recognition software, Alexa-type devices that work better, make self-driving cars more feasible due to higher image classification accuracy.

But these continued improvements in the accuracy of algorithms, e.g. "deep learning" hype - just improves devices and tech like described above - it doesn't fundamentally bring us close to AGI.

It's not like we have the structure of AGI mapped out and it's like oh damn - if only we could squeak out a 10% improvement in classification accuracy.... THEN this AGI would be working. No - the entire structure / mapping of AGI is insanely complex and requires large breakthroughs in several fields.

All the hype as of lately has been due to advances with several specific variations of neural networks which are specifically well suited to high dimensional ML problems with complex feature interactions (and outside of these specific applications they actually perform very poorly relative to "traditional" models or just say random forests or boosting methods / ensembles) - and these advances got tons of hype because they're useful for image classification problems which self-driving cars, facial recognition etc. all rely upon.

But the difference is we knew very well how to structure these problems then when the accuracy of the underlying ML algorithms improved enough these problems (e.g. self-driving cars) were feasible to put into application.

By contrast there's no such structure of AGI that I'm aware of that we're just waiting around for some improvements to take place which will allow us to solve AGI.

The problem of AGI is far more fundamental - it's not a matter of algorithm performance, it's that we don't even know what to do.

If someone more on the AI side (I'm on the statistics / ML side) wants to tell me I'm mistaken I'm more than willing to listen......

edit: In terms of automation / job replacement - you don't need anything remotely close to AGI / AI, those terms are pure marketing terms in the corporate world. To automate jobs - say a chatbot that helps you with customer service - you simply need sufficient accuracy of the algorithms. All these products that can replace people's jobs are remarkably simple in structure and are just using applied stats / NLP bundled together in some software product being sold as "AI". Automation of jobs will continue to happen even when true AI stays far beyond the horizon - because to automate many jobs you just need a few algorithms and a bit of engineering. This is counter-intuitively actually a sign that we're far away from true AI - because companies are building highly specialized domain-specific products to automate processes for companies. Any guess on when true AI becomes possible is worthless IMO.

(http://www.autoadmit.com/thread.php?thread_id=4149157&forum_id=2#37359074)

Some people, including myself, think that there is no free will. We also believe that what we call human consciousness is really a processing hallucination that is generated epiphenomenally once neurons hit a certain level of complexity in broadly adaptive skill sets.

For these people, they (and I) believe that human-level or "true" AI will inevitably occur. At some point in the near future we will hit on nonbiological complex neural development (probably through a combination of whole brain emulation and planned broad learning, TBH). When we do so, what we consider "consciousness" of some type will appear to manifest in that AI. And that will be an epiphenomenal hallucination on the part of the poor little AI, in much the same way as we suffer out own hallucination of consciousness, and of agency in others.

In the end, who knows, though. We could all be wrong. And this viewpoint seems to defy common sense. But it is basically where the neuroscience appears to take us at present.

(http://www.autoadmit.com/thread.php?thread_id=4149157&forum_id=2#37360199)

Can you explain this more? What exactly is the "individual" having this hallucination?

(http://www.autoadmit.com/thread.php?thread_id=4149157&forum_id=2#37360231)

This gentleman has some interesting work on the subject of consciousness as hallucination. https://aeon.co/essays/the-hard-problem-of-consciousness-is-a-distraction-from-the-real-one

Here is an interesting video on point, nice and simple:

https://www.youtube.com/watch?v=lyu7v7nWzfo

(http://www.autoadmit.com/thread.php?thread_id=4149157&forum_id=2#37360288)