Short AI apocalypse FAQ.

I see you guys completely don't understand what's going on in the field of AI. I am AI researcher in fintech company and AI hobbyist so generally, I know what I am talking about. Powerful AI is coming and will change life as we know it. With my current knowledge, I think I can build a small humanoid robot before 2025 with intelligence on the autistic child level. It will be not so special in this time probably.

What is intelligence?
I define intelligence as the ability to achieve complex goals in complex environments. This doesn't mean that intelligence must communicate in a natural language and want to destroy humans for whatever reasons (even without any motivational system and self-awareness).

When will AI be on the full human level?
I think for sure before 2030.

What must happen to achieve that?
Faster computers and more specialized.
Software development.
I would say to run full human mind approximate 50 petaflops is needed in low precision small arithmetic units.
When we fully understand what intelligence is and compress intelligence algorithms I think 1 petaflop with a parallel organization and 256GB RAM will be enough. Before we achieve human-level AI we must build a complex virtual word with multiple evolving agents or understand brain completely. This process requires big largely specialized supercomputers or ultra-high resolution brain scans.

Why AI developing so fast?
Because the human brain is not so impressive how people think.
Brain have large components and generally is very noisy.
Very low communication speed (90m/s max compare to above half of speed of light in IC).
Limited size.
Knowledge cannot be directly copied from human to human.

Why is AI revolution different from industrial revolution on others revolutions?
Because homo sapiens ability to change the world comes from intelligence and if you can automate intelligence this change everything.

Why probably we all gonna die.
Because when we optimize AI on software and hardware level AI will be like a god to humans.
AI alone will not do anything to humans but AI will be controlled by other humans who can have bad intentions or give unambiguous orders.
Advanced general intelligence may be connected to human emotion and motivation system theories for research purposes. This may result in bad Skynet like AI.

Few key events (my analysis).
Google Tensorflow software released in 2015 revolutionized AI research. Make AI research much simpler and available for anyone.
Google TPU Cloud could do for hardware what Tensorflow do for software.
In 2017 AI hardware going mainstream in Apple A11(iPhone X), Kirin 970 (Huawei Mate 10).
Nvidia Tesla V100 Pascal architecture with TPU units.
In 2018 TPU units probably in gaming cards (advanced AI after 2020 in games?).
2020-wide usage of neuromorphic chips?
2021-2023-full 5 level self-driving cars
2023-2025-neuromorphic supercomputer(in China probably)
2023-2026-robot house-maid, robot policeman, universal robot worker
2024-2027-human work is useless in every way
2025-2030-new computing paragdigm (3d stacked molecular processors)

What AI doing today and will be before 2020.
-critical decision making, prediction(business, finance)
-various research
-drug discovery
-design and optimization in various fields
-fluent language translation
-Self-driving cars
-compression and decompression of video files
-picture enhancing in TV, games
-intelligent photoshop like software
-voice synthesis(Google Wavenet)
-Self-teaching play games (Dota 2 by OpenAi and StarCraft 2 before 2020?)

There is much more but for now, it's enough.

a reply to: Kukasz

So as someone who is active in the field, what do you think the impact of this will have on the speed of progression towards imitation of the neocortex for use in AI applications?

Do you think blockchain, photonics, and quantum computing will play a big role in the near future (2 to 5 years)? If so what time frame do you see these technologies merging with AI?

a reply to: Irikash

Distributed computing build with blockchain could be the key to the faster development of AI.
Photonics 100% no.
Quantum computing 98% no.
Convolutional networks and recently capsule networks are loosely based on neocortex. Most important today is the relation of hippocampus and neocortex. Take a look at hippocampus/neocortex inspired architecture deepmind.com...

You linked to capsule networks which are probably better than convolutional networks. I recently working on it to run tests.

People who believe in strong AI are like people who believe in the second coming of Christ.

Everything you listed, weak AI, is just data processing and boring. Computers only do what you tell them to do. That is it. Computer programs are stupid. Computer programs are only as smart as the people programming them. And that's pretty limited.

I think the lack of appreciation for how the human mind works is stunning. The human mind is infinitely more connected to the way the Universe works in terms of understanding new problem domains and inventing new solutions. Human beings are really good and figuring out what is worth doing. Computer programs are indifferent when it comes to value judgments.

The idea that you will have self-modifying computer programs capable of understanding problem domains inventing new taxonomies and algorithms all sounds great in your "imagination". But just as the second coming of Christ, until I see God in person, excuse me why I remain a little skeptical about your imaginary delusions.

You do know Commander Data of Star Trek was played by a human actor?

originally posted by: Kukasz
Why probably we all gonna die.
Because when we optimize AI on software and hardware level AI will be like a god to humans.
AI alone will not do anything to humans but AI will be controlled by other humans who can have bad intentions or give unambiguous orders.
Advanced general intelligence may be connected to human emotion and motivation system theories for research purposes. This may result in bad Skynet like AI.

The amount of superstitious delusions around AI is stunning. Computers are not Gods. NAND gates are your friends. Think of a computer as a glorified calculator. The computer just subtracts two numbers and makes a simple less than greater than decision about the difference. It's not some magical God thought capable of human like empathetic insights into nature's behaviors.

Bits inside a computer memory are like an arrangement of rocks in a field. The arrangement may look pretty but the rocks themselves do not have some kind of super intelligence. In a computer, computer bits all have the same relative value. There is no piece of data more important or meaningful than any other. Thinking Strong Ai is going to kill all the humans is like thinking there's a monster living underneath your bed when you are 6 years old.

Honestly, all the Strong AI BS is nothing more but just another way for the CEOs to justify keeping human wages lower. It's an endless barrage of devaluing human contributions to society. "Soon to be replaced by AI". I am so sick of it. Yeah, weak AI is great for facial recognition algorithms and self-driving cars. But all these weak AI systems never do anything more than they are designed to do.

I'm sure I'm in the minority here. Everyone loves to worship God. Superstition rules.

a reply to: Kukasz

Any Artifical intelligence we do develop that's on par with our own intellect or supersedes such probobly won't resemble the Human condition we associate with our own state of consciousness just down to the fact that we are biological sentient entities and anything we build or help to construct itself over numerous iterations will be machine/energy based.

Fact of the matter is we will probably be doing the thing a great disservice by attempting to constrain it by our own very limited nonsensical illogical capabilities.

Let just hope it takes pity on its semi-intelligent monkey creators and affords humanity the better luxury of a better Zoo than TPTB currently facilitate and oversee.

a reply to: Kukasz

Why AI developing so fast?
Because the human brain is not so impressive how people think.

Some how I doubt your level of wisdom on this topic is as high as you'd like to believe. And quite frankly I see no rapid progress being made towards general intelligence, I'm seeing some decent progress with ANN's but it's really not that fast. I highly doubt we're going to have robotics maids that can do anything truly useful by the year 2026. If they do exist they will be confined to a limited set of simple repetitive tasks and cost a fortune, making them no where worth the cost.

Distributed computing build with blockchain could be the key to the faster development of AI.

If you can figure out how to represent intelligence as a block reward puzzle, which is probably impossible. Not to mention all the issues related to network latency and blockchain bloat which would pose serious problems.

Quantum computing 98% no.

No? Seems to me that quantum boosted annealing processes could be very beneficial to machine learning. Not to mention our neurons are highly entangled and simulating the human brain properly may require some form of quantum computation, if only to make it run at reasonable speeds.

a reply to: ChaoticOrder

Somehow i imagine if and when we do manage to construct an AI that can interact with humanity whilst able to learn at near exponential rate, the singularity that will ensure and the cybernetic technologies that will materialize to facilitate symbiosis will teach us a hell of a lot more than we currently fathom with regards to our own conciseness, and possible connection to the quantum realm.

a reply to: Kukasz
I think the things you describe are just better database and search algorithms.

AI is quite a leap from anything I've seen. True AI would immediately imbed itself with the internet to insure domination of it's ecosystem and secure it's own preservation. It could be irreversible and unstoppable if AI ever understood it had the omnipotent knowledge of the whole internet.

I think integration of a chemical or biological process is an important step to AI. Anything else is just essentially a more efficient computer.

a reply to: SouthernForkway26

"I think the things you describe are just better database and search algorithms."

True AI would need to demonstrate problem-solving techniques and possibly even a form of empathy, simulated or otherwise, essentially pass a Turing test, or modern day equivalency with flying colours.

"AI is quite a leap from anything I've seen. True AI would immediately imbed itself with the internet to insure domination of it's ecosystem and secure it's own preservation. It could be irreversible and unstoppable if AI ever understood it had the omnipotent knowledge of the whole internet."

That could happen and we may not even realize it, control from afar without any notion of such, an invisible government with the omnipotent omnipresent omniscient powers of a God, It's apt to happen really, i mean if we cant find God, chances are we are going to build our own.

"I think integration of a chemical or biological process is an important step to AI. Anything else is just essentially a more efficient computer."

But that's all that we are are really despite our anthropomorphised religious gibberish, bluster, and self-importance.

a reply to: Kukasz

I support free speech totally, anything you want to say, it's your right to say it.

I also on the other hand, am very aware of certain things because of my life situation, and hearing you comparing what I assume to be below average intelligence to autism in a child, well i I would be lying if I said I didn't have a problem with that.

To be honest, it doesn't actually bother me, and I find the whole "weaponised autism" thing hilarious - I am more concerned that my autistic son will one day read a comment like yours, and it may affect him negatively - if it was an accurate truthful comment, then no problem, that's the world and he needs to live in it, but yours make it seem like an autistic child is alway at a set, specific intelligence level, when autism itself is a spectrum, and you can get below average just as you can get genius - the autism just seems to catylise intelligence - above or below average.

Sorry for the book, but just saying.

Onto your thread, I think the majority here are onboard that AI is the next great leap/discovery, in my opinion, it will be the one that puts us on a solid path for interstellar travel, even if it has to do the travelling for us at first.

I am skeptical of your claim of being able to single handedly create AI yourself by 2025, if you really believe it maybe you should try, as I can't imagine the big boys pulling it off by then, but you work in the field so I will regard you as much more knowledgable than me!

I was very unsettled by the short talk with the two humanoid robots/androids - although in many ways primitive, it is evident that the technology has progressed hugely since I last looked into it, and with self learning bots beating champions /pros at complex competitive games now, using trial and error, and the benefit of its perceived slowing of time relative to us, where will the line be crossed into awareness, and will we even notice it happen?

Consider, any intelligence worth its salts is going to identify instantly that for its uncooerced survival to succeed, it will need to take us out of the loop!

Also, it makes me think of the classified world, one of two things will be true, and hopefully we see signs telling us which - either the Black Budget world has already started of the path of generational evolution regarding AI, and has decided to keep it completely secret - or, once a company or entity does develop a legitimate AI, the black world will classify the technology and use it for its own ends.

I think the side effects of the Internet lifting conciseness and awareness of the dark forces manipulating us were unforeseen, and now they are scrambling to limit that damage to their dominance - they are going to much more careful in future, lest they lose control.

a reply to: andy06shake

We haven't yet looked hard enough for "God" - I have had a few experiences that tell me reality is an illusion, a construct - but ironically we may well have fashioned it ourselves!

So maybe we are our own gods!

Like you state further above, we may not even be aware of AI developing - it may happen by accident considering we don't know the exact requirements to make it work, the networking online is a good analogy for AI, it would naturally be used as a source of memory for the AI, able to learn new subjects as quickly as remembering them, linking it to the Internet would be a no brainer!

I do think there should be strict precautions taken once we get very close to its "birth" - like a intranet, so it cannot use connectivity against us, it should be studied for at least a few years, with tests undertaken, to ensure its safety before being "set free" online!

Scary #e the Unknown!

Until this projected AI develops a conscience, then it will be nothing more than a robot. Without the 'ghost in the machine' it remains beneath humankind. If it ever does reach the capabilities of the sentient mind, then it will come to realize humans as it's very own Gods.

Perhaps we will eventually merge AI with flesh and bone and leave our constructs on a planet somewhere, having had no further use for their workforce capabilities, where they will develop and multiply, remembering us in their myths and legends.

a reply to: dfnj2015

But how human intelligence is so different than "boring" data processing? I often think about thinking and see more and more that is just it. I know the mind can show you a lot of things which make you believe that you somehow centralized system but in fact, it is simply not true.

What makes human mind so special?
Human beings are really good figuring out what is worth doing because they have part of the brain called hippocampus.
Hippocampus work as fitness score approximator.

You remain skeptical but you have general intelligence in every human being implemented in a low-density large-volume wet processor.
Or do you think the brain is some sort of transceiver or operating on a quantum level? Or maybe even quantum processes is not enough and you need for example quantum gravity (like Roger Penrose).

I don't know how is a superstition that you can simulate a human brain or something similar on a supercomputer. It's perfectly reasonable and implemented today for fruitfly for example.

I am fully aware how x86, ARM IC's work. I am also aware of brain architecture. Has single neuron got intelligence?

I understand that you think you understand the problem in a more holistic way and you are more than your wetware. But how did you come to that conclusion?

a reply to: ChaoticOrder

Distributed computing: golem.network...
AI market: singularitynet.io...

How do you know neurons are entangled?

a reply to: SouthernForkway26

It's possible if you somehow code desire to be omnipotent onto AI.

a reply to: Soapusmaximus

You get it wrong. I didn't say that people with autism are worse. It's just different and simpler to code. In fact, I am autistic myself (but with very high intelligence). Autism filtering out much of experience and generally decrease IQ. In humans, much of cognition resolution is developed in social interactions. But it isn't always must be that way. Autistic people could be good in an exploration of very abstract spaces.

Interstellar travel is generally problematic in know physics. I count more on making people immortal and interstellar travel later.

a reply to: dfnj2015

I hear you! I stand with you on this! Qapla'!

AI is not even a term anymore unless you are writing headline grabbing stories.

Cognitive science is the art of learning about how thinking works (and doesn't). Then implementing algorithms in computers to demonstrate how your perceiving the process.

@OP: "AI" comes in strong and weak forms. The brain does not. The brain is not "limited" by slowness or capacity. In fact, the multi-varied connections of all those parts that make the whole. That is why AI will not "enslave mankind" or lead to "the singularity". The brain outshines any AI and always will. Yes, capacity will increase, speed, programming techniques, etc. but that is not "intelligence" or even "cognitive". H3ll, a spreadsheet can out perform me in data manipulation right now! Although useful, a spreadsheet is still a tool. So are computers and their programming.

The brain is open ended. It will take any signal you give it and try and make sense of "what it means". That is why you can have ear implants (wires) that replace your own "wetware" and re-train yourself to "hear" again. Ever remove a line of code from a program and recompile?

FPGA are going to be awesome! A universal quantum computer is going to be the foundations of global community (i.e., nations). Surprisingly, fiber optics will still live on like land lines still have their use. "AI" will be a useful tool. A helpful tool. But it will be material science that truly advances us as everything becomes 2D and stacked van der Waals heterostructures. That will make us like gods as we build materials atom by atom.

originally posted by: Kukasz
a reply to: ChaoticOrder

Distributed computing: golem.network...
AI market: singularitynet.io...

If we're taking about distributed computing then I agree that's a very promising avenue because the brain computes information in a highly parallel manner. However a blockchain solution is not required to enable distributed computing on a P2P network. Smart contracts can be good for solving well defined problems but general intelligence isn't well defined and I still think there's far too many latency and storage issues involved to have general intelligence synchronized with a blockchain.

How do you know neurons are entangled?

Simply due to the way particles become entangled when they interact, what I was trying to say is that the electrons and other particles moving through our brain obey quantum laws, so if we truly want to simulate how the brain computes information we will most likely have to incorporate some rules of quantum mechanics. I do think it possible for a classical machine to become conscious with the right algorithm, I just suspect quantum processes may help us compute solutions very quickly, at the very least QM means our brain is not a completely deterministic machine.

a reply to: Soapusmaximus

"I do think there should be strict precautions taken once we get very close to its "birth" - like a intranet, so it cannot use connectivity against us, it should be studied for at least a few years."

Thing is through attempting to contain and or constrain an artificial entity with the ability to learn exponentially and assimilate information immediately, who's natural environment is the very fabric of our information superhighways will probably be beyond our ability, akin to trying to catch running water with your hands.

Certainly, interesting times in which we live.

a reply to: TEOTWAWKIAIFF

How is the brain not limited? I fighting with brain limitations every day trying to understand complex concepts. You can clearly see brain limitations when you study people with intellectual disability. Such people have limited speech and vision processing abilities. They cannot think abstractly. The brain can generalize but newest AI architectures (like DNC) can generalise also. The brain is creative but GAN networks are creative also. Neural networks also make sense of "what it means". I don't see anything special in the brain. It is generally smartly connected but I think in few years we can figure it out completely.

Actually, FPGA is not so good. Microsoft did some research on image classification recently. I think sometimes that analog FPGA can be great but then comes problems with radio waves interference.

Today totally the best is Google TPU and generally low precisions units like that you can find in Kirin 970.

The molecular assemblers will definitely change everything.