Nvidia Launches AI Computer To Give Autonomous Robots Better Brains (theverge.com)

An anonymous reader quotes a report from The Verge: At Computex 2018, Nvidia unveiled two new products: Nvidia Isaac, a new developer platform, and the Jetson Xavier, an AI computer, both built to power autonomous robots. Nvidia CEO Jensen Huang said Isaac and Jetson Xavier were designed to capture the next stage of AI innovation as it moves from software running in the cloud to robots that navigate the real world. The Isaac platform is a set of software tools that will make it simpler for companies to develop and train robots. It includes a collection of APIs to connect to 3D cameras and sensors; a library of AI accelerators to keep algorithms running smoothly and without lag; and a new simulation environment, Isaac Sim, for training and testing bots in a virtual space. Doing so is quicker and safer than IRL testing, but it can't match the complexity of the real world.

But the heart of the Isaac platform is Nvidia's new Jetson Xavier computer, an incredibly compact piece of hardware that's comprised of a number of processing components. These include a Volta Tensor Core GPU, an eight-core ARM64 CPU, two NVDLA deep learning accelerators, and processors for static images and video. In total, Jetson Xavier contains more than 9 billion transistors and delivers over 30 TOPS (trillion operations per second) of compute. And it consumes just 30 watts of power, which is half of the electricity used by the average light bulb. The cost of one Jetson Xavier (along with access to the Isaac platform) is $1,299, and Huang claims the computer provides the same processing power as a $10,000 workstation "AI, in combination with sensors and actuators, will be the brain of a new generation of autonomous machines," said Huang. "Someday, there will be billions of intelligent machines in manufacturing, home delivery, warehouse logistics and much more."

Comments

The Moon Is A Harsh Mistress/"Mike"

[Rick Schumann]

I swear to you all.. it's like everyone read TMIAHM, and actually believes that if you hook together enough hardware, it'll magically become a sentient self-aware synthetic mind like Mike in the book. Sorry, doesn't work that way, and until we solve the riddle of sentience, all these 'deep learning algorithms' will always fall short of whatever expectations you might have, no matter how much hardware you throw at it.

Re:The Moon Is A Harsh Mistress/"Mike"

[mark-t]

Nature managed to do it, supposedly without understanding anything about what it was doing.

Re:The Moon Is A Harsh Mistress/"Mike"

[Anonymous Coward]

Here's the thing:

Intelligence != Sentience

Re:The Moon Is A Harsh Mistress/"Mike"

[Rick Schumann]

Nature managed to do it IN A FEW MILLION YEARS OF EVOLUTION

Fixed that for you. Also these machines don't evolve, we build them. They don't reproduce, there's no natural selection, just us idiot arrogant humans, thinking we can do an end-run around all of it.

Re:The Moon Is A Harsh Mistress/"Mike"

[Rick Schumann]

So-called 'intelligence' without 'sentience' is worse than nothing.

Re:The Moon Is A Harsh Mistress/"Mike"

[taustin]

"Deep learning algorithm" actually means "statistical analysis applied in a slightly fuzzy way." And not really much more.

Re:The Moon Is A Harsh Mistress/"Mike"

[mark-t]

Well, natural evolution was undirected... I don't think it's a far reach that an already existing intelligence directing some new inteliigence's course may be able to expedite that duration by many orders of magnitude.

Re:The Moon Is A Harsh Mistress/"Mike"

[mark-t]

Machines evolve.... that they don't do so by entirely natural processes on their own is immaterial.

Just look at the past 50 years of computing... you can't tell me that machines don't evolve. The fact that we are the ones evolving them is immaterial.

My point is that I expect that an intelligence-directed course of evolution should be much faster than an undirected one... so I don't think there's any reason that we won't see AI happen some day, and it might not be even very long form now. Honestly, I'd say that I'd be surprised if we still don't have it by the end of this century except that I'll probably be dead by then unless I live to be about 140.

Re:The Moon Is A Harsh Mistress/"Mike"

[Rick Schumann]

We can't 'direct' something that we don't understand. We haven't got a clue how the phenomenon of 'consciousness' 'self awareness' or 'sentience' actually works in our own brains; the current approach in my opinion is fatally flawed and will not ever produce these phenomenon; it's like a million monkeys mashing keys on a million typewriters for a million years, expecting to get a duplication of Shakespeares' works. If and when we have the instrumentation to understand how our own brains produce these phenomena, then maybe we'll be able to build machines that can do that, too, but so far as I'm concerned not until then. It's wishful thinking at best otherwise, and delusional at worst, and falling for marketing hype the whole way.

Re:The Moon Is A Harsh Mistress/"Mike"

[mark-t]

We can't 'direct' something that we don't understand

Well, it's only true to say that we don't understand it fully. That's not the same thing as saying we don't understand it at all. And it's certainly not the same thing as saying that we are necessarily far away from understanding it sufficiently to artificially replicate it.

I mean, starting from the time that man started building machines, we managed to create flying machines in a barely negligible *fraction* of the time that it took nature to pull it off, for instance.

Again, given that nature managed to create something intelligent within a a billion or so years without *ANY* intelligence applied to the process whatsoever, I'd suggest that's not a far stretch to suggest that we might be able to outpace it in that regard as well.

Re:The Moon Is A Harsh Mistress/"Mike"

[ShanghaiBill]

an already existing intelligence directing some new inteliigence's course may be able to expedite that duration by many orders of magnitude.

Indeed. Moore's Law is driving increases in complexity at a rate about TEN MILLION TIMES faster than Darwinian evolution.

Evolution is undirected, converges on local minima, and does a lot of overfitting to narrow niches, and retrograde movement. There is little evidence that greater intelligence leads to more procreation, which is the only thing evolution cares about.

"Intelligent Design" can do way better. Unless you believe in some sort of "vital life essence", it is hard to argue that computers won't achieve sentience in the next century or so.

Re:The Moon Is A Harsh Mistress/"Mike"

[jma05]

> all these 'deep learning algorithms' will always fall short of whatever expectations you might have

Actually, they have exceeded expectations so far,

None of the actual researchers who work on these worry about these going sentient at this level, only the Gates and Musk types.

Actual researchers simply see these as slightly deeper statistical learning systems and are just concerned about building models with better classification accuracy. That itself is tremendously powerful TODAY to keep working on them.

Re:The Moon Is A Harsh Mistress/"Mike"

[angel'o'sphere]

Just look at the past 50 years of computing... you can't tell me that machines don't evolve.
No, they don't evolve. We get better skilled in making them.
And as long as not a human is building a sentient machine, machines never will be sentient.

Re:The Moon Is A Harsh Mistress/"Mike"

[Anonymous Coward]

"Artificial Intelligence" is not the same thing as "consciousness, self-awareness, or sentience."

You are fixated on a definition that is wrong, and white-knuckling an enormous straw-man fallacy.

Yes, we don't have synthetic consciousness, and perhaps never will. But we already have artificial intelligence, and it is getting better every day.

Re:The Moon Is A Harsh Mistress/"Mike"

[mark-t]

I said that they don't evolve on their own, but that's not the same thing as not evolving at all. Or do you think that the term is invalid outside of any context of reproductive evolution?

Re:The Moon Is A Harsh Mistress/"Mike"

[AHuxley]

Just doing fast sorting on costs of data is not good evolve people are expecting. Thats just more hardware and software funding providing faster results from larger existing data sets.

Re:The Moon Is A Harsh Mistress/"Mike"

[AHuxley]

The only understanding needed is how to get more gov/mil funding for AI this decade.

Re:The Moon Is A Harsh Mistress/"Mike"

[religionofpeas]

the current approach in my opinion is fatally flawed

And you have a better method ?

Re:The Moon Is A Harsh Mistress/"Mike"

[religionofpeas]

Humans can do magically a lot more than apes, and pretty much all that happened is that our brains got bigger.

Re:The Moon Is A Harsh Mistress/"Mike"

[Anonymous Coward]

Kind of leaving out that there were also *trillions* of experimenters in that there process aren't you? You have a rather large multiplier you're ignoring.

Re:The Moon Is A Harsh Mistress/"Mike"

[chthon]

Whales got the biggest brains...

Re:The Moon Is A Harsh Mistress/"Mike"

[AHuxley]

Every decade the US mil and gov get sold on AI.
Someone has to pay for a really fast GPU. The experts to then say it an AI project as to the funding
Remember the AI winter AC? https://en.wikipedia.org/wiki/...

Re:The Moon Is A Harsh Mistress/"Mike"

[Iamthecheese]

Not quite. It takes a known and finite amount of computronium to use CAD to emulate the electro-chemical processes in the human brain. When the structure of the human brain is known -- even if it's not understood -- a computer with that amount of power will be able to emulate human intelligence. That's the high end. There are much more efficient methods, but at that point a computer will definitively have the intelligence of a human.

Re:The Moon Is A Harsh Mistress/"Mike"

[jimmydevice]

Trillions in a gram of mud.

Re:The Moon Is A Harsh Mistress/"Mike"

[zifn4b]

Nature managed to do it, supposedly without understanding anything about what it was doing.

The complexity of the universe is far more complex than your brain. Unfortunately with the Dunning Kruger effect running afoul (we should teach that to the machines too) it's hard for you to comprehend.

Re:The Moon Is A Harsh Mistress/"Mike"

[zifn4b]

Well, natural evolution was undirected... I don't think it's a far reach that an already existing intelligence directing some new inteliigence's course may be able to expedite that duration by many orders of magnitude.

Ugh, intelligent design again? Whether something is guided by a entity we can categorize at intelligent or un-directed is not really that important. The people who think that is important are worried about the superstitious consequences of death and the afterlife. All of the processes the led to the state of things today are indeed mathematical and based on a complex intertwined series of cause and effect. The complexity of that is very large, much larger than the human brain It still amazes me that you and I are made from the dust of the stars and yet here we are having this conversation pondering how that could be so.

Re:The Moon Is A Harsh Mistress/"Mike"

[angel'o'sphere]

Evolve implies it is evolving on its own. Well, at least in German, might be different in English (we use the same world).

Re:The Moon Is A Harsh Mistress/"Mike"

[Rick Schumann]

Flight is easy by comparison. All you need to observe how flight works, really, is just your eyes. A camera would help. A movie camera would help even more. A wind tunnel helps with experiments, as would a number of other purely physical measuring devices. Do you see where I'm going with this? Flight is a purely physical thing that happens on a macroscopic scale. How a human brain works is on a microscopic scale. Furthermore you can't dismantle a brain and expect to get anything useful from it so far as how it functions when it's alive, it's too dynamic and complex for that, and we do not have sufficient instrumentation to sufficiently observe the incredibly complex system that is the human brain while it's in operation. Functional MRI is stone knives and bearskins compared to the task of mapping all the functions, in real time, of a human brain. Seriously, if we could do it, we'd have done it already, and the world of I, Robot would be a reality. Clearly it's not. Maybe someday, but that day is not today or anytime soon.

Re:The Moon Is A Harsh Mistress/"Mike"

[Rick Schumann]

Congratulations, you've got the message: Marketing hype drives all of it. There's already so much invested in something the likely mistakenly thought was going to be another simple development cycle, and like the bridge in Zork, they get 99.999% of the way to the other side, only to find that they can't seem to get to 100%.

Re:The Moon Is A Harsh Mistress/"Mike"

[Rick Schumann]

What I'm 'fixated' on, if I actually am (which I'm not) is that the whole approach to so-called 'AI' is completely wrong and a dead end. Furthermore for some of the tasks they want to use it for (SDC's I'm looking at YOU), which need to operate in a world purpose-built for humans, not machines, it NEEDS to UNDERSTAND US, not be the vague half-assed 'deep learning' crap they keep trotting out. It's not cutting it now and I do not believe it will cut it, EVER. Disagree with me all you want, argue with me all you want, insult me all you want, it will not change my opinion.

Re:The Moon Is A Harsh Mistress/"Mike"

[Rick Schumann]

You don't have to have all the answers to see that something just isn't going to work.

Re:The Moon Is A Harsh Mistress/"Mike"

[JoeDuncan]

We haven't got a clue how the phenomenon of 'consciousness' 'self awareness' or 'sentience' actually works in our own brains;

Well, this is just utter bullshit. We have plenty of clues to how it works, and some pretty good ideas on how to recreate it and what properties it must have.

This is the kind of unmitigated nonsense that philosophers usually spill when they are completely ignorant of the current state of scientific understanding.

Lastly, you obviously don't know what "sentience" means: it is simply the ability to have sensations, and we understand that quite well.

Re:The Moon Is A Harsh Mistress/"Mike"

[JoeDuncan]

Ding ding ding^

Re:The Moon Is A Harsh Mistress/"Mike"

[JoeDuncan]

... the whole approach to so-called 'AI' is completely wrong and a dead end.

LOL - only if you don't understand the difference between Artificial General Intelligence (AGI) and Artificial Intelligence (AI), which you've made painfully clear that you are totally ignorant of.

Re:The Moon Is A Harsh Mistress/"Mike"

[Rick Schumann]

Oh fuck off. You don't work in any such field yourself, you're not an AI researcher, you're not a neuroscientist, so you don't know a goddamned thing yourself.

Re:The Moon Is A Harsh Mistress/"Mike"

[JoeDuncan]

Oh fuck off.

Ad Hominem, so clearly you're an idiot

You don't work in any such field yourself,

I do

you're not an AI researcher,

I am

you're not a neuroscientist,

I earned an MSc at the Center for Neuroscience at Queen's University

... so you don't know a goddamned thing yourself.

Wow, you make so many unwarranted assumptions you'd be hard pressed to argue your way out of a wet paper bag. You don't know shit about me and everything you assumed you got wrong. All you've done here is prove how much of an idiot you are.

Re:The Moon Is A Harsh Mistress/"Mike"

[yuriklastalov]

You say that as though some crap made up by a pulp sci-fi author in the 50s as plot points in his stories are some kind of normative guide to actual technology. For Christ's sake, "muh three laws of robotics" were violated routinely in order to demonstrate the fact that they wouldn't actually work and yet here we are, with a bunch of sci-fi addled technophiles jerking each other off about muh three laws and an ignorant sense of the immanence of sentient AI.

Whenever AI is finally achieved, it's certainly going to have been developed for military applications in which anything even resembling muh three laws would be a detriment to its purpose. Mindless morons dreaming of their favorite sci-fi fantasies coming true refuse to realize that the scenarios in their treasured stories do not reflect reality in any way, that these "amazing technological innovations" are going to be used to subjugate humanity with a totality the 20th century totalitarians could only dream of.

Technophiles and sci-fi quacks have their heads in the sand. The history of technological development is not one of expanded human liberation, unless you think being shackled inextricably to managerial capitalism is liberation. I'm so glad to have the corporate overlords orchestrating every facet of our existence. At least we've got our precious "self actualized" identities based on consumption of pop culture and social justice to carry us through being treated like the disposable cogs we've become.

bwahahaha yeah right

[coolmoe2]

Ill belive that once they get something that works in the real world outside of kicking as at video games.

In the real world computers can't even drive a car which even the dumbest among us can do.

Re:bwahahaha yeah right

[Bryansix]

Dumb people don't drive cars very well at all. In fact if we didn't have laws against it and extreme enforcement, the levels of drunk driving would be significantly higher.

Much better link

[rthille]

https://developer.nvidia.com/e...
TLDR: 10x better power efficiency than the TX2 and 20x the performance.

Average lightbulbs don't use 60 watts

[Bryansix]

It's 2018 and most of by lightbulbs use between 8 and 15 watts each and contain only a single LED. Yes this computer is energy efficient but lets ditch the anachronisms.

Re:Average lightbulbs don't use 60 watts

[Anonymous Coward]

The original point wasn't exactly pedantry, a decent portable laptop often uses below 30W on average in use and that is powering the full device; the only reason for comparing to an element based bulb is to try and exaggerate the energy efficiency.

Let's repurpose this

[RhettLivingston]

It strikes me that any system for automating a robot that is relatively small and low power is likely to have great attributes for creating personal augmentation systems. The basic requirements of personal augmentation systems include analyzing the surroundings and increasing the person's situational awareness or responsiveness in some way. The only thing this is missing is the features required to deliver the feedback to the individual. Presumably, they aren't missing the outputs to help control devices augmenting personal actions.

Some augmented applications this could be great in include training in virtually any professional sport but especially outdoor ones, motorcycle helmets (especially in sports and off-road contexts), hunting (many hunters would shell out thousands for a full augmented reality system geared for hunting), snow skiing, guided aircraft repair activities, guided vehicle repairs in general, an EMT aide, controlling artificial limbs and exoskeletons, many military applications ranging from personal battlefield computers to personal warehouse and flight deck computers that could tie in with exoskeletons. All of these applications require realtime processing the live environment and presenting analyses and optimal plans based on that analysis and the goals of the tasks being performed to the user.

If we don't want to be replaced, we should stop trying to slow our replacement and instead speed the development of systems that help to make us less replaceable.

How About

[jmccue]

How about releasing the source under an Open/Free Source licence so all non Microsoft/Apple OSs can use your cards. That will help a lot more people than any AI will.

I got burned once (had to buy a new video board), will never buy anything with Nvidia until the above happens

Re:BRAINZ!!!!!!

[LifesABeach]

What a great idea for a game and movie plot!

Re:9 billion transistors?

[DivineKnight]

Shhh, is marketing. Always find impressive numbers that may or may not contribute to a positive view of your product, and cite them.

When the Last Tree Is Cut Down... etc etc

[timerider]

Am I the only one who sees "Someday, there will be billions of intelligent machines in manufacturing, home delivery, warehouse logistics and much more." and translates that to "some day there will be billions of unemployed"?

Can someone PLEASE wake up those idiots and make them understand that ROBOTS do not buy PRODUCTS?

Re:When the Last Tree Is Cut Down... etc etc

[JoeDuncan]

... ROBOTS do not buy PRODUCTS?

Huh, and I'm sure we were all going to pay them *salaries* until you pointed this out. You must be a genius or something.