Inside Baidu's Bid To Lead the AI Revolution (wired.com)

mirandakatz writes: China's search giant missed mobile: As WeChat and Alibaba deftly transformed their companies to suit mobile, Baidu stayed stuck in browser mode. It can't afford to make that mistake with the AI revolution -- and, as Jessi Hempel writes at Backchannel, it just might have an edge in its bid to come out on top. There's huge governmental support for AI in China, including a plan to make the country the world leader in AI by 2030, and it has double the number of people online than America does -- AKA vast quantities of raw data. Hempel traveled to Beijing to chronicle this tenuous moment in Baidu's history, and has delivered a deep look at Baidu's AI be on AI, speaking with key leaders including CEO Robin Li and COO Qi Lu. She writes that 'Robin Li is doubling down on a future beyond 2017. In that future, Baidu is not a series of products, but rather an engine that belongs inside everything -- an engine that powers Baidu back to dominance in China, and possibly far beyond.'

Comments

Not to mention...

[SuperKendall]

.. in addition to the larger online population, the huge state-sanctioned lack of respect for user privacy that should let Baidu collect even *more* tasty data from everyone!

Re:Not to mention...

[DNS-and-BIND]

Oh, that's so precious that you think Facebook, Google, and Equifax are any better, if not magnitudes worse.

I'll bet you Baidu has a lot better security than any of those, and the executives will go to jail if they get hacked. So please continue with your nativism and looking down at another culture.

AI evolution

[Botnet-of-People]

The way I see it. There won't be an AI revolution. Slowly the pieces of the weak-to-strong AI puzzle will fall into place, and only someone who's been comatose all the time will see the arrival of strong AI as a revolution. Examples of the pieces, weak AI that can do crude image recognition, synthesized voices like SIRI that actually sound better than a lot of non-native speakers of a given language (English in particular), weak AI that can defeat the best human minds in games like chess and go, self driving vehicles that can already drive better than a student driver.

3 Laws Safe?

[LifesABeach]

I cannot but wonder if this new AI can answer the question, âoewho can step on my Blue Swede Shoes?â

Re:3 Laws Safe?

[darthsilun]

Only blonde Carl Perkins can step on your Blue Swede Shoes.

Your Blue Suede Shoes are a whole 'other matter.

No shortcuts.

[Gravis Zero]

There seems to be some misunderstanding by the business types that we are on the verge of making AI that can do more than parlor tricks. Even if you formulated a way to make a generalized intelligence tomorrow, we are like the cutting edge researchers that created modern computer graphics algorithms: we have the idea but the means to realize are far beyond our reach.

If you really want to redefine the future of humanity then you need to turn an ultra-low power electronic scheme (like Quantum-Dot Cellular Automata) into a reality because then we'll be able to build the ultra-complex systems that AI need. We need billions of neurons per chip just to get started. Things will be interesting when we get quadrillions of neurons per chip.

Re: No shortcuts.

[Anonymous Coward]

We're nowhere near the scale of billions of neurons on a chip yet, the largest neural nets we've ever trained are still only at the number of neurons of, say, a frog. We're a few orders of magnitude away from the 200 billion odd neurons in the human brain.

At the moment, the power requirements for AI are still very high, but that's because we're emulating neural activity in software. Trainable hardware networks are coming, and should reduce power requirements significantly (though probably not the training difficulties).

The next revolution beyond so-called parlor tricks is going to be trainable robotics. Rather than building a robotics model using hand knowledge, we get the robot to learn the task we want it to do using RL (this is in fact already possible).

Beyond that it's anybody's guess. My bet is on policies that are able to take a dynamically changing environmental context into account. Methods to do this exist, but they're in their infancy. The idea is to factorize the environment into higher level abstractions that can be used to make decisions beyond mapping a statically sized state vector to an action.

It's likely that this type of framework will need to be solved if we want to beat StarCraft. Interestingly, a number of quirks and facets of human intelligence seem to fall out of this stuff. AI is not going to be some ignition like the atomic bomb, it's going to be a long, slow climb to something that very closely approximates what it is that we do.

Chinamen going to imitate us U.S. guys again?

[Anonymous Coward]

You Chinamen going to imitate us US guys again as you did MY work http://www.theregister.co.uk/2017/04/26/boffins_supercharge_the_hosts_file_to_save_users_plagued_by_dns_outages/ as it's part of what APK Hosts File Engine 10++ 32/64-bit https://www.google.com/search?hl=en&source=hp&biw=&bih=&q=%22APK+Hosts+File+Engine%22+and+%22start64%22&btnG=Google+Search&gbv=1/ does!

* IMIITATION = the SINCEREST form of FLATTERY!

APK

P.S.=> Yes folks - it's NOT easy being "world-class" (lol, like ME!)... apk

What AI is

[neoRUR]

When all these companies talk about AI, what they really mean is a way to provide back-end servers, neural net chips, GPUS and such so that they can charge for the deep learning systems that everyone wants to use to process the data to find those little nuggets. They are not working on building general AI systems or even working on how to architecture one. They want to make money and the money is in the service. Just like Amazon, Google and IBM. Facebook uses AI and provide some services. But the future where you need fast real-time detection of things that your automated cars and planes will be doing, where it sends all that to a massive computer to do the processing and sends the results back is what they are talking about (at least in these last few and next 5 years).
The AI systems, based on neural nets as still so simple compared to what humans can do, it will be a little while before real general AI gets around. And the Robotic technology and hardware to house it is so far behind the software at this point.
It will all come, bio-robotics, Bio-AI, might be here first.
You can't stop technology and you won't stop developing AI systems, we learn more and more each month, but there are still a lot of unsolved hard AI problems and formalizing all that is even harder.

Re:What AI is

[m00sh]

When all these companies talk about AI, what they really mean is a way to provide back-end servers, neural net chips, GPUS and such so that they can charge for the deep learning systems that everyone wants to use to process the data to find those little nuggets. They are not working on building general AI systems or even working on how to architecture one. They want to make money and the money is in the service. Just like Amazon, Google and IBM. Facebook uses AI and provide some services. But the future where you need fast real-time detection of things that your automated cars and planes will be doing, where it sends all that to a massive computer to do the processing and sends the results back is what they are talking about (at least in these last few and next 5 years). The AI systems, based on neural nets as still so simple compared to what humans can do, it will be a little while before real general AI gets around. And the Robotic technology and hardware to house it is so far behind the software at this point. It will all come, bio-robotics, Bio-AI, might be here first. You can't stop technology and you won't stop developing AI systems, we learn more and more each month, but there are still a lot of unsolved hard AI problems and formalizing all that is even harder.

Then why does Baidu have Apollo?

The cutting edge of R&D on all AI is often supported by tech giants like Google, Facebook, Microsoft, Baidu.

I think you are confused

[SuperKendall]

Oh, that's so precious that you think Facebook, Google, and Equifax are any better, if not magnitudes worse.

You don't seem to understand - all of those guys are terrible but limited in what they can collect by U.S. laws and and technology.

Baidu will have the advantage of the government *helping open* up any data that might be there to collect, in return for eternal and direct government access.

Imagine for a moment that Google was tied into every U.S. government security camera, for example...

So it doesn't matter if the motivation of the companies you list are worse, when the CAPABILITY of Baidu to collect data will be so much greater...

I'll bet you Baidu has a lot better security than any of those

From who? When a service is run BY the Chinese hackers, you can be a little more lax.