Tech CEOs Declare This the Era of Artificial Intelligence (fortune.com)

You will be hearing a lot about AI and machine learning in the coming years. At Recode's iconic conference this week, a number of top executives revealed -- and reiterated -- their increasingly growing efforts to capture the nascent technology category. From a Reuters report (condensed): Sundar Pichai, chief executive of Alphabet's Google, said he sees a "huge opportunity" in AI. Google first started applying the technology through "deep neural networks" to voice recognition software about three to four years ago and is ahead of rivals such as Amazon.com, Apple, and Microsoft in machine learning, Pichai said.
Amazon CEO Jeff Bezos predicted a profound impact on society over the next 20 years. "It's really early but I think we're on the edge of a golden era. It's going to be so exciting to see what happens," he said.
IBM CEO Ginni Rometty said the company has been working on artificial technology, which she calls a cognitive system, since 2005 when it started developing its Watson supercomputer.
Artificial intelligence and machine learning will create computers so sophisticated and godlike that humans will need to implant "neural laces" in their brains to keep up, Tesla Motors and SpaceX CEO Elon Musk told a crowd of tech leaders this week.Microsoft, which was absent from the event, is also working on bots and AI technologies. One company that is seemingly off the picture is Apple.

Comments

Outsourcing Me

[SuperKendall]

Lots of people have long had the dream of putting together a chatbot that would represent them in online forums...

Well I'm going the opposite route. I'm attaching a chatbot to my source code editor for work, leaving me free all day to do nothing but post in online forums!

As for the work quality, I wouldn't worry about that - one of the neural inputs is StackOverflow recent answers.

Re:Outsourcing Me

[Applehu Akbar]

These CEOs are kidding themselves. It'll be another hundred years or more before people have anything approaching real AI.

A hundred years for strong AI is about what I would expect. As that time approaches we will see the broadening of today's narrow AI into more and more new places.

Re:Outsourcing Me

[Hairy1]

Only it isn't specific. Thats the point. The same neural net approach is being used in multiple domains. It exhibits the same kind of robustness that we do. Sure, we are not there yet, but if you look at the rate of progress it is evident that we are at most two or three doubling away from general purpose intelligence in machines, This means four to six years away. One hundred years? Not a chance. The chance of having genuine universal AI within ten years is about 90%.

And even if the technology progress stagnated, as can happen in technological progress, the AI we have right now is already at the point it can drive a car and perform many jobs we think of as the domain of humans. What we do not have right now is a serious effort to identify an alternate economic model which accounts for this. There is the universal income, but there has been no detailed analysis of various economic models. It is all very well stating the obvious; that the machines are coming, but the hard questions really remain open. How should we adapt?

Re:Outsourcing Me

[fluffernutter]

Meh, they were talking about neural nets back when I was in school. If it has only progressed from Chess to Go in that time, then it has a long way to go.

"Increasingly growing"?

[jeffb (2.718)]

So, "accelerating"?

Tech companies spend more resources on trendy topic because tech companies spending more on a topic makes it trendy. Film at 11.

Re:"Increasingly growing"?

[AK Marc]

We are no closer to AI now than we were 70 years ago. All we have now is better dB lookups. *yawn*. Call me when someone creates an approach that has a possibility of creating AI.

Re:"Increasingly growing"?

[Anonymous Coward]

Deep neural nets are not dB lookups. They mimic the way the brain stores and recalls patterns and responses to patterns. Specifically they (both the deep neural nets and the brain) store the patterns and responses to patterns in the form of synaptic weights of multiple layers of neurons. If that is what you want to call dB lookup, then well the brain just works through dB lookups too.

Re:"Increasingly growing"?

[Etcetera]

We are no closer to AI now than we were 70 years ago. All we have now is better dB lookups. *yawn*. Call me when someone creates an approach that has a possibility of creating AI.

We can train neural networks far, far faster than we used to be able to. But outside of that, yeah, nothing except standard evolutionary improvements. Eventually standard improvements will be enough, but we won't notice that happening. 3D printing is starting to move on from plastic and we'll get a nice boost out of that far sooner than any AI improvements coming down the stream.

^ This. Vastly increased networking and data collection makes some of the previous growth requirements predictions meaningless. With enough data coming in (sensory-level), you don't need strong AI algorithms, you just need a neural network that can be bootstrapped.

Alphabet has clearly demonstrated the rapidity at with AI can be developed if you just shove data at it. Who need to sit there training a voice recognition product for years when you can just turn on the microphone in Android and have 100M people in the US providing voice samples all day long? There's petabytes of data out there waiting to be analyzed. Eventually, you don't need to hack your way to AI programmatically, when you can "brute force" your way to it by shoving it the data we're all providing the AI companies with on a daily basis.

And let's be honest; that's what Alphabet is: an AI company. That's the future of information.

What next?

[Threni]

It's great people are getting excited about AI. I'm looking forward to reading about it every fucking day, just like I did about voice recognition, how apps would change my life etc. At the very least, I hope it means it will become slightly easier to say things like "set an alarm at 2.30" and not end up with a calender entry which reads "self harming - tooth hurty" or whatever, but can we sort of pre-empt the whole thing and start thinking about what comes after AI so those of use who find it a little dull already can read about something else?

Don't we have to, you know... *HAVE* something....

[mark-t]

.... before we can say that this is the era of that thing?

What passes as AI so far is still just all smoke and mirrors.

Great

[decipher_saint]

Let's replace CEOs and stupid tech blogs with AI and put them on their own internet

Been there, Done that (Erin & Zach)

[Kevin by the Beach]

We have been creating Intelligences running on organic processors for all of human history. The two I helped to create have some bugs, but I blame the team programming effort with the wife. (we still argue about who introduced which bugs, and if a patch would ever be effective).

A newborn is simply a set of default starter programs that interact with an increasing number of inputs over time.

Partly cloudy and warm by the Beach

Relax

[Okian Warrior]

Relax everyone, we're nowhere close to having, what is commonly perceived to be, intelligent programs.

What we have, and what we have finely honed, are clockworks: algorithms that perform a single specific task.

Granted, a lot of what humans do can be replaced by a sufficiently well-designed clockwork. Lots of human tasks are repetitive, boring, and uncreative. Driving, for example, is repetitive, boring, and uncreative, and appears to be well suited to a clockwork.

And this will bring about massive changes in how we view human activity. We will eventually have to change our notions of entitlement and human worth, and found a new sect of economic theory.

But each of these is only a clockwork, suited to only a single task. Humans, the only example of intelligence we have, can learn to do any of these tasks, and as far as we can tell there is no wiring in the human brain specific to any of them. Humans can learn to play chess, checkers, poker, or any of a hundred other games, but so far as anyone can tell there's no wiring in the brain specific to chess.

A chess program can't learn to play checkers, but the human algorithm is universal.

We're starting to automate our world, that's all.

Tech CEO's

[Archfeld]

Tech CEO's famous for spouting techno-babble, raising and losing enormous amounts of venture capital, and utilizing golden parachutes declare something incredible is about to happen, just invest some money with us.

Re:All hype

[thinkwaitfast]

backblow will be when it fails to deliver. Read up about the AI winter which happened in exactly the same way in the 1980s.

I vividly remember this and attended colloquia with Dr. Hecht-Nielsen.

Concerning backblow, I used to like to temper people's hype with reality, but found it more entertaining to add to the hype and watch the downfall eating popcorn. Evil, I know, but I found schadenfreude much less stressful than living with a cassandra complex.

Re:Don't we have to, you know... *HAVE* something.

[mark-t]

where did I allege that I hated progress, exactly?

Re:BS

[ShanghaiBill]

We need about a 100 billion (thats billion with a B) times more advanced AI than what we have today

This is almost certainly not true. The human brain has 100 trillion connections. Some artificial neural nets (ANNs) have over a million. So the brain has 100 million (with an M, not billion with a B) fewer. But the synapses in the brain fire 100 times per second, while ANNs can clock a million times faster. So now we are within a factor of 100 ... but that is not all. As far as we know, a brain stores ALL information in synapses. So you are using synapses to remember what your third grade teacher looked like, your mother's voice, and what freshly baked cookies smell like. None of that is useful when you are, say, trying to ride a bicycle, and none of those other synapses are being used. But a computer only needs to load the synaptic data needed for a particular task, and leave the rest on a HDD. An array of HDDs can store petabytes of synaptic data (far more than a brain), and swap them in and out as needed. Furthermore, much knowledge can be stored in tabular or text rather than synapses, and use associative lookup that is way faster and less error prone than a NN. A human brain has to use NNs for everything, because that is all it has got, but a computer can only use it where appropriate, and use simpler algorithms when possible. There is little reason to believe that hardware capability is the limiting factor in AI. Of course faster hardware will help, but we mostly need better algorithms and more data.

Yep, still years away from a True Scotsman.

[jeffb (2.718)]

Driving is repetitive, boring, and uncreative? You should show up in some of the autonomous-vehicle threads and use that statement to confront the "machines will never be able to share the road with humans" crowd.

I'm pretty sure that human brains are no less "clockwork" than any of the things you mention -- just with more complex works, that are perhaps less reliable/predictable due to their implementation.

As far as the "universality" of the "human algorithm", well, greater human minds than mine have foundered on that question. How would you go about proving that there is nothing a human mind can't learn? At least, without falling into circular arguments ("since humans can't do that, it's not really learning")?