Stay on Top of Enterprise Technology Trends
Get updates impacting your industry from our GigaOm Research Community
Artificial intelligence might be the most misunderstood term in technology. It conjures up images of malevolent robots and self-aware computer systems capable of outwitting — or at least matching wits with — human beings. It is not that. At least not today.
Google’s acquisition of artificial intelligence startup DeepMind for $400 million sent the tech world atwitter earlier this week. Everybody wanted to know what the mysterious company was up to and why Google was willing to pay so much for it. After a day or so of mystery and even speculation that Google wanted to turn its new robots into sentient beings, the probable truth finally began to emerge.
Google just wants to build a better search platform, and talent isn’t going to come cheap with everybody in the web vying for it.
DeepMind was pricey, but it wasn’t novel
DeepMind was working on some form of artificial intelligence technology, although the details are still somewhat murky. It had filed patent applications around image search, and the Re/Code post linked to above quotes AI expert Yoshua Bengio, who described a DeepMind paper about teaching a computer to learn the rules of Atari games as “essentially using deep learning.” Here’s a primer (albeit one in need of an update) we did on deep learning in November.
Whether or not its methods or its people are worth $400 million is up for debate, but one thing is not: DeepMind was just the latest in a string of similar acquisitions of artificial intelligence talent and technology by large web companies. And it won’t be the last.
We recapped the recent activity earlier this month when Pinterest bought a computer vision startup called Visual Graph, but here’s an abbreviated version of moves that happened throughout 2012 and 2013: Dropbox bought Anchovi Labs; Google bought DNNresearch (and its co-founder Geoffrey Hinton); Yahoo bought LookFlow, IQ Engines and SkyPhrase; Facebook hired Yann LeCun to head up its new AI lab. Various students of Hinton and LeCun (who are also professors), as well as of their peers from other top universities, are floating around companies like Google, Facebook and Microsoft.
A betting man might wager that a Silicon Valley startup called Vicarious is one of the next up for acquisition. It launched in August 2012 with $15 million in venture capital and in October 2013 claimed it has passed the Turing test by successfully cracking CAPTCHAs at up to a 90 percent rate. Vicarious is clear to point out that it doesn’t do deep learning (nor do some of the other startups mentioned), but it’s trying to accomplish a similar task. If someone is interested in buying it, the acquisition price might depend on who else is making offers.
AI means automation, but it starts small
Which, actually, brings up one other thing that’s not up for debate: deep learning, artificial intelligence and similar technologies are not what we instinctively want to think they are. Largely, the companies and researchers being acquired by Google, Facebook et al are focusing on two things: computer vision (usually object recognition) and natural language processing. Their algorithms try to learn the features of objects and the meanings of words and phrases so computers can automate tasks such as classification.
(In other areas, companies are applying machine learning for pattern detection in fields ranging from cybersecurity to music recommendation.)
We’ll discuss many of these use cases at our Structure Data conference in March because, well, they’re very interesting. However, they are not yet, at the the risk of being trite, SkyNet, HAL or anything even close to as smart as a human.
In the near term any type of advanced machine learning represents a way to offer a stickier platform for companies such as Google, Microsoft and even Facebook. Whether we’re talking about mobile devices, laptops, gaming consoles or wearables, features such as better image search and recognition, better voice controls, better natural language interaction (think search or text message completion) and more-accurate translation will make for a better user experience. Exposing these capabilities via API will make for a better developer experience that, in a virtuous cycle, should make for a better user experience.
No matter who’s working on these technologies, they also represent a method for making use of the previously unanalyzable data they’ve accumulated over the years. Namely, that means photos, videos, and text in the form of blog posts, wall updates, captions, you name it. Being able to extract objects from images and concepts (or sentiment) from text will help users organize their content and advertisers learn what people are interested in. The companies themselves should be able to get a better sense of what their users have been up to and how that has changed over time, for example.
Here’s Google Fellow and deep learning expert Jeff Dean talking about the technology at our Structure conference last June.
The future is brighter — and possibly pricier
Fast-forward a few years, though, and we could be seeing some major advances taking place in terms of how powerful and accurate these AI systems can become, and what types of problems they can address. Deep learning research, especially, benefits from lots of data, lots of computing power and lots of smart people — and Google, Facebook and Microsoft have those things in spades. It’s not inconceivable that artificial intelligence could power smart homes, smart cars and smart robots, especially when combined with other data sources and systems.
Put in the hands of developers via APIs and cloud computing platforms, who knows what types of uses we’ll find for these capabilities. Turned on cancer research, crime prevention or other societal problems, they could change lives.
So, yes, Google does have big plans for the DeepMind team, but it’s neither the first nor the last investment the company will make into artificial intelligence. It appears that web companies and platforms that want to remain relevant will have to find new ways to bake AI into the consumer experience. Until pool of available talent in the AI world catches up with demand, that means the best intellectual property and the brightest minds will come at a high price tag.
Feature image courtesy of Shutterstock user Sebastian Kaulitzki.