DeepMind Shows How Expensive the AI Race Is & Google’s Strategy to Win It
4 Oct, 2019 15:02
Author: Michael K. Spencer
Editor’s Note: Michael K. Spencer, who is a Blockchain Mark Consultant, tech Futurist, a prolific writer.
Can startups compete with BigTech when it comes to artificial intelligence? DeepMind is a fairly interesting case study here.
In the summer of 2019 there’s been a flurry of articles on DeepMind’s achievements and its financial losses. Since the second round of Softbank’s Vision Fund is supposedly going to be AI-centric, I thought to look into it a bit more.
I love DeepMind, I think it’s one of the hottest talent beds in AI doing important work.
DeepMind is incredibly promising, and sits as one of Alphabet’s “Other Bets”. DeepMind had 700 employees as of mid 2018. On LinkedIn, in 2019, it shows over 850 employees.
DeepMind is like a Multi-Billion Dollar AI Startup Hiding in the Backyard of Google
That’s a pretty huge growth in headcount by all reckoning. So the Google-owned DeepMind lost $572 million in 2018. That’s a small price to pay for AI supremacy and all the money it can save Google, for Waymo or in the great ambition that is Google Cloud.
DeepMind’s achievements are dazzling in the realm of healthcare and game simulations and show new ways of training deep learning systems.
Indeed in March 2018, Google’s cloud business announced a new service that converts blocks of text into natural-sounding speech, the first product containing DeepMind code that’s for sale. The first, but not the last.
I love reading DeepMind’s blog about all the cool projects AI is up to these days. Google is a world leader in AI, and DeepMind has been our bet on why it can remain in the lead. There’s no way a $500 million burn is too expensive, this is the AI arms race we are talking about. Remember, Softbank’s Vision Fund second installment will be giving out $100 billion in funds.
Research costs money, and DeepMind is doing more research every year. The dollars involved are large, perhaps more than in any previous AI research operation, so I agree with the Wired and Forbes spun PR on this account. This is important work.
Especially if the United States wants to stay ahead of China in AI in the years to come. DeepMind’s AI has beaten chess grandmasters and Go champions but you know the end-game of this stuff may be a bit more ambitious! Fancy new King’s Cross locations aside, I’m very curious to see what DeepMind will do next.
Google Beat Facebook to Acquire DeepMind in January, 2014
Not many people realize this but Google got DeepMind before Facebook could snap them up. Who controls the future of AI is really important for the future of humanity, to be honest. As a futurist, this is the sort of AI-regulation governance issue that most concerns me.
Alphabet operates other AI research groups, but DeepMind has been doing more futuristic work and is more R&D meets the future with things like Waymo or deep learning in Healthcare. It’s not Microsoft or Apple level machine learning that’s more product-focused, it’s bigger stuff.
DeepMind was founded by neuroscientist Demis Hassabis, a former child prodigy in chess, Shane Legg and Mustafa Suleyman. Alphabet can afford such “other bets” here, and who can compete with that kind of scale, realistically? Only a handful of other Tech companies and the Chinese Government perhaps.
One of the Best Acquisitions Ever for a Tech Company
The rising magnitude of DeepMind’s losses are not to be discounted though, the costs are real. Let’s see now: $154 million in 2016, $341 million in 2017, $572 million in 2018. Should be near $700 million in 2019, I’d wager. More or less what a WeWork is losing but doing far more important work for our global future.
DeepMind’s accountant said it had “written assurance” from Google that it would be supported for at least another year and helped to pay its debts. DeepMind could save Waymo $billions ahead of its Waymo One product, and has already improved its pace in an industry that’s going to be huge even by 2026.
DeepMind is such a concentration of AI talent Google has a pristine advantage over all other BigTech companies in artificial intelligence. It’s besides YouTube, one of the most important and paradigm-altering acquisitions a tech company has ever made. That’s, in a nutshell, how bullish I am about DeepMind.
In the 21st century, I’d give DeepMind around a 3% probability of becoming something that could lead to artificial general intelligence (AGI). As unlikely as we are to see it within our lifetimes, this is the sort of hub that you’d expect has the best chance. These the dreams of the sci-fi agents in some of the most brilliant men today.
Towards New Paradigms of Deep Learning
DeepMind is evolving and is a global leader in deep learning. In a nutshell, DeepMind has been putting most of its eggs in one basket, a technique known as deep reinforcement learning.
That technique combines deep learning, primarily used for recognizing patterns, with reinforcement learning, geared around learning based on reward signals, such as a score in a game or victory or defeat in a game like chess. Quite funny how prominent training on video games has become then in this kind of deep learning. If you Google the December 2013 paper “Playing Atari with Deep Reinforcement Learning”, you’ll see what I mean.
DeepMind went on to beat all of the best humans at playing games like Chess, Go and video games. That opened up the eyes of a lot of skeptics about the potential of this generation’s iteration of machine learning swag.
Google is a Coin Machine for the Future of AI
Google is putting down the Alphabet of the future with this exciting team. Even as DeepMind continues to hire hundreds of expensive researchers and data scientists but isn’t generating any significant revenue, Google has what amounts to an unlimited budget on this particular goal.
DeepMind has fundamentally changed how we see the future of AI.
We don’t even understand fully the end game of this company in five years’ time. The company made headlines in 2016 after its AlphaGo program beat a human professional Go player, Lee Sedol, the world champion, in a five-game match, which was the subject of a documentary film. That was just 2 years after Google gobbled it up.
Google’s advertising revenue will lead to the future of artificial intelligence. How Google and Facebook decided to game the internet with Ads will lead to the kind of AI that will likely dominate the Earth. Think about what that means for a moment. It will have Google-like ethics!
That could be a good or a very bad thing.
As dazzling as deep learning seems on the outside, it’s also a bit dumb. This kind of deepreinforcement learning is a kind of turbocharged memorization; systems that use it are capable of awesome feats, but they have only a shallow understanding of what they are doing. Machine learning seriously has a lot of limitations in 2019.
But whatever DeepMind is today is nothing compared to what it can and likely will become.
You can be an optimistic or a pessimist about AI, but it doesn’t matter. Companies tethered to the future of AI have a special kind of aura: Google, Nvidia, ByteDance, Baidu, Box, etc…. First and foremost among these would have to be Google though. First, in Google, lives DeepMind.
In a world owned by massive global corporations, who can afford to even play or compete with Google in AI?
Original Source: https://medium.com/artificial-intelligence-network/deepmind-shows-how-expensive-the-ai-race-is-googles-strategy-to-win-it-7e588fa08cb