Thinking Deep

DeepMind is Google’s AI arm. With DeepMind, Google has developed technologies that we might have been using on a regular basis. They have made and are developing groundbreaking innovations that not only benefit the IT industry, but also inventions useful in healthcare and in biotechnology. They are also an AI company that pushes for responsible creation and development, with their promotion of ethical AI policies.

With the cool portfolio of DeepMind, let’s look further into some interesting facts about the company:

DeepMind was not Google’s Idea

DeepMind is the brainchild of two independent AI technologists, Shane Legg and Demis Hassabis. They launched the startup AI research laboratory in 2010, which was located in London. Their primary goal was to make artificial intelligence tech that is multipurpose, stretching it to do almost anything, because at present robots only function on one specialty.

The AI company was only then acquired by Google Inc. in 2014, and then promoted as a subsidiary unit of Alphabet Inc., Google’s mother company, the following year. Since then, they have produced promising products and technologies such as The AlphaFold, AlphaGo, AlphaStar, AlphaZero, WaveNet and WaveRNN. Some of their inventions are now part of many Google apps.

The Initial Founders are Computational Neuroscientists

Hassabis and Legg met at Univesity College London, the former doing his PhD (cognitive neuroscience) and the latter employed as a researcher in the Computational Neuroscience Department. Hassabis mentioned that he went back to university to study how brain intelligence functions and apply it in artificial intelligence.

Hassabis has outstanding achievements since age 13, having a 2300 Elo rating (one considered as a chess master standard). His next feat was in his teenage years, completing secondary school around age 16, then at 17 began building his video games: Syndicate, and Theme Park, which was awarded with Golden Joystick Award). His career in artificial intelligence then began in making games.

While Legg is a Kiwi who worked on artificial intelligence since his university years. His masteral thesis is about Solomoff Induction, a theory of an algorithmical physical universe. His doctoral dissertation, “Machine Super intelligence,” tackles about intelligence (its nature and how it’s measured), universal general intelligence, computational agents’ limit, and temporal difference learning.

DeepMind Thought They Weren’t Ready About WaveNet

Notice the difference of Google Assistant’s text-to-speech interface compared to other AI virtual assistants? What about the amazing Google Translate and Google Search speech recognition?

You might have just LoL’ed or rolled eyes at the amazing part, but in all seriousness DeepMind has had a hand with the technology. Called WaveNet, the commercialized version was used in these three Google apps.

WaveNet is the speech recognition API created by DeepMind. Google has had doubts it will be integrated into their commercial products because they admitted that it was not easy doing the work, computationally speaking, but it was just in time in 2017 to be included in Google Assistant. Cool, yeah?

Their Bots Organically Learn How to Play Video Games

The technology somewhat sounds complicated as deep reinforcement learning uses deep learning (on a neural network that is convoluted), plus q-learning, plus reinforcement learning (model-free). Basically, they program a non-complicated code then let the bot do its thing in finding out what works and what does not.

With this, they created Agent57, a bot with a genius playing level on 57 games of Atari, the video game console. It is a pretty smart group of AI techniques that make artificial intelligence a little scary. Imagine, robots learning from virtually scratch…

Their Bots Beat Top Human Go Players

In 2016, DeepMind’s AlphaGo has beaten a 9th Dan in Go, Lee Sedol, in 4-1. It was very sensational it has been made into a movie. The following year, it has beated Ke Jie, the world no. 1 at Go at that time. If chess is complicated and has many possibilities, the game Go has many more possibilities of moves than chess, making a champion bot hard to program. But Google did. And they even improved it with AlphaGo Zero and AlphaZero.

AlphaGo et al played with itself many times and also learned from the past games of real human players. That is where it developed its intelligence.

They Give Importance to Robot Ethics, but…

The third founder of DeepMind, Mustafa Suleyman, is an advocate of making responsible technology and IT products. He made a subunit within the company named “DeepMind Ethics and Society,” which researches about the wide scale impact of AI in the real world and promote ethics into the workplace and in their works. He is also a co-chair of Partnership on AI, an organization like G7 but in the IT industry, where they talk about AI’s positive public awareness and responsible practice.

It is ironic, however, that Suleyman has been marked on probation in his last days with DeepMind because of bullying accusations by his employees. His managerial methods might have been very different. He then left the subsidiary company and transferred to Google Inc.

They Have Significant Contribution to Protein Folding Predictions

This is one of the great feats of DeepMind: AI protein folding. It was just demonstrated in a competition, but the DeepMind AI protein folding prediction showed outstanding results in finding out what parts of a whole string of protein will bend and fold.

Why is this vital? This type of prediction is needed for advanced research because a protein behaves differently simply by having a different whole structure—where it folds and bends as it is synthesized. It has its importance in the biology and medical field, but protein folding prediction is quite expensive and laborious to do using physical laboratory methods. With DeepMind AI, protein folding prediction can greatly cut on costs and efforts.

Elon Musk Considers DeepMind as His Rival AI Company

Along with FAIR (Facebook AI Research), DeepMind, and OpenAI are the top three IT companies are that are leading in the artificial intelligence invention race. Though Elon Musk openly stated OpenAI is in a loose competition with DeepMind, as they both utilize seemingly strong AI methods and because they are capable of doing so.

The ironic thing is that Elon Musk was one of the key investors in funding DeepMind. Though he co-founded OpenAI later and set it to rival DeepMind, Google AI Arm.

Wrap Up

With acquiring DeepMind, Google has become a key player in developing artificial intelligence technology, and in pioneering mindful AI inventing. The founders: Hassabis, Legg, and Suleyman, along with a great roster of talents, have exceptional skills and abilities that have shaped DeepMind into what it is today.