How I see ‘DeepMind’ as a promising future for Machine Intelligence.
A state-of-the-art company advancing in AI.
“Does this implicate that the finite human intelligence is transformed into infinite intelligence by the creation of a human-like machine?”
It has nearly been ten years since the company, ‘DeepMind’ broke new ground in the Artificial Intelligence (AI) domain. Although the past decade has seen a rapid development of smart machines in many companies, still I specifically chose to write on DeepMind because I have been inspired by it for about two years now. I chanced upon an article on some website while I was struggling with formulating an idea to kick start my very first research paper: I knew what to do but the pressing question at that time was, ‘How to do’? Browsing indefinite web-pages, tired of clicking anywhere and everywhere and still not finding the precise machine learning branch to incorporate into my research, I was exhausted by the time the clock struck midnight. Flagging by now, I opened another link and voilà! I found DeepMind. At first, More than the kind of work it does, I was far more intrigued by the name ‘DeepMind’. I am still fascinated by this name, tell me in the comment section below how do you feel about it. Anyways coming back; At that point, I knew it had to do something with Deep-learning or Neural-networks and hence I was relieved that I might have reached the stop. Little did I know that this stop would be my start to every new research.
If you are unrelated to the field of computer science, then you probably must be thinking what is this company all about. Well, it started in 2010 when inventions in the field of artificial intelligence were not as buoyant as they are now in 2020. The pedestal intentions were clear to the founders: Demis Hassabis, Mustafa Suleyman and Shane Legg, to build a powerful company. It was all about bringing experts from variegated branches of AI and making safe and smart machines for solving problems through machine intelligence; primarily producing expert AI tools to combat the issues related to humankind. It soon became a highly reputed artificial intelligence startup with its early but massive success in computer games. They built an AI program which could learn how to play ‘forty-nine’ different Atari games right from scratch just by learning from its previous scores, to be literal, its past ‘experiences’. The AI agent later came to be known as ‘Alpha Go’ the well renowned first AI program to beat a professional player in the classical ‘Go’ game.
“The kind of learning, which is based on experiences, is called Reinforcement Learning (RL). These RL algorithms were the base to build the program of Alpha Go”.
Side note: Like, my any other research fascination, I was infatuated with the use of ‘Deep RL’ algorithms in Atari Games after studying this research paper by DeepMind; To my surprise, I realized it later that I always had an Atari sweatshirt. Even though it has been years now, I still love it because it is reminiscent of how I started to pursue research.
Coming back to the context, the success DeepMind gained was perhaps indicative that the winds of change might begin to blow soon. Understanding the incredible potential DeepMind withholds, in 2014, the tech giant ‘Google’ bought this company for more than 500 million dollars. Again in 2015, DeepMind became a whole owned subsidiary of the company — ‘Alphabet Inc’. This completes a brief history of the company. So, in a nutshell, it is a UK originated Artificial Intelligence-driven company and a research lab which is the hub for formulating quicker AI solutions.
Now that you are familiar with DeepMind let us dive a little more deep. While working on my research project and taking inspiration from the deep reinforcement learning algorithms of DeepMind, I learned that some of the remarkable research areas this company works on, are: Deep Learning, Control and Robotics, Reinforcement Learning, Unsupervised Learning, Neuroscience and Safety. These are some of the most thriving research areas of 2020, and DeepMind has pioneered the art of combining all the advances in machine learning for scientific discoveries and overall public welfare. This exceptional and innovative ability is very much evident from their significant research contributions:
1) Alpha Zero: It is an AI agent which learns to play Chess, Shogi and the game of Go, right from scratch, on its own. Not to mention that this artificial mind has defeated many grandmasters in their own games, starting with a random play with absolutely no knowledge embedded in its system regarding the rules of the game. This program has achieved exceptional and faster results as compared to any other smart system developed ever before. Studying this research paper, I realized that having Reinforcement learning algorithms as a backbone, it won 91.2% of overall games. Here is what a former world chess champion, ‘Garry Kasparov’ had to say about Alpha Zero:
- “I can’t disguise my satisfaction that it plays with a very dynamic style, much like my own!”
- “The implications go far beyond my beloved chessboard….not only do these self-taught expert machines performs incredibly well, but we can actually learn from the new knowledge they produce.”
This impeccable intelligence poses two questions: ‘Has the human intelligence produced something from which we might learn in the future?’ and ‘Does this implicate that the finite human intelligence is transformed into infinite intelligence by the creation of a human-like machine?’
2) Differentiable Neural Computers (DNC): This was the next breakthrough in the field of AI. It is a memory augmented neural network. Non-technical readers can consider this DNC as a mimic of human memory. For example, if your new neighbour asks you something like this- “If I cross the main road, take a left, then walk for about a kilometre, which of the gas stations will I be nearer to — A or B ?” To answer such a question, you would generally reply by simultaneously plotting a map in your mind while the question is being asked. Retrieving this map from your memory and responding to the problem whilst you are imagining, is the process which DNC can replicate.
Isn’t this so fascinating? For all those ‘NeuralHeads’ who wish to read further in-depth, check out this complete research paper and this GitHub repository.
3) DeepMind has played its card not just in modern inventions but also in one of the most prevalent issues since decades — Energy Consumption. In 2016, DeepMind helped Google reduce its data centre cooling bill by 40%. By incorporating super-efficient energy savers, they could utilize 3.5 times the computing power used before 2016 with the same amount of energy. Honestly speaking for an enormous company like Google, a 40% reduction is phenomenal. I can’t help but think about a time in future where regular households can benefit from such a technology.
These were just some of the significant researches done by DeepMind.
Quite often, people tend to perceive the nexus of artificial intelligence as a far-fetched possibility. To comprehend AI’s ability, they tend to draw a parallel image to sci-fi movies. The real reason behind it might be the failure to perceive its power in their day to day life. Of course, it is evident that in 2020, smart machines are not an integral part of every regular household but there is no hiding from the fact that AI has started to establish its roots all around us. The actuality that we are using smartphones equipped with AI assistants (Google Assistant, Siri, Bixby, Cortana, Alexa) and bots (Echo by Amazon) which we use these days to perform simple activities are indicative that we are already involved in AI’s network. Then why is it a common notion that AI is still a distant possibility? Almost ten years ago, it wasn’t easy to see the advancements we are experiencing now since it was just the start. But now that we have stepped into the ever-changing AI world and have learned to be its part then it is not apt to ignore the highly advanced researches going on to build machine intelligence so that it might be an “inherent” part of us someday. This is the reason why I see DeepMind as a promising future for machine intelligence. The state-of-the-art research it carries out, speaks about the various possibilities that the artificially created intelligent minds might be a crucial part of our future lives.
“Artificial intelligence will reach human levels by around 2029. Follow that out further to, say, 2045; we will have multiplied the intelligence, the human biological machine intelligence of our civilization a billion-fold.”
— Ray Kurzweil, American inventor and futurist.
For the nonce, I can say that if something like this is stirring in the company’s pot right now, then it is not hard to comprehend the kind of potential it is holding on for the future. Perhaps, this is one of the primary reasons why Google invested in this company — to be a super-competitive power in the AI industry.
