This article was written by Zachary Shipman, a Financial Analyst at I Know First.

Deep Learning Revolution: Google’s DeepMind

“Humans were stuck in a local maxima for 3000 years”- Google’s DeepMind CEO, Demis Hassabis.

Summary:

• DeepMind: The collaboration between Humans and Artificial Intelligence
• Why are games such an effective simulation tool for Artificial Intelligence?
• Inventive Strategies that allow AlphaGo to exploit human game plans
• Neural Networks and their function in the future
• I Know First’s Implementation of Artificial Intelligence technology

(Source: All About Circuits)

DeepMind: The Collaboration Between Humans and Artificial Intelligence

DeepMind is the world leader in artificial intelligence research. The idea is to develop AI programs that can solve any complex problem without having to be taught how. The algorithm can find structure on its own through experience and data collection. It can then apply it to all sorts of domains where there are huge amounts of data, complexity and volatility. So much complexity that the best human scientists can’t understand it all unaided. Ideally, it operates on the belief that once intelligence is solved, this technology could be applied to practically anything. The goal is to apply the artificial intelligence technology towards a positive worldly benefit. From climate change to the need for radically improved healthcare, too many problems suffer from painfully slow progress, their complexity overwhelming our ability to find solutions. With AI as a multiplier for human ingenuity, those solutions will eventually be possible.

(Source: Linkedin/AI)

Google’s DeepMind CEO, Demis Hassabis, shows that AI doesn’t only learn from human knowledge, but also creates new knowledge. The DeepMind program uses general purpose learning machines to run its complex system. General purpose learning machines allow the system to operate across a wide range of tasks. It learns automatically by analyzing the raw inputs. The algorithm is therefore entirely flexible and able to adapt to changes, because it is not pre-programmed to solve a specific task. The first job of the system is to build the best statistical model possible. Secondly, the system must then pick the best action based on available options that will drive a new observation.

DeepMind offers a human created system that can also teach humans new information, strategies, and techniques. It doesn’t just learn and copy what humans do, instead it actually innovates on traditional knowledge. This is a massive tool for future human intellectual progression. The algorithm can pick up on patterns that have been completely ignored by the human mind. The program is better strategically and thematically than humans. This creates new ideas that can be analyzed, absorbed and potentially used. Humans can reverse engineer the techniques used by DeepMind, and use those techniques to further advance future methods. DeepMind’s system allows humans the ability to explore a seemingly endless intellectual universe. Altogether, this allows for greater potential innovation. “Humans were stuck in a local maxima for 3000 years”- Google’s DeepMind CEO, Demis Hassabis. Humans and machines collaborating together in a complimentary way is a powerful tool that could change the way society operates.

Why Are Games Such An Effective Simulation Tool for Artificial Intelligence?

Games are the perfect way to test the progress of an artificial intelligence-driven algorithmic system. They give scientists the ability to test the same scenarios thousands of times without the risk of major consequences. Additionally, they help develop an understanding of pattern recognition. The great thing about games is that they usually keep some sort of score. It is easy to evaluate who is winning or losing. It is also easy to measure incremental changes in progress. There is also no bias when scientists use previously created games to test their algorithmic creations. The games were not created with future artificial intelligence simulations in mind. The system must adapt its learning to the game and not vice versa. Therefore, it is easy for scientists to evaluate the performance and ability of the algorithm. Games allow researchers the ability to run millions of experiments or simulations, measure progress, game scores, while measuring if algorithmic tweaks are gaining an advantage based on the performance in these environments.

(Source: New Scientist)

One of the main breakthroughs for artificial intelligence technology was the creation and ensuing effectiveness of DeepMind’s algorithmic program known as AlphaGo. Go is an extremely complex ancient game from China. This offers DeepMind a chance to test their program’s value. There are over 2000 professional Go game players in the world. The game is taken extremely seriously through Asia. Children who show initial promise in the game of Go are taken out of school at an early age and sent to study in privately specialized Go schools. In the past it was inconceivable that a computer program could outperform players who have dedicated their lives to the game of Go. Despite its relatively simple rules, there are more possible board configurations in Go than atoms in the universe. Potentially, AlphaGo could use neural networks to solve all sorts of different games. If AlphaGo could be successful at solving the game of Go, it is possible that it could also serve useful in many other facets of life.

(Source: Go Game Guru)

DeepMind originally started their testing with the Atari games from the 1980s. The system could play any Atari game directly from the pixel inputs. It did not even know the rules to the game. It had to solve the game by itself through a series of trial and error. After several 100 game play trials, DeepMind was able to play any Atari game at a level higher than any human. The system would develop innovative strategies that would exploit the rules and objectives of the games. An example of this can be seen when Google’s DeepMind created an innovative strategy when playing Atari Breakout. As seen in the video link provided, DeepMind was capable of creating a strategy that generated the perfect angle for every move. Games have been proven to be an effective tool for analyzing algorithmic performance.

Inventive Strategies That Allow AlphaGo To Exploit Human Game Plans

Elite human Go players rely on their experience, intuition and instinct to initiate a certain decision. They often have a somewhat dynamic strategy of how to exploit their opponent. The end-game is ultimately conquering the board by surrounding your opponent’s stones, or “walling off” empty areas of territory. In Go, all of the pieces are worth the same value. This creates a game where there are an incredible amount of potential strategic decisions. In an average position there are 200 possible moves. The game’s best may explain their past moves through their feelings of intuition. These moves may not always bring positive results, however it is essential that a player is always adaptive and confident in their instincts.

It is difficult to program functions that create specific strategies for a game where all pieces operate as equals. Programmers have tried for decades to write codes that could optimize game play. AlphaGo has its own creativity and intuition, inventing new knowledge and strategies. DeepMind is not pre-programmed and thus it learns through experience. The creators of AlphaGo allowed it to play itself 30 million times on their internal servers. The system learns to improve itself incrementally through reinforcement learning by avoiding previous errors. This allows AlphaGo the ability to improve its win rate against older versions of itself. After all of the simulations are complete, the latest version of AlphaGo is capable of beating the original version upwards of 90% of the time.

(Source: Go Game Guru)

Solving the game of Go became the holy grail of artificial intelligence progression. Lee Sedol is a grandmaster Go player, and is widely considered the best player over the last decade. He is the pinnacle of success in Go, and is arguably the best player of all time. DeepMind’s AlphaGo system took on the challenge of playing Lee Sedol in a widely viewed and publicized match. Amazingly, AlphaGo defeated Lee Sedol 4 games to 1 in the heavyweight match up of man versus machine. This was the most significant gaming victory for a computer system over a human entity, ever. It also marked a giant shift in gaming strategy. Instead of studying the moves of the top players, the top players began to study the moves made by AlphaGo. Some of which were so unorthodox that commentators believed the system handlers had accidentally clicked on the incorrect piece positioning. One AlphaGo move that was incorporated into its strategy was normally attempted by a human player only 1 in 10,000 possible opportunities. This means that AlphaGo was adapting its strategy to incorporate occasionally contrarian moves. These moves will help to grow and solve the game at a rate much faster than before. AlphaGo gave professional players a glimpse at an assortment of unique strategies.

Neural Networks and Their Function in the Future

Artificial Intelligence and neural network capabilities have been an intriguing topic in the high-technology community. Self-learning algorithms have the potential to make changes to society that humans have not been able to achieve alone. The neural network theoretically should be able to access an external memory in a way that imitates the short-term memory of a human brain. Algorithms that can optimize efficiency would be a huge boon for the planet. Think of all the money and resources that could be saved if there was a program that could solve sectors such as healthcare or energy. Making any major industry even 5% more efficient could significantly alter the socioeconomic landscape. Climate change has been a widely debated political issue for years. It is possible that artificial intelligence could find the best possible route of action for combating climate change. Artificial intelligence could be human’s greatest aid. By solving fundamental problems and allowing society to have technology and structure that may have taken humans centuries to discover alone.

(Source: Linkedin: Artificial Intelligence)

Artificial intelligence is already changing the world in incredible ways. We are on the verge of an information breakthrough that will alter society throughout all sections of life. Google’s DeepMind is at the forefront of this technological revolution. Video games have become the vessel for algorithmic performance testing. It is amazing to think that these potential technological leaps could have been initiated through testing on publicly made video games. Also, it is relieving that DeepMind continues to try to apply their technology for a positive impact. They have established an A.I. ethics board that will analyze problematic factors regarding their technology. It would be easy for the company to focus solely on making a profit. As constructed, DeepMind can turn a profit while also considering the potential problems for civilization that could spawn from their technology. Artificial intelligence is in great hands with Google’s DeepMind.

I Know First’s Implementation of Artificial Intelligence Technology

AI technology has started its worldwide industry takeover. Consequently, this has created a mandate for more AI based methods within the financial sector. Researchers have been focused on developing innovative techniques for financial market prediction and forecasting. Founded in 2011, I Know First was one of the first FinTech companies to implement artificial intelligence with deep learning neural networks with the function of market analysis and forecasting. I Know First developed a prediction system that uses artificial neural networks that are self-learning, flexible, and adaptive to the capital markets. The algorithm features a Decision Support System (DSS) to optimize the information produced by the years of data inputted. Each day, as the market changes, the system uses the new data inputs to learn from past data information and adjust the weighting of the hidden nodes to improve its predictive performance.

Understanding stock market predictions using artificial neural networks and their adaptations was analyzed further by Tali Soroker, a Financial Analyst at I Know First. In the article, she explains the different type of ANNs (Artificial Neural Networks) and how they are applied to a particular industry. I Know First is using artificial intelligence technology to exploit the stock market. They provide consumers with quantitative information that will help them make concise and profitable trading decisions. The financial industry is experiencing a technological overhaul, and I Know First is one of the leaders of the economic transition.

Google’s DeepMind: Self Learning A.I.

“A victory of a program over a human in the ancient board game Go has sparked intrigue and in some cases concern. It shows that a machine has approximated human intuition and outsmarted the best human brain in the game. It is something that scientists had not expected to happen for at least another decade. And it is a giant leap for artificial intelligence, showing that machines can learn on their own,”- Ali Mustafa, TRT World.

Below is a ColdFusion video explaining the background and application of Google’s DeepMind: