Nature has given human beings an amazing ability to learn. We learn complex tasks, like language and image recognition from birth and continue throughout our lives to modify and build upon these first learning experiences. It seems natural then, to use the concept of learning, building up knowledge and being able to model and predict outcomes and apply that to computer related processes and tasks. The terminology used to describe the technologies involved in this paradigm in computing are Artificial Intelligence (AI).
It’s Just a Game
In the late 90s, a defining moment in the world of artificial intelligence happened. In 1996 chess master Garry Kasparov played IBM’s Deep Blue, originally built to play chess using a parallel computer system, and won 4-2. A year later, Kasparov and Deep Blue played another match, this time, Deep Blue won. This win created a sea-change in the attitude towards the idea of AI. Chess masters minds have to perform highly complex calculations, evaluating multiple moves and strategies, on-the-fly. They can also take their own learning and apply novel moves. Being able to mimic this process, even if applied to a specific task like chess, opens up real potential for the technology.
Out of this success, new developments in AI have brought us to the point of maturity and sophistication. DeepMind, now owned by Google, uses deep learning algorithms. These algorithms are based on the same idea that allows human beings to learn, i.e. neural pathways or networks. Again, AI has been applied to gaming to prove a point. DeepMind has taken the idea of ‘human vs. machine’ and this time used it in the highly complex game of ‘Go’. DeepMind, the company, describe the game of Go as having “more possible positions in Go than there are atoms in the universe”. So then, this is the perfect challenge for an AI technology. DeepMind uses deep learning algorithms to train itself against known plays by expert players. The resultant system is known as AlphaGo and has a 99.8% win rate when pitted against other Go programs, and has recently won 4 out of 5 games against the Go pro player, Lee Sedol.
It may seem that it’s just a game being played, but in fact, this is proving the technology, showing it can learn how to model and predict outcomes in much the same way that a human being does. In almost 20 years AI is already 10 years ahead of what was anticipated of the technology. The games have proven the capability and now the technology is entering a stage of maturity where it is being applied to more real-world problem solving. Following the AlphaGo success, Google has understood the benefits of these technologies and has promptly integrated AlphaGo technology in its cloud based Google Machine Learning Platfom.
Some Definitions in the World of Artificial Intelligence
At this juncture, it is worth looking at some of the terminology and definitions of AI technology.
It can be viewed as this: Deep Learning is a sub-set of Machine Learning; Machine learning is a sub-set of Artificial Intelligence.
Artificial Intelligence: This is a general term to describe a technology that has been built to demonstrate a similar intelligence level to a human being when solving a problem. It may, or may not use biological constructs as the underlying basis for its intelligent operations. Artificial Intelligence systems typically are trained and learn from this training.
Machine Learning: In the case of the games we used earlier as examples, machine learning is trained using player moves. In learning the moves and strategies of players, the system builds up knowledge in the same way a human being would. Machine learning based systems can use very large datasets as training input, which they then use to predict outcomes. Machine learning based systems can use both classical and non-classical algorithms. One of the most valuable aspects of machine learning is the ability to adapt. Adaptive learning gives better accuracy of predictions. This, in turn, facilitates the handling of all possibilities and combinations to provide the optimal outcome from the incoming data. In the case of game playing, this results in more wins for the machine.
Deep Learning: This is a sub-set of machine learning, a type of implementation of machine learning. The typology of the system is vital; when learning, it’s not so much about ‘big’ but it’s more about the surface area or depth. More complex problems are solved by larger numbers of neurons and layers. The network is used to train a system, using known question and answers to any given problem and this creates a feedback loop. Training results in weighted outcomes, this weight being passed to the next neuron along to determine the output of that neuron – in this way, it builds up a more accurate outcome based on probabilities.
Real World Applications of AI
We’ve seen the use of AI in gaming, but what about real-world commercial applications? Whenever it comes to predict, forecast, recognize, clustering, AI is being used in a multitude of processes and systems. At Sopra Steria, for example, we use AI components in industry solutions, including banking and energy. We are integrating Natural Language Processing (NLP) and voice recognition capabilities from our partners’ solutions such as IBM Watson or Microsoft Cortana. NLP, voice recognition – and image recognition in a near future – are now widely used and integrated in a multitude of applications. For example, for banking industry, text and voice recognition are used in qualification assistants for helpdesk and customer care services. More generally, some of the best-known modern applications include everyday use in our smart phones. Voice and personal assistance technologies like Siri and Google Now brought AI into the mainstream and out of the lab, using AI and predictive analytics to answer our questions and plan our days. Siri now has a more sophisticated successor named VIV. VIV is based on self-learning algorithms and its topology is much deeper that SIRI’s more linear pathways. VIV is opening up major opportunities for developers by creating an AI platform that can be called upon for a multitude of tasks. Google recently announced a similar path to its widely acclaimed assistant Google Now becoming Google Assistant.
Machine Learning is also used in many back-end processes, such as the scoring required to allow things like bank loans and mortgages. Machine learning is used in banking to specifically offer personalization of products giving banks using this method a competitive edge.
Deep learning is being used in more complex tasks, ones where rules are fuzzier and more complex. The era of big data is providing the tools that are driving the use cases for deep learning. We can see applications of deep learning in anything related to pattern recognition, such as facial recognition systems, voice assistance and behavioral analysis for fraud prevention.
Artificial Intelligence is entering a new era with the help of more sophisticated and improved algorithms. AI is the next disruptive technology – many of Gartner’s predictions for technology into 2016 and beyond, was based on AI and machine learning. Artificial Intelligence holds the keys to those unsolvable issues, the ones we thought only human beings could do. Ultimately, even the writing of this article may one day, be done by a machine.