In this special guest feature, Eliya Elon, Director of Product and Business Development at Razor Labs, discusses a new technology that is starting to trickle from the purely theoretical academic world to the business world, one that aligns with your company objectives, that draws a clear line between your business question and the insights generated. This new technology is called Deep Reinforcement-Learning, and it is gaining significant success in different use-cases. Eliya is the VP of Product and strategic partnerships at Razor Labs. He is an experienced tech entrepreneur, selling his last AI company in 2017. Since joining Razor Labs he is focused on creating AI products that bridge the multi-dimensional gap of business needs, user experience, and academic research, hopefully at scale. He is excited by human psychology and new technology, but what excites him the most is where they meet. That is where the black magic happens.
You have invested hundreds-of-thousands of dollars on analytics software, you believed the ROI would be right around the corner, you were promised “Simplifying business analytics for complex data” or “Answer questions as fast as you can think of them,” but in reality, insights we’re non-related or minor at best.
The Data Analytics market is overwhelmed with forecasting solutions that perform predictions based on uncorrelated data. The problem with those solutions is that the questions they answer are hardly related to the business challenges. Challenges the organization is facing, making their ROI very hard to define or measure.
Many organizations, commercial and governmental alike suffer from this issue, where current analytics tools have no clear ROI for their end-users. For context let’s look at an example. Suppose you are a coal mining company, the question that current solutions answer can be “What will be the coal throughput tomorrow?”, when the business question for the coal mining company is “How can I optimize coal throughput?”, so I can maximize profits or reduce resources usage.
A new technology is starting to trickle from the purely theoretical academic world to the business world, one that aligns with your company objectives, that draws a clear line between your business question and the insights generated. This new technology is called Deep Reinforcement-Learning (here on DRL), and it is gaining significant success in different use-cases. In essence, DRL is a form of Machine Learning and is the new promising “tool” in the AI revolution.
These technologies are being used to recommend specific actions that easily solve complex business problems, keeping the whole system continuously optimized. Deep-Learning for instance is a “type” of AI that is special in its resemblance to the human biology of neurons in the brain.
As we said earlier, Deep Learning algorithms are inspired by the human neurons in our brain, allowing for pattern recognition and learning of complex patterns based on raw data and systems that were impossible until recent years, but Deep Reinforcement-Learning (DRL) uses an even more interesting – and very human – approach: reward and punishment. The technology agent (An entity that can relate and act on its surroundings) has one goal in “life,” to maximize its “score” on a task. When a behavior that increases the score occurs, the agent learns that this is a positive “pattern” and vice-versa. You must be saying to yourself – “this sounds much like a child playing a game.” Moreover, you are entirely correct. This analogy is helpful.
Imagine DataMind as a Super-Mario game where the business process is the game “world.” Allowed operations in the game, as moving right/left and jumping follows operations permitted in the business process, such as changing input parameters in a manufacturing process. Also, the physics of the game mimics the physics of the process – the speed of movement, the height of jumping and so on.
If you were back at your 10-year-old body, playing Super-Mario for the first time, you would start to experiment around, “discovering” your new virtual environment. With time you would learn that some behaviors are better and result in a higher score. The only limitation to become “best” is time, what we achieve with DRL agents is virtually removing that limitation, let us see how.
What if you could model your specific business process with high accuracy and low bias, to the digital world (Digital twin concept), effectively removing the intuitive concept of, experience as a function of time? DRL coupled with strong parallel computation and significant raw data achieves precisely that. The agent constructs a digital model by consuming all data created by the existing business process, using data as a single source of truth reduces human bias to zero, allowing for a much more accurate model. Then the exciting part happens, you let the DRL part of the system “play” on your model for millions of iterations, getting better and better every time, learning the hidden subtleties of your use-case. You gain the insight of hundreds of human years in a matter of days. That insight comes in the form of a simple, readable and interpretable output – “Increase input Y by 20%”, “Reduce temperature by 3 degrees.”.
Deep Learning technologies are the core of the modern AI revolution impacting every industry in amazing rates, these group of technologies and algorithms are so effective and Deep Reinforcement-Learning is taking place as a new technology that will impact business processes that were considered until this day “highly intuitive” – Driving a car, playing golf or becoming best in the world at GO.
These algorithms are starting to prove very useful in the industry as well. For example, for the Oil & Gas industry similar techniques have resulted in 50%-75% more accurate anomaly detection and 15% reduction in full stream OPEX expenditure.
As Razor-Labs Head of Product I have seen first first-hand, how companies have gained up to 40% increase in throughput, this is an unprecedented improvement, based on data utilization and algorithms entirely. I am excited by the new use cases that are still awaiting. More then that I am excited by the rapid dynamics of what we consider intuitive and human to be and I can’t wait to see how these concepts evolve as the human-machine interface expands
Please, feel free to leave a comment below and share your thoughts about this piece, this is just a conversation starter…
Sign up for the free insideBIGDATA newsletter.
Machine learning and expert systems have grown dramatically in complexity and capability over the past three decades. “AI” has become a substantial, mainstream discipline that is creating real value, even though its name does not have much to do with what it really is. The awesome computing power at our fingertips provides the means to accomplish amazing things. As the CBINSIGHTs’ State of AI 2018 report notes, there is still no evidence of general intelligence in any computer applications as of 2018. Will there ever be? It seems unlikely that computers will ever demonstrate real intelligence, but it would be foolish to say that it never could happen. It will become increasingly difficult to distinguish between computer generated results and real intelligence. However, there is no evidence of any real intelligence in computers today and no indication that traditional intelligence will be seen in the future. It is still just a dream. Beam me up Scotty! There is no sign of intelligent computers down here.