Over the past 60 years, Computer Scientists have relentlessly searched for ways to create intelligent machines through the application of complex rules-based algorithms. Recent advancements in Machine Learning (ML) algorithms in combination with vast quantities of training data have made wide-spread Artificial Intelligence a reality. By successfully modeling many functions of the human brain in silicon, experts believe that one-day computers will actually be able to think for themselves.
AI is a classic success story of convergence. Neuro-science, Computer Science and Robotics have all converged to create neurological models of the human brain and body. Understanding how the human brain is organized, its underlying building blocks and how it interacts is now fueling massive cognitive and perception-based innovation. Basic examples include hand-writing and voice recognition, computerized Jeopardy contestants, autonomous vehicles and devices that pass the Turing Test of the 1950s. While amazing achievements have been made, AI and Machine Learning are in their infancy and about to provide massive economic change through exponential innovation. Currently, three of the most popular Machine Learning models are:
- Reinforcement Learning
Each of these models have played a very important part in AI’s evolution. Increasingly fast graphical processing units (GPUs), Cloud Tensor Processing Units (TPUs), hyper-computer scalability, parallel-processing and massive data stores have all enabled super-human cognition and perception. The Internet and IoT technologies have enabled richer sets of training data. Interestingly, the current level of training data is nascent in comparison to the amount of data being captured everyday via Social Media, stock markets, E-Commerce and the Internet of Things (IOT).
AI’s Holy Grail
The ultimate Holy Grail of AI is the achievement of “General Purpose Intelligence”. Through the use of multi-layered algorithms that are modeled after our human brains, computers can do something very human, learn and improve their performance. The impact of General Intelligence is the ability for machines to augment worker’s cognitive power and or fully automate both repetitive and non-repetitive complex tasks. The underlying technology is based on learning algorithms (hidden layers) and historical training data to determine next best actions.
As an example, Google’s DeepMind recently demonstrated its AlphaGo machine learning capability on the board-game called Go. Go is a game where points are awarded for surrounding vacant areas on a board or capturing enemy pieces. The result was that AlphaGo beat the #1 Go Grandmasters (Lee Se-dol). For the first time, AlphaGo simultaneously demonstrated that computers can:
- Learn in an unsupervised environment
- Self-improve performance beyond human capabilities via structured feedback
- Generalize and learn other games using the same algorithms
The reason that General Purpose Intelligence is incredibly important is that it enables the rapid compounding of innovation to produce additional exponential improvements. This model is a paradigm shift because machines learns from examples, rather than being explicitly programmed for a particular purpose.
Existing Artificial Intelligence has already surpassed current human cognitive and perception capabilities in many areas. Fraud detection, oil and gas exploration and cancerous tumor detection are all areas where this super-human expertise has been demonstrated. In many areas in medicine and genomics, there is just too much data to humanly process and machines can embrace a much broader and wider data set. AI capabilities come in a wide variety of applications ranging from the military’s ability to identify tanks and missiles on the battle field to smart toys from companies such as ANKI. Human-sized robots such as Azimo can run, walk and dance and can also display intelligence and simulated emotion.
Making sure AI has a great future
The danger from combining autonomous AI applications and robotics is real. Programs that can modify their own code or reset their goal or reward functions need to be controlled at all cost as they can easily become detrimental to humans. With the recent convergence of Neuro-Science, Computer Science and Robotics, many autonomous machines are already a reality.
Hollywood had created multiple examples of useful AI machines as well as AIs that have gone extremely bad. Movies such as Terminator, X-Machina and Star Trek either evoke horror or wonder. No one has a problem with autonomous cars that can successfully navigate a congested street or robotic nurses, however AIs that eliminate jobs such as doctors, lawyers and accountants or potentially even go rogue against humanity have to be managed carefully. The next generation of AIs are likely to drastically enhance task efficiency and make many professional business processes obsolete, but AI will also give rise to many new business models and industries.
In order to protect against potential adverse effects of Artificial Intelligence, such as hidden biases and unpredictability, a Global AI ethics board needs to be formed to prevent misuse. The OpenAI Foundation funded by Elon Musk is one such group that is focusing on controlling development of AI for good. The goal is for machines to better service and empower humans, not replace or harm them.
How it Works
The first step in designing a neural network is to identify what problem you are trying to solve and then identify the data sets that can be useful in solving it. Once that has been completed, the next three steps to establish a useful Neural Network are 1) build it, 2) train it and 3) test it.
Neural networks are excellent at drawing signals out of large or noisy data sets. The secret sauce of neural networks is in the design of their hidden layers of neurons that build increasingly accurate rules based on reinforced learning. Digital input from existing data sets provide either positive or negative feedback to establish complex pattern recognition algorithms. Often times, the data can be self-generated by having the neural network play itself. Platforms such as Google’s Tensor Flow, provides an excellent Cloud-based infrastructure for neural-network development.
Modeling the Brain
Ironically, the blue-print for Neural Networks was always in our head (literally). Neurons are nature’s computing machines that when networked in layers can computationally produce powerful Machine Intelligence. By modeling and combining various processes of the Human Brain into complex chains of algorithms, computer scientists are able to replicate a new level of data awareness.
The brain is the most complex organ in our bodies. Modeling of the Brain is Silicon is still very different than true thinking machines though. The brain is an organ that serves as the center of the nervous system and is located close to the sensory organs for senses such as vision, smell and hearing. Science has been able to mimic neurons (information processing units), Dendrites (information collectors), and Axons (signal propagator) to form multi-layered neural networks. The main difference is that biological computational models are still superior in most ways to silicon based models. In addition, worker productivity will be greatly enhanced by the ability to better harness AI through new brain / computer interfaces such as Neural Lace.
In theory, if you collected all the thought patterns and experiences of a human throughout his or her life and were able to train an AI with all the circumstantial and behavioral response data, then it would be possible reproduce the mannerisms and speak patterns nearly perfectly. While this process could fool many people as to its origin, it merely replicates human behavior, but does not represent human thought, yet.
AI’s Pervasive Economic Impact
As the power of AI improves our ability to perform tasks and predict outcomes, it will cause all companies to re-think their business strategy and our overall economy. In a few short years, AI will be as pervasive as electricity and drastically alter all aspects of our society.
If your company is seeking to better understand and embrace AI technology, please contact HBSC Strategic Services. We help clients select and implement cutting-edge AI technologies to achieve competitive advantage. For more information, please visit us at www.hbsconsulting.com or call 1-800- 970-7995.