Artificial Intelligence (AI) is a packed term that has evolved over the past 40 years. In fact, if you Google ‘AI’ an amazing 4.4 Billion results will be generated in 0.61 seconds – showing the vast scope of AI - and also AI in action by fast tracking the search results.
First up, let’s quickly define the different terms that often get used, sometimes interchangeable, when discussing AI but have important nuances between them. We’ve used the analogy of a car and driving to help simplify these technical terms.
Defining Terms in AI
Artificial Intelligence (AI)
At a simplistic level, AI helps machines to perform tasks in intelligent ways by having the ability to adapt to different situations and new data. Examples in today’s modern cars will be many of the driver assist functions, such as emergency braking, cross-traffic detectors, blind-spot monitoring, and driver-assist steering to help avoid accidents.
Machine Learning (ML)
ML is seen as a subset of AI and enables machines to process data and learn on their own, without constant supervision. However, if the machines return wrong results, a programmer would need to step in to adjust the code. An example here will be your navigation system, whereby it will intelligently map out traffic patterns, to better predict arrival times and help you to avoid traffic jams.
What is Deep Learning (DL)
DL is a subset of machine learning using neural networks, which requires more data and training time than most ML approaches. Although Deep Learning takes more data and computation, it is better at extracting subtle patterns, such as non-linear relationships. An example of this is how autonomous cars detect pedestrians and street signs using image detection of extract patterns like outlines of people, or the octagonal shape of a stop sign.
What is Reinforcement Learning
RL is based on dynamic programming that trains algorithms using a system of reward and punishment. A good example is an autonomous driving car, which interacts with its environment and receives a reward state depending on how it performs. In this example, driving without intervention will earn it a reward; but if the driver needs to intervene by braking or changing lanes, then the agent receives a penalty. Reinforcement Learning is not explicitly told how to perform a task but works through the problem on its own by learning through its environment.
So now we have a high-level definition of AI, ML, DL, and RL, let’s go through a few examples of what AI is not.
A Black Box of Magic
You heard it here folks! AI cannot magically conjure up an unexplainable solution out of thin air, but instead combines math and statistics with theory and human subject matter expertise to refine the model before the true value of AI can be achieved.
What is depicted in iRobot, The Terminator or The Matrix
In the same vein, this will not lead what’s known as “Artificial General Intelligence” or AGI, where the advancement of super- intelligent robots will exceed a human ability or extinguish the human race. Even with the promise of Reinforcement Learning, applied AI is considered very narrow. This means applied AI can perform well at the task it was designed for, but not well at other tasks. For example, AI used in car navigation is not good at identifying pedestrians. Instead, autonomous cars are comprised of many narrow AI solutions that are brought together by the many people who design the system.
In the industrial sector, AI is shepherding in a new era where industrial operations can move from situational/post-production awareness to comprehension and prediction. And most importantly – insights come in near real-time so that the appropriate action is taken before production is impacted and operators are empowered to continuously measure and improve their performance. It’s important to note that implementing AI into your operations is not a set and forget project, but instead should be viewed as a journey that continuously optimizing and improves as it learns and adjusts to environmental factors.