Artificial Intelligence remains an emerging field. It has entered the mainstream consciousness through the abstract discussions and concerns among folks like Elon Musk and Mark Zuckerberg regarding future risks, advertisements that put forward an image of AI capability that evokes science fiction, and the increasing rate of movies exploring AI in robots that often end badly for their human progenitors. All of this has resulted in a flawed public perception of the actual state of AI. The following briefly summarizes Aegis’ view on the current technical capability and application value of AI, and then where Aegis thinks this capability can have practical application in manufacturing.
At present, AI is not ‘thinking.’ It cannot simulate or approximate the human capacity for intuition, inference, creation. In reality, it is not yet even close. It likely will be at some time, and we need to think toward that and plan for that, but to envision current applications predicated on a false belief in actual cognitive ability in AI is wasteful and unproductive. So what is AI good for? One need only look slightly below the surface of even the marketing for AI applications one might see on television. Finding snow leopards on a mountain face against which they are nearly perfectly camouflaged—finding the pattern within an archeological site image that would suggest where something of interest is located. These image processing applications are ideal for AI—because AI is good at finding patterns within images that, while humans also can do it, we find it tedious, slow, and are sloppy at it. AI algorithms don’t bore, or degrade in their diligence and are even better than humans at reliably finding pattern matches at great speed in huge volumes of images. This pattern-matching capability is a great application for AI.
Other current excellent applications of AI involve speech recognition, text-to-speech, and noise detection/cancelation. Once taught and given parameters, AI is excellent at evolving automatically to cancel noise from content more fluidly, and refining the way it understands a speaker, and how it might speak back.
The above examples, while not exhaustive, have commonality. Whether sound waves or raster images, AI excels at learning patterns, automatically evolving its comparative patterns and making matches and suggestions. AI is differentiated from traditional algorithms in its ability to self-learn. In the snow leopard example, it would be possible to write a traditional algorithm that finds snow leopards, of course. But for this algorithm to improve, coders would have to continually refine the algorithm, and publish new versions regularly. The AI version gets better on its own because it learns and continuously refines the algorithm itself. AI is also adept at high-speed acts of tedium humans loathe, but this learning also enables it to become better at it--on its own--over time. The more it does, the better it becomes, so to speak. It is in this area of detecting anomalies or changes in patterns, becoming better at it with greater use and exposure to the problem, and pointing them out that AI has current and powerful potential.
This pattern recognition and mapping extend to multidimensional, large volumes of patterns within data sets. Humans are woefully lacking in this area. Imagine extending the snow leopard search, not to one mountain but thousands, then add a third dimension to the data set of each.
That brings us to practical applications in manufacturing. Aegis entered this exploration from the perspective of what is lacking in MES/MOM and the digitization of factories, rather than from what AI can do or cannot do. Another element to the consideration was the fundamental tenet of Industry 4.0, which is a fully automatically adapting factory environment. This scenario led Aegis to believe that, ironically, it is factory digitization that has opened an opportunity—or more accurately, a need—for AI. IIoT and factory digitization has caused a massive increase in data sets available about everything happening in production. The precise real-time nature of each detail of production, materials, quality, operator actions, test data, throughput, errors, etc. is all entering giant data reservoirs. So, the question becomes, we have this data why are we not seeing the grand benefits of Industry 4.0? The answer is simple—we aren’t doing anything with the data because there is too much for any human to make sense of it quickly enough and to derive any actions as a result of it.
This data gap is where Aegis believes AI will add value. The application of AI to continuously monitor and learn from the massive incoming data about production and do a few things with that to assist its human colleagues. First, its value comes in detecting areas for improvement to the processes by revealing causal relationships that humans could never detect. Detecting multidimensional cause/effect from factors not just on an assembly line but also from incoming materials, out to shipment, and back to design. This is a massive data set continuously in flux—which is where AI excels at establishing baseline patterns and then the deviations from it. This would enable AI to reveal opportunities for continuous improvement, to ulimately direct human colleagues to focus their investigation and apply their creativity, etc.
Another potential value is in the area of scheduling and simulation. Once the AI has developed its baseline pattern of ‘normal’ or even ‘ideal’ operations, it is in a great position to apply that to ‘what if’ scenarios to help its human colleagues to see the impact of a sudden schedule change. However, it would not be doing this in the current algorithmic way, but more based on reality borne of extremely thorough historical patterns as well as the reality from production at the moment. This could usher in true, instantaneous production adaptation to nearly continuous flux in demand, materials supply, and product types.
In the end, all of the envisioned applications are using AI to augment human shortcomings in the areas of pattern detection and the analysis of massive data sets in real-time to both see patterns, see changes to baseline patterns, and reveal the cause/effect of changing a pattern that is already understood.
Sign up for our blog
Stay up-to-date on the latest in manufacturing trends, insights and best practices.