By Pat Dixon, PE, PMP
President of DPAS, (DPAS-INC.com)
In 2023 during a particularly demanding day on the Texas electrical grid, a bitcoin mining operation in Rockdale was told to shut down in exchange for $31.7 million.
At first glance this may not appear to be pertinent to the subject matter of this newsletter, but this incident is an attribute of the 4th industrial era. In this era of big data and server farms hosting applications requiring intensive processing, it is easy to assume there are no limits. The hosting and energy costs seem cheap until they are scarce.
Artificial Intelligence may be the biggest buzzword in industry. Data centers are filling up with AI applications for a myriad of objectives. The very nature of AI gives the impression that you can load it, and it will automatically make things better.
In the same way bitcoin miners demand computing resources and energy, AI applications can eat up demand. Beyond the resource impact, there are other reasons to be judicious in deploying AI in industry.
We need to begin with what I have stated so many times that you may be tired of hearing it, but a castle built on a faulty foundation will crumble. Many facilities are so far behind in their base infrastructure and maintenance capacity that any AI that sits on top of it will fail. Instrumentation will produce garbage data without sufficient maintenance, which will render any AI application useless or dangerous. Systems relying on eBay for spare parts, multiple layers of gateway bottlenecks, and insufficient network security are vulnerable to failures that will take AI applications offline. A system needs to be solid under 3rd industrial era criteria before it is ready for the 4th era.
Assuming you have a solid foundation, before investing in AI we need to know what it is. It is a term like many others that can be so broadly and randomly applied that it can be hard to know what we are talking about. Some consider AI to be anything that replaces or behaves like a human. A human can watch a flow gauge on a pipe and manually throttle a hand valve to set flow at the desired rate. Does that mean every PID loop is AI? Perhaps a better definition is an application that can adapt its model (learn) based on continuous realtime data. That definition often applies to Machine Learning, which is a subset of AI. There is ambiguity in defining AI, so it is important to know how an application works before investing in it.
Assuming there is clarity on the AI attributes, it needs to be acknowledged that AI is in its infancy in industry. While applications like ChatGPT are mind blowing, they are not applicable for deterministic industrial automation. Some industrial AI applications do not require determinism, such as an open loop machine learning prediction model. The point is that first principles and well-established algorithms should not be dismissed with the notion AI can do anything. In the same way a baby can't drive a car, AI is not mature enough for every industrial application.
Now we come to the bitcoin mining analogy. While a lot of AI applications can and should be locally deployed with an on-premises server, it is very enticing to deploy in the cloud. Especially for big data applications, the cloud is the ideal architecture for many reasons. However, the growth in server farms can outstrip the growth in energy capacity. The 4th industrial era coincides with an era of mandates to cut fossil fuel usage. While AI applications may not have the resource demand of bitcoin mining, there can come a time where cloud resources can become scarce for any application.
Some have fears of AI and others enthusiastically embrace it. For those that are bold there is a high ceiling of unknown potential. Even for the bold, caution will serve you well by considering the risks described above.
