What AI Is… And Isn’t

To make the best use of AI, understand its limitations.


AI is very good at some things. It may never be good at others. The challenge is figuring out where it can help the most, and then making the cost calculation for how it can be applied.

Cost sounds like it should be fairly straightforward, but it isn’t. For instance, what is the cost of continuing to doing something the same way if you aren’t making necessary changes? Delaying those changes can be more costly in the long run, so a decision to make necessary changes can be beneficial.

In this sense, AI can be a disrupter. It can replace continuous improvement, which is what the semiconductor industry does very well—sometimes too well. Continuous improvement has allowed us to build chips at new nodes with twice the density for close to the same price. And it has allowed us to improve reliability across entire systems, regardless of what they are used for.

We see this demand from the supply chain. Our suppliers have moved from 5 parts per million impurities to 5 parts per billion, which may not even exist in the real world. What exactly that number means isn’t clear. It’s theoretical. At some point, they may move to 5 parts per trillion, which is another order of magnitude removed from reality.

And this is where AI really shines. It allows us to look at our processes statistically, digesting large quantities of data and making sense of where problems creep into the process. Not everything requires change, and AI helps pinpoint what does need to be fixed and how to fix it. We’re looking for anomalies in statistical patterns, and there is no better or more efficient way to find them.

This is just the start, too. Imagine what can happen if some of this “intelligence” is built into materials themselves. Those materials can generate data that is sent out to a server. If something goes wrong, those materials will send an alert before it becomes a problem.

Think of this like a bridge. A crack inside one of the bridge supports cannot be observed until the crack reaches the outer surface of the support. But with smart materials, that data can be relayed well before it reaches crisis stage. That same strategy can applied to a car, a medical device, an industrial control system, or anything else where the integrity of a material is critical. And on the receiving end of those signals, AI can determine when something needs to be replaced based upon the data received and the weights applied to those algorithms.

On the other hand, AI will not replace the people or the cognitive skills of the people working with that data. An AI system isn’t an artificial brain. We don’t understand how the human brain works today, and it’s possible we never will. It takes the collective effort of the entire semiconductor industry to be able to identify and reduce simple defects in a process. Understanding why that’s good or necessary, and utilizing years of expertise to predict the future, is a lot harder than measuring movement or temperature shifts or signal throughput.

A machine can be trained to differentiate a cat from a dog. And it can create statistical models for unbelievably low contamination of materials. What it cannot do is tell us why we should do that.

In conclusion, we still do not know exactly what role AI will play in the future. As we learn more and more about it, we are beginning to see that AI will not take over a majority of the jobs like the movies AI and I, Robot demonstrated, but rather it will aid people in their decision-making. AI will be an enabler for quicker and more fact-based decision-making, and a more precise decision maker.

Leave a Reply

(Note: This name will be displayed publicly)