by Michael Schwartz
The next three technologies that will change the world as we know it will be artificial intelligence (AI) running in the cloud, working with the Internet of Things (IoT). Yes, that is quite the mouthful of technology and the basis for many science fiction films. But the reality of the situation is, these technologies are here to stay. So the question is: How can metrologists take advantage of them?
First, I would like to point out many of you have used these three technologies and didn’t even know it. For example, you use Google Maps when you need to get from point A to point B. Just type the address into Google Maps and press Start Navigation. On the back-end, the AI algorithms are looking at all the possible routes and choosing the fastest one. The fastest route is being continually updated based on the data from all the people traveling that day. Your phone is telling Google’s Cloud database about the traffic at that time. Then that cloud database builds both real-time traffic information, as well as creating some traffic patterns. Your phone is an IoT device, updating the cloud, so the AI can give you driving directions.
Another implementation is voice recognition software and Google Translate. I have personally witnessed this technology improve over the years. Six years ago, when I first went to Turkey on a business trip, I tried to use Google translate—epic fail. My colleagues told me about another app called Tureng Dictionary that would get the word right more often than not, but it could only do one or two words at a time. When their accreditation body sent me some documents in Turkish, I needed whole sentences translated, so I tried Google Translate again. I was amazed how much it had improved, so this last trip I used it and the voice recognition to order a sandwich at Subway. It worked both ways as we had a two way conversation about how to make me sandwich.
So how did the programmers make so much improvement in the technology in such a short amount of time? Truth be told, they gave the AI more data. Its accuracy is based on having millions of documents where it learns grammar, rules, and sentence structure. Then the AI uses statistical analysis, rather than rule based algorithms. It gets smarter the more data you give it. Then it figures out how to translate a sentence it has never seen.
The next phase of the IoT is autonomous control of systems. This will have a great effect on metrology. Writing automation software for the past 30 years has taught me one thing: Computers don’t know good measurements for bad ones—and autonomous systems will be plagued with this problem. AIs will have to play a part in evaluating measurements; they will eventually learn how to best make a measurement.
Let’s take three measurement approaches. The first is simple, take 5 measurements and average them. The second one is to take 5 measurements, throwing away the high and low measurement and then average the other three. And finally my favorite, keep taking measurements until the system stabilizes, then use that measurement. Each method has its advantages and disadvantages. The first two produce great results as long as the measurement time is sufficient, whereas my favorite method has a tendency to increase the total measurement time and can result in several minutes to take a measurement on something like a high resistance test.
Now, if we were to switch from a rule based measurement methodology (i.e., pick one of the three and code it) to a statistical analysis method of measurement, much like Google Translate, the AI could then be optimized for measurement speed or measurement accuracy. An AI with data about measurements could learn how to best make a measurement achieving the uncertainties required. The measurement methods could be thought of as needing x number of samples over x amount of time when using x device to measure a value. The AI would be able to learn from its data and past measurements, and update the measurement requirements and instrument settings.
Over time, the AIs could also help us discover dogs and gems—what instruments are making better, more reliable measurements and what instruments can’t be trusted much. It could also help us discover new things like contributors to our measurement inaccuracies. Large populations of measurement data could be correlated with other data to discover new contributors to our measurement uncertainties: a butterfly flaps its wings in New Mexico and my measurements are off by 5.6 ppm.
As I try to position Cal Lab Solutions for the future, I think about the effects of these technologies on metrology. What can I do to insure our technologies are not obsolete the same day they are introduced to the market? What I learned from the book Technopoly, by Neil Postman, the best we can do is speculate on the future and technologies. My idea about AI’s effects is just a vision of an unknown future.
May 21, 2018 Correction: The spelling of “Tureng” was corrected and link added.
