How Predictable—The Second Life of AI
Artificial Intelligence, or AI, has a long history. Early research in the field was done even before the advent of electronic computing, and some of the first attempts to commercialize AI were made in the 1980s. The technology is seeing a renaissance of late, although today it is more likely to be referred to as machine learning or deep learning. Some of the largest technology providers are providing branded technology platforms and applications (IBM Watson, Salesforce Einstein, multiple Google Cloud Platform products) and consumer services (Apple Siri, Amazon Alexa).
In this article, I’ll focus on the role of AI in the lifecycle of manufacturing and construction. Before I dive into the benefits of AI for these areas, however, let’s discuss some of AI’s limitations.
Previous Experience Needed
2016 has shown two clear examples of the failure of predictive models—the Brexit vote and the US Presidential election. In both cases, poll data was interpreted based upon models constructed using past voter behavior, and these models were used to predict the outcomes. We all know how that turned out. But what does that have to do with AI? Both traditional models and AI depend upon prior experience to interpret new situations.
AI platforms use a variety of methods to sample and filter data, fuse (combine) data from multiple sources, make inferences, predict ranges of possible outcomes and their probabilities, and ultimately make decisions or recommendations. The process of developing AI models, whether through supervised training or unsupervised “deep learning”, depends upon access to data that represents the behavior of a system under a wide variety of conditions. The AI model must also know how to characterize outcomes (i.e., learning “good” from “bad”).
Just as with the recent failures of electoral predictions, AI algorithms can go astray. So long as new situations correlate with patterns observed from past experience, a well-trained AI model can do a remarkably good job of developing inferences and recommendations. In some cases, AI can even produce novel insights that would be difficult to uncover through other means—the technology is that impressive. Unfortunately, AI isn’t always adaptable to new patterns. When major shifts take place that move beyond the realm of experience, the results from AI become much less reliable. To be fair, many humans have trouble with novel situations as well.
Artificial Intelligence, Real Results
The word “artificial” can carry negative connotations, notably a sense that whatever it is, it’s not “real.” In the case of AI as applied to manufacturing and construction, that would be a mistake. A capable AI model can digest far more data than even the smartest human, and can work tirelessly to develop diagnoses or solutions that would be difficult, if not impossible, to derive any other way. While the technology is only now starting to be applied to design, manufacturing, and construction—and some of the work is still exploratory—some opportunities are becoming clear.
One area of opportunity is optimization. AI is well suited to computational design problems. Emerging solutions provide the ability to evaluate different design and configuration options to identify one or more “best” solutions, considering such factors as design layout, structural optimization, and material cost. In construction, at least one commercial solution already exists to apply AI to the problem of construction planning. Perhaps the richest opportunity is in operations and maintenance, however. AI has the potential to diagnose incipient failures before they occur, and to provide service technicians with clues to confirm diagnoses. Similarly, AI is well suited to optimizing maintenance based upon predictive metrics.
Get Smart
Given the potential of AI, it would be a mistake to ignore the technology. But where to start? The answer will depend on your specific business needs, but there are some good guidelines to help you identify the best opportunities. First, consider the problems or opportunities that have the greatest potential impact to your business today. Do you have a good understanding of what the desired outcomes are, and can you quantify them? Do you have extensive data (structured or unstructured) that can be used to characterize prior experience? The best opportunities are those with significant business value, clearly articulated desired outcomes, and extensive underlying data.
If you’re able to identify a project that meets these criteria, the next step is to work with an independent expert to outline your strategy and identify potential solution providers to interview. Consider starting small, with a pilot project that allows you to become familiar with the technology and project cycles. This experience will be invaluable as you target higher-value opportunities.
CEO at Falc1 Tech.
8 年dear Ed, we already got it. Finally we already got it . First of all thank for your article, very interesting and, for this reason, I would take some minutes(fortunately here in Italy it is night time) and try giving you a different side of the question. I would start from your sentence; in the last year Twe, my society, is testing one IT application which reset to zero ( yes reset to zero with no possibility of failures, and I consider it an added value) the design of the tooling clamps for gear's shaving operation. When I say reset to zero, I declare that this application develop the mechanical system much more faster then the timing necessary to open the same system already realized. Even if ReCreate (IT Platform name) is amazing, I consider it just the final result of the way that I start twenty five years ago; it is useful to demonstrate that all together we can change radically the direction, we have to change it, we need it. Now my starting consideration (and questions for you). Is it moral that thousand and thousand of designers in all over the world they design the same components for all their work life? Is it sane that a designer work like a machine (repetition) for a machine? Isn't would it be useful for all of them to value their activities with higher activities which allows the correct use of their experience? Yes, because just as the Human Intelligence is the capability to use the experience in the "better way", so the Artificial Intelligence (simplifying extremely the concept) is a big database containing abstract "dynamic & plastic" informations capable to renew themself in innovative concepts. It is just question of time (in my opinion no more of a couple of years) and the world will radically change; no more IT experts which prepare Applications for the market, but the market itself will create the toolings containing design ( for example) specifications. There is a lot to say... I would say that AI is not based only on predict or experience but primary it is the integration of all the instruments that contribute to get the target; brexit or Trump were easily predictable if aside of the history there had been also a "termometer" of the population. For the moment I thank you very much for your time and I wish to go on this argument with everybody want to discuss it with us. SORRY FOR MY POOR ENGLISH; I WILL IMPROVE IT AS SOON AS POSSIBLE. PROMISE Best regards Antonio