Digital Engineering magazine surveyed its readers late last year about the future of design and came up with a slew of interesting findings. You can read their summary of the overall results here, but I’d like to focus on just one aspect: When answering the question, “Which technologies do you think will have the biggest impact on product design and development over the next five years?” 47% of respondents checked “Artificial Intelligence (AI) and Machine Learning” — and 47% checked “Simulation Software”. Both at 47%. That’s slightly ahead of additive and cloud and a lot ahead of IoT, virtual and augmented reality, and lots of other things. Funnily enough, generative design (which uses CAE and could leverage AI) was 24%. Check the source material for details; let’s explore the AI and CAE findings.

Before we go any further, whenever we look at survey results, we need to understand how many respondents there were, how broadly distributed they are, and if they’re likely to know what they’re talking about. Digital Engineering said it received more than 330 responses from design engineers, engineering managers, and others; no one demographic dominated. Nor did any industry. So, while not a huge sample, these results represent people who do design rather than talk about it.

And they see the potential of AI and CAE as similar over the next five years. Huh.

To me, CAE and its benefits are known values and have been for decades. CAE replaces physical tests and their cost and time lag. It enables exploration of alternative designs to see what is most fit-for-purpose, potentially far earlier in the design cycle than might otherwise be possible. All we need to start with CAE is a design concept, an understanding of what we want the design to do and the forces it needs to withstand, software, and (maybe) hardware. After a while, we get results that show us what steps to take to move forward in our design.

On the other hand, AI is still essentially an unknown quantity in the world of design and engineering. We know that we can train machine learning algorithms with real-world data and simulation results to understand product behavior and save on test and simulation cycles, but that’s not at all common today (though that will change as vendors work to tie these technologies together). We need to understand the inputs we need and interpret the outputs. We might need to gather and sanitize a lot of data points. All of this will get easier over time; we’re not there yet.

Perhaps the survey respondents who are excited by ML’s potential are thinking of ML-backed CAD solutions, changing how designers work. For example, the ML engine monitors what the user is doing and offers priority access to the tool palette or options likely to be used next after a specific operation: “Monica usually messes with the fillet after creating it, so we’ll show her that set of options next”. And then, during my next fillet operation, it tells me how to do it right the first time. Or perhaps the people surveyed are looking for ML-based training, where the CAD product captures the user’s level of expertise and then offers suggestions to boost effectiveness. All of the vendors are working on this, and some implementations are already on the market.

That said, making a gnarly CAD product sing has been a point of pride for decades. Model-building contests and benchmarks showcase expertise, and many people don’t want that badge to disappear. This type of ML may not be for you … (I believe you can usually turn it off if it’s bugging you.)

There’s also the more significant issue of “I don’t want my boss to know / don’t want to be monitored this way” … Yes. There is a price to pay for this background assistance. Who has access to this data, what they will do with it, how long they keep it, are specific humans identified, are all questions to be answered.

Applying AI/ML in the design and engineering realm in other ways is still largely unexplored. What do we want an ML algorithm to tell us, that’s relevant for design? We could incorporate more data into design decisions, like supplier on-time performance, that guides us to/away from specific suppliers. Or an algorithm could calculate all of the possible permutations of component cost, logistics, and eventual part disposal to develop a weighted sustainability score. There are so many opportunities for this technology, but I haven’t seen any one concept emerge as a definite must-have.

Perhaps the answer to the “Really? Simulation and ML tied for impact” conundrum isn’t all that complicated. Most designers and engineers know about simulation, its capabilities, weaknesses, and how it fits into their workflows. It’s a known and, therefore, more easily overlooked. There’s been so much hype about artificial intelligence and machine learning that it’s like the hot new toy: people don’t really know what it is but might want it anyway. Seeing the two technologies as equally beneficial is part hopes and dreams and part reality.