First steps to AI: data, man

Apr 11, 2019 | Hot Topics

I was speaking to a group of AEC professionals a while back about the technologies they should be investigating. I covered everything from 0D to 6D, from conceptual sketches to advanced BIM, conceptual design to operations — but, as is often the case today, we got into an involved discussion on AI, artificial intelligence.

I said that I don’t see AI as commercially ready for most implementations. Yes, you and a handful of data scientists can scour your digital universe for data, enter it into a unifying system and start creating algorithms to search it for meaning. But wait just a little while, until the end of 2019 for Autodesk’s IQ and similar commercial off-the-shelf offerings, and things become simpler and much more accessible. I’ll write more about Autodesk’s Construction IQ next week, but for now, I’d like to keep this at a higher level.

Talking about AI always leads to data. What data do we need? How do we get it? Ensure its accuracy and timeliness? Guard against incorrect data? After all, if the point is to draw conclusions from the mountain of information out there, bad data can lead us all down a rabbit hole of misleading answers.

A few weeks ago, one frustrated person in the crowd to say that her job relies on data that she often believes to be incorrect. In her case, it was GIS data–geographic information system data that combines spatial references with some sort of attribute, such as this water main is HERE–but really, it could be any data relevant to any job. And it could be incorrect for a million different reasons: It could simply be old, which means it may not represent current conditions. It could be incomplete, leaving out something that would lead to a different conclusion. It could be formatted poorly, using a synonym or acronym that the AI tool might not recognize. Or, of course, it could be maliciously incorrect, when someone tries to hide poor job performance or something else.

What to do?

I suggested that she gather that data anyway, but assign a confidence rating to it. That way, if nothing better every surfaces, she can use it with full knowledge that it may be incorrect. In her specific case, that may mean an extra day of surveying to capture accurate positional data. That update can replace the questionable data. If GIS isn’t your thing, find another way to gather confirming data, perhaps putting another sensor on a machine to check on the first, or gathering sales data in another way.

The point is this: start gathering data. Figure out what your main questions are and how you would answer them. On a construction job site, it could be overall performance. That might lead to a deeper dive on what tasks usually fall behind schedule — who, how, why, what equipment, tasks immediately before and after. On an AEC design project, it might be more successful bidding: what jobs have we won, at what price/profit, with how many design changes. On a manufacturing line, it could be quality related: after how many items are we out of compliance? In what part of the process? Is it related to material, machine, human? In retail, it could be comparing store performance or theft rates. It all starts with the question.

AI will fundamentally change all our jobs by making it easier to find the one or ten pieces of information that need our attention. But to identify those, AI engines will plow through more data than we humans ever could, looking for connections.

As we wait for the commercial solutions to hit the shelves, start thinking about this. What one problem can you solve to help your business the most? What questions will get you to those answers? What data will you need to start answering those questions? Start gathering that data now (or at least, think about how you would gather that data) so that you’re ready when the tools hit the market. If you want to be an overachiever, use a test sample of data and apply human intelligence: can you answer the question yourself with that data? If you can, you’re on the right track.

Do you remember that awesome 1967 movie, The Graduate? A party guest told a young Dustin Hoffman that the future was “just one word … plastics”. Now that word is data. “There’s a great future in data. Think about it. Will you think about it?”

Yes, all of this talk about data leads to a discussion about formats, engines and so on. Not all data will be in the “right” form for whatever AI tools will be used. That’ll be a gnarly problem, for sure, but easier to solve than not having any data to start with.

So, GO! Stop overthinking the end-game and just start.