The past seven days have seen the construction industry erupt in heated discussions concerning the use of artificial intelligence (AI) for site monitoring applications. This is a complex subject - here’s my two cents on the matter. The Construction Index news article entitled ‘Eye-in-the-sky to police construction sites’ sparked an industry outcry (in their words). The article tells the story of a new software product (EasyFlow CV by Agmis) that uses Computer Vision to ‘automatically detect safety violations or any decline in productivity’.

Before we dive in, what is computer vision? ‘Computer Vision’ (CV) refers to the process of training a computer to make sense of pictures and video. It’s the artificial intelligence technology that auto-tags you and your mates in Facebook photos, and keeps your (fictional) Tesla on the road.

We’re beginning to see more CV applications within construction, so it’s about time we had a proper conversation about how to and how not to deploy such technology. What are the uses cases? What are the boundaries? What are the ethical considerations? How can avoid getting it wrong again in future?

So, why the outcry?

• All about Trust: The best, most productive sites I’ve worked on had three themes: openness, trust and great communication. This specific application of CV technology sits these three themes on the back of a grab lorry, buries them under 3 tonnes of London Clay and ships them down the A13 to Wallasea Island.
• It’s a People game: The construction industry puts food on the table for over 2 million people in the UK. A primary role of construction tech should be to make their lives easier. To augment there human intelligence, not insult it. As construction software providers, we need to plan meticulously how the tech we produce can be woven seamlessly into this existing fabric with minimal disruption, and it’s clear that this software release has fallen short of this.
• When all you have is a Hammer, everything looks like a Nail: The safety of our workforce is our biggest priority in construction. However, is this the winning solution? I’m not convinced. We know Computer Vision technology is powerful, so I ask: is flagging PPE really the best use-case for this technology we can think of? Again, I’m not convinced.
• Double standards? In fact, I wasn’t convinced back in 2017, when Autodesk invested heavily in smartvid.io for pretty similar technology. Interestingly, I remember the reaction from the industry being very different back then. I seem to remember smartvid.io being heralded as AI pioneers and safety leaders in the industry. So where was our collective ethical conscience back then? Or is it that we are only starting to think through the repercussions?

So, what can be done about it?

This is not just a mild rant. I’d like to end by offering my thoughts on how I think Computer Vision could be applied to bring benefit to our industry:

• Monitoring things on site, not people: Rather than monitoring the person building a wall, can we not just monitor the wall itself to understand progress? It’s then easy to compare the imagery from site to our construction BIM model. Note: Things in this context could mean mapping built assets from imagery (see SenSat’s Mapp software for great work in this space), plant and equipment movements (see indus.ai’s product for this), or monitoring defects (see the Andreessen Horowitz-backed doxel.ai).
• Educate and inform the workforce: It’s fair to say the average construction worker is not expected to keep up-to-date with Hinton’s latest NIPS paper. If we are serious about wanting to apply AI tech on our sites, we need to have a conversation with our workers. Make it explicit what the technology can do, and categorically rule out what it will NOT be doing. Otherwise we run the risk of letting their imaginations run wild, based on their AI depictions from the media and sci-fi films.
• Think of better CV solutions: to me, the PPE example is lazy. We need to get creative with how we think about construction, and how we apply AI technologies to it.

Call to action: Enough of my views, I’d really like to get other people’s views on this. If you liked the post - please share it! Do you agree with what I’ve said? Do you see it differently? If so I’d love to hear from you - leave a comment below and let’s begin the conversation.

Originally posted here