Blaming AI is like saying the dog ate my homework

“[On] confidentiality, most of your contracts, if not all of them, will say something like: ‘You will not share your client’s information. You will not share project information in public unless you have permission to do so.’ Putting something onto ChatGPT is effectively throwing it out into the public. And someone theoretically could ask the right question to ChatGPT and get the information back.”

Turning to copyright, she said: “There are two aspects to that. One being if you have something that was produced only by the AI without a human working with the AI, no copyright attaches to it. You can’t stop anyone using it. And if you use it for a particular project, the client might be quite upset if they see it reproduced everywhere.

“There’s also the issue that you don’t know [the origin of] the information that the AI is being taught from. Most are scraping their data off the internet. You can’t say: ‘Tell me the provenance of information when I asked you to produce this document or I asked you to improve this image.’”

Can you rely on AI’s output?

Winfield then addressed the reliability of AI output: “Our industry is not great at having complete datasets. If the AI is learning from incomplete datasets, and it provides you with a response, it may not be completely accurate. And also there is the issue of hallucinations, which is where the AI completely makes something up. Saying the AI got it wrong is like saying the dog ate my homework. You can’t blame the tool. You still have to check it.”

Dave Philp, chief value officer at Cohesive and chair of the Chartered Institute of Buildings’s innovation advisory group, said: “We’ve started to look at data quality in the context of digital twinning, but when you think about AI, that adds a new dimension. How do we start talking about addressing biases presented by the data, and how it’s going to avoid perpetuating or amplifying any biases in the AI outputs?

“So we’ve got to make sure that we are thinking about diverse and representative datasets that enhance the AI’s ability: it’s got to be able to understand the different situations and outputs. So data quality is going up the agenda. We’ve got to get rid of the ‘garbage in, garbage out’ concept.”

Auditing the data in and out

Murillo Piazzi, digital consultant at BIM Academy, added: “This is artificial intelligence right? With real intelligence – humans – wouldn’t you check their work? What I see in the future is that we’ll let the AI do the menial and time-consuming tasks. Most people will be checking the AI’s output: we will all become, more or less, auditors and quality checkers of the work that the AI is doing for us.”

Andreas Galatoulas, data, analytics and AI director at Aecom, said: “You don’t want to be doing the same checks over and over again. You have to identify where the problem is, where your data is becoming corrupted [if there are issues with the output].”

He noted that two different algorithms with different biases generating different outputs from the same data is no different from two human professionals of different backgrounds having differing views about the same subject.

Finally, Xerini chief technology officer and co-founder Alex Luketa reinforced the importance of the human element: “It depends on what you’re asking the AI tool to do. There are some applications where it matters more than others. And if it’s important enough, you need a human to double-check things.”

Partners for the Digital Construction Summit 2023 are the Chartered Institute of Building, Cohesive, Elecosoft, Ideagen Mail Manager and NavVis. The Summit is organised by BIMplus and Construction Management.

Watch the webinar on demand.

Don’t miss out on BIM and digital construction news: sign up to receive the BIMplus newsletter.