Innovation-July-August-2023

F

E A T U R E

I llustration : AI creation

"What are we trading off here?" he adds. "Are we trading off an hour of somebody’s time to do the calculation manually or in a spreadsheet versus one minute. To me that trade-off is not appropriate if the AI could get it wrong. But if it’s going to take 10 million years of person-time and AI chugs away and spits out an answer in five minutes, well maybe that makes more sense.” He stresses it’s essential to understand the type of AI, as well as what it’s doing and what task it’s trained to do. “Sometimes that piece is actually tricky to tease out because the AI might be part of a larger tool,” he says. “Recognize that AI can be a lot of things. It’s not necessarily obvious as a user where the results are coming from. I think engineers know their tools quite well, but this is another case where they might have to stop and think, what is AI or machine learning really doing for me here?” Machine learning can be sensitive to small changes practically imperceptible to humans; meaning that, if the input changes a tiny amount, it could alter the output completely. That creates a need to thoroughly explore how sensitive the tools are to small changes in input. “I don’t think we should be relying exclusively, or at all, on AI or machine learning unless the tool’s outputs have been independently verified by a professional,” he says.

“When people are using an AI tool, sometimes there’s a tendency to jump into it because it’s flashy and exciting. It’s the next big thing. It’s usually good just to pause and think, ‘OK, do we really need this or are we just adding more uncertainty to our life?’ If you could do the same thing with a manual calculation, do you really need AI to do it for you, especially if the consequence of a failure is quite high?

Our IP lawyers are here to help you build the future.

Oyen Wiggs Green & Mutala LLP patentable.com

PROTECTING INNOVATION

1 6

J U L Y / A U G U S T 2 0 2 3

I N N O V A T I O N