In Tokyo I had a great chat with a very smart man.
We had two pictures of apples and he said, “why do people think they look alike?”.
Immediately I thought it was a riddle, and the sweat started to bead down my neck, but he was being serious.
We’ve always been computer vision experts, and years before it was sexy we adopted AI techniques to take the visually similar to another level to include “semantic meaning”, that is, what objects are in the image. Is it a car, a circle, a star or in our case above, an apple. The aim was to get inside the user’s head and return similar results for brand protection by extracting features of images a brain would find similar.
We’ve had a lot of success, but as the technology has progressed over the years, so have the cowboy copycats. Those people that ride in, connect an open source model to a set of data, and think they are the fastest gun in the west. They shoot, but they only need to miss once to be dead.
There are a lot of open source deep learning models you can use. And yes, if you upload an apple image, it is probably going to be smart enough to go, “apple” like my kids could do before they were 1 (well, definitely one of my kids, the other maybe around 1 and a half, but hey, I’d never compare).
You then just show apples and the user goes, “wow”. But that’s where the problems start. It’s not just the apple that’s important, it’s the shape of the apple, the fine details of the apple that might actually look more similar to a funnily drawn macadamia. That macadamia needs to be found, so it’s the 1% that are now the real problem cases.
AI models have drastically improved similar searching, but they create problems when used poorly and it’s what those cowboys are missing that will cause grief.
My answer to my Japanese friend was simple, “they aren’t alike”. They were both apples, but in terms of brand they were completely different. There was no infringement. We need to identify that semantic meaning, but we need to combine it with a million other things, and that accuracy is what we strive for every day.
We have teams across the US and Aus now. We’re growing fast and we’re working with several government bodies and huge commercial partners. It’s awesome though when a potential client gets it straight away and appreciates the real power of AI is in conjunction with impeccable data and their/our knowledge to make amazing decisions. We use AI (our machine learning experts experiment non-stop), but we use it by combining it with years of experience and testing to improve a specific problem facing the world’s brands.