My thoughts on suggested swaps – Gurushots

Tom Bourdon 15 Comments ,

So, over the last few weeks or so I’ve read several posts on Faceboook from people who are upset/disappointed/annoyed/frustratedโ€ฆ about the suggestions, that Gurushots make regarding swaps. And at first light, I can kind of see where this is coming from. But, if we look at it a little closer I think we can understand things a little better.

The way the system works

The system uses Artificial Intelligence (or AI for short) to label each and every photo uploaded (I’m guessing there are multiple billions by now). Then to make a suggestion it searches those labels to see if any of them match the challenges that are currently running. If it finds a match, it then does some calculations on the votes those photos have achieved in previous challenges and then finally makes, what it believes, is a good suggestion that will give you more votes.

Now, it’s the AI that is causing the issue here. When I say AI I am, of course, referring to the code behind the Gurushots image recognition system. The code as in the ones and zeros that can ‘look’ at a photo and distinguish, with a high degree of accuracy, between thousands of different objects. For me that alone is mind-boggling. And as far as I’m aware this technology isn’t being used on this scale by anyone else. *I’m fearing the comments to say I’m wrong here.

Occasionally the system does find something in a photo that is obviously (to the human eye) not in the photo. It might recognise a small log as a bird for example. This can sometime produce amusing results, There was one example in the recent ‘Animals with horns’ challenge where the suggestion was a man playing a trumpet. Technically a man is an animal, and a trumpet… well it could be considered by many as a horn instrument (although, according to Google it isn’t). But surely this wasn’t what the challenge was about. And the photo would have no doubt been reported as being off-topic.

So, what’s the point?

Simple, to give you more options and no extra cost (in terms of money or time). This image recognition also allows you to search your photos too using these labels. I can’t count the number of mind-numbing hours I’ve spend keywording/labeling photos for stock websites. I guess it will only be a matter of time before the stock websites catch up and offer a similar service. Knowing some of them I guess there may even be a fee for using this service.

ones and zeros that can ‘look’ at a photo and distinguish, with a high degree of accuracy, between thousands of different objects… mind-boggling

I would also love, love, love to have keyword/label suggestions in Lightroom. I have hundreds of thousands of photos in my catalogs and manually labeling/keywording them is a pain in the butt, to say the least. No doubt I’m missing many many useful keywords. The addition of which would bring to light photos that I may have forgotten about or not considered for a particular project/client.

Don’t forget…

There are 2 things that we should also remember. Firstly you can edit these labels as much as you want. So, if you see the AI has mistaken your photo of a turtle for a rifleย (I’ve never seen a case this extreme on Gurushots) then you can simply remove the tag ‘rifle’ and add the tag ‘turtle’. Simple.

Secondly, these are only computer generated suggestions and they are very simple to ignore. Even humans make bad suggestions sometimes, and what do we do when that happens? We simply ignore it and move on.

When I studied software engineering at university AI was a thing of sci-fi films, something that seemed like an impossible dream. But here we are in 2019 using it to play a gameโ€ฆ Personally, I still find the whole thing amazing.

Is the Gurushots image recognition system perfect? No way! Will it ever be? Well, I have my doubts that technology will ever be able to recognize every object 100% of the time. But then 20 years ago I couldn’t imagine AI doing what it does now – so who knows.ย It’s still early days for this type of cutting edge technology.