Shutterstock and AI: Is it a really bad deal?

Shutterstock logo
Shutterstock® logo

Several weeks ago, Shutterstock informed its contributors that it collaborates with several projects connected with artificial intelligence. Because AI is considered as a threat to many jobs, many photographers and illustrators consider this step as an another “exciting news” which is actually bad for us. The question is if it is really so bad. I will try to find an answer for this question.

First of all, we have to admit that we do not know how the whole concept will be working in the final versions, but Shutterstock has given us some hints.

First, there is Shutterstock Dataset where the agency is providing our content to companies and research organizations to train artificial intelligence algorithms. AI needs a lot of data for its training in which it is trying to find some patterns during the training phase. After that (as an already trained AI), it can be used for various purposes where you expect that it would give you reasonable outputs for all the possible inputs.

For example, in order to make a self-driving car, you would need to record thousands and thousands samples of a driving situation just to train the car for that one particular situation. For a complex system, one would need a huge amount of samples for a huge amount of different situations. In the training phase of many of the AI algorithms, AI checks if it computes a reasonable output for given conditions (if it behaves similarly in comparison with the examples from the training data set for the same/similar conditions, otherwise it changes its properties in order to lower the error). If it works well, there is a high chance that the AI would reasonably work even for situations which were not in the training data set.

So, Shutterstock may provide our images for any company or organization which wants to use labeled images (images with keywords) to train their AI for almost any purpose. Do you need to train a detector of various type of animals? Just use the database of animals from Shutterstock.

The second thing is AI-Generated Content. In this case, if there is not reasonable image for a buyer, (s)he can let the system DALL-E to generate it according to the description. Do you want an alien in front of the Eiffel Tower? No problem.

What about the contributor?

The question is how does it influence the contributor. Here, I have to say that I am a little bit more optimistic than the majority of contributors (and pesimistic in some regards as well). The fact is that in both Shutterstock cases above, the contributor will get some financial compensation for the use of his/her images. No one knows “how much” though. And I really do not think it will be 0.01 USD and such non-sense values which we may sometimes find on the internet. Probably, it will be much, much, much,… much less. Nevertheless, I do not think it will be so bad as it seems. Why?

Disclaimer: My numbers below are just created from my head, nothing specific. Their purpose is to “show the concept”, not the actual numbers.

Let’s say that you want to sell a photo of a cat. It is so oversaturated topic (3.7 million images of cats on Shutterstock), so it is highly probable that even with a great photo, the new image will never ever sell. But it can be different with AI. If there is a million of cat images used for training the AI, each of them is eligible for a commission. Of course, this commission is going to be almost zero. Very very close to it. But there is one detail which everybody forgets. This commission is going to be paid for each image of a cat which was generated through this AI (at the end of each half-year). And it gets interesting here.

A small kitten
I do not think this image of a cat has ever been sold. On the other hand, if it is going to be used for training any cat-related AI recognition system, or for generating any cat-containing image, there can be some commission. A very very very small one. But for each such an image!

Imagine a situation today. For example, let’s say that there is one million of sold images of a cat in some time frame for $0.5M for contributors . The lucky ones may sell some images hundred times and earn few hundred bucks for one of their image of a cat, some can have tens of sales, some only few of them and probably tens of cents only. The majority of people will never sell majority of their cat images and will receive… zero.

Now, let’s go to the possible future few years ahead and let’s sell 1 milion of images of a cat which were created using AI. The buyers will probably pay the same, the contributors will probably get less (something has to be paid to OpenAI). Let’s say that there will be $0.4M for contributors for this one milion of sold images of cats. The contributors have provided 1 million of images for training, so each of these images would be worth 0.0000004 USD only. That sounds horrible. But something is missing. A very important detail. Really, a very important one.

The contributor is “selling” his image in each(!) of this 1 miliion of sales. Every time when there is an AI image of a cat sold, (s)he will be eligible for a commision. In other words, if the image was used for training the AI, one has achieved 1M of these mini… micro… nano-sales × 0.0000004 USD per image for training = 0.4 USD per each image of a cat which was used for training. Just to remind, majority of cat images are never sold and get zero. In our old non-AI system, we have to be lucky to be in the first pages of the search results to get some sale. In the new AI system, we are going to sell an image every time when there is a sale of this topic and one of our images was used for training.

Great? Yes and no. If you have a great image of a cat which has earned you a thousand of dollars, AI will probably hurt your earnings. On the other hand, if your image of a cat is not a great one or you have been just unlucky and your image in not somewhere on the first page of a search result (which is basically everybody in this crazy set of millions cat images), you may probably benefit from this.

Bad thing? It is going to be even more cruel number game again. Quality is not the thing that will automatically bring you more sales and more money. The important thing is to have more images which are going to be used for training. So, probably more image spam with the same object taken from tens of different angles to increase the amount of images used for the training. Do you have one image of a cat which was used for training? Maybe 0.4 USD for that for one million of sales covering this topic. Do you have 20 of them? 8 USD. (Once again, the numbers are just created in my head, not actual values).