With the advent of AI-based software applications, the question increasingly arises who owns texts, images, music or similar results created by AI. It is difficult to answer this question since it implicitly postulates what it expects to be answered, namely whether AI is capable of creating something. In other words, there is a petitio principi, as occasionally occurs in AI discussions[1].
Improperly phrased questions rarely lead to useful answers. Although even appropriately put questions do not per se guarantee suitable answers, they offer at least a fair chance that the search for an answer may lead to some knowledge gain, be it only the insight that an approach pursued turns out to be the wrong one.
In the present case, we are dealing with the image of a dog. The primary interest is not to know who owns this image, but rather who created it, because the answer to the latter question determines in turn the attribution of the image’s intellectual property rights.
To better understand how something can be created, it is helpful to look at some concrete examples:
Example 1
A painter fills three tin cans with oil paint and hangs each can one by one on the same string that he has attached to the ceiling. He places a large sheet of white paper on the floor below the suspended can of paint. He then drills a small hole in the bottom of the can and gives it an impulse by hand so that it performs more or less circular movements and leaves corresponding traces of colour on the paper. As soon as the can comes to a standstill, the painter repeats the same procedure with the other two colour cans. By doing so, a fascinating three-colour picture of overlapping circular figures appears on the paper.
Who created this picture? The colour in the tin cans? The gravitational force of nature? The painter by implementing a simple idea of how to create pictures? One will probably tend towards the last answer.
Example 2
Two days later, the painter is looking for new artistic inspiration. He lays again a large white sheet on the floor, fills three paint buckets with one colour each and fetches his goose Gerda from the garden. He then dips her feet into one of the paint buckets and lets her waddle around freely on the paper. Subsequently, he repeats this process with the other colour buckets. The result is a three-coloured jumble of goose footprints on the paper, from which the imaginative eye may recognise, or at least thinks seems to recognise artistic figures.
Who created this picture? The goose or the painter? One will probably assume the latter.
Example 3
The following day, the painter's daughter comes to visit her father's studio with her three-year-old son. Daughter and father have a lot to talk about and while both are engrossed in their conversation, the boy takes the opportunity to reach into his grandfather's paint pots and leaves colourful traces on a ready to use canvas with his bare hands.
Who created this picture? The boy himself? His careless mother? His father or his grandfather, the painter? One would probably assume that the child created the picture.
Example 4
The painter often takes his faithful dog Dada to the studio whom he has taught to grasp a paintbrush with his snout, to dip it into a bucket of paint and to move the brush back and forth on a canvas. Thus, upon the painter’s command, there results a multicoloured jumble of paint splashes, strokes and dots which, when viewed insistently, inspires the feeling of light-heartedness.
Who created this picture? The dog or the painter? At first blush, one may have some doubts.
Example 5
The painter can no longer get the question out of his head whether his trained dog Dada or he himself painted the picture. The longer he thinks about it, the deeper he falls into a creative crisis, because the mere thought that his own dog could be a painter fascinates him on the one hand but calls his artistic self-image into question on the other hand. He therefore searches for new sources of inspiration and finds them on the Internet, where he comes across an interesting AI application. Just for fun, he types in the words “dog paints a painting” and is somewhat surprised to see that after a short moment, the image of a painting dog (as shown above) appears on the screen.
Immediately, the thought flashed through his mind: Who created this picture? He himself? The AI application? The AI developers or the creators of the training data used? One may have serious doubts on how to answer...
The above examples illustrate how people may use different means to create something. In example 1, the natural forces of gravity and Coriolis are used to create consistent line patterns on paper with cans of paint. In example 2, the instinctive movements of an animal (lacking legal capacity) and in example 3 the play instinct of a toddler (having legal capacity but lacking capacity of judgement) leads to a random colour pattern. In example 4, upon command a trained animal generates coloured structures on canvas and in example 5, an AI application, trained with large amounts of data, generates a digital image in response to corresponding prompt.
In these five examples, who created a work protected by copyright? According to the “creator principle”, which prevails in Switzerland, copyright law protects natural persons who have generated an immaterial performance of individual character and thereby an intellectual work.
a) In example 1, this readily applies to the painter. Although he uses the gravitational and rotational force of the earth to create a coloured image on paper, this image results from a process that he has devised and implemented himself and from which emerges an individually characterized and identifiable performance.
b) The situation is similar in example 2: The goose Gerda is not a legal subject that can acquire authorship of a work. Rather, she serves for the painter as a tool to produce a colourful painting. Although the goose's waddling on the paper – in contrast to the earth's gravitational force – is of a random nature, the choice of the paints and the placement of the goose on the paper are creative activities by the painter, which are reflected in the colour pattern as an individual performance.
c) In example 3, it is the toddler who qualifies as the author of the colour patterns he created on the canvas with his bare hands. Every natural person having produced an intellectual work is considered to be the author of it, regardless of whether a person is capable of judgement and action. Indeed, creation is conceived as a purely factual act and not as a juridical act. This explains why the author does not need to have a conscious mind of creation but his will to undertake actions, from which the work emerges, is sufficient.
d) In example 4, the dog Dada differs insofar from the goose Gerda (example 2) as the dog was specifically trained by the painter to apply colour to the canvas with a brush. Here, the authorship belongs again to the painter, given that he employs the dog (who lacks legal capacity) as a tool for a creative act.
e) In example 5, the painter uses an AI application freely available on the Internet and was not created by him[2]. The above image of the painting dog came into being by entering the words “dog paints a painting”; the rest was done by the AI application trained with large amounts of data. Should nevertheless the painter be regarded as the author of the digital image? This view would be supported by the fact that without the selection of a specific AI application and without the specific instructions by the painter, the digital image would not have been created, or at least not in the form in which it actually exists now. In fact, the painter essentially used an AI application as a handy new tool instead of working with a brush and oil paint. A certain creative activity on the part of the painter, reflected in the individual character of the digital image, is still present. By comparison, this is in plain contrast to entering the multiplication “13 x 13 =” into a pocket calculator, in which case the result will always be “169“, regardless of the person acting and the calculator being used.
What would be the alternatives to the painter's authorship? To grant the developers of an AI application the authorship of all the results produced by the use of his or her application, would ultimately mean that, by analogy and consequentially, one would equally have to grant to the manufacturers of brushes and oil paint the authorship of all paintings made with these products. This is neither a realistic nor a sensible approach. It would be even less realistic to attribute authorship to the creators of the training data used for the AI application as these creators will hardly ever be individually identifiable and, what is more, would be even further removed from the actual creative process of the digital image of the painting dog than the AI developer himself.
In sum, from a sober point of view, the most plausible approach is to attribute – if at all – the authorship to the painter when he employs an AI application as a tool[3] for his artistic work. Although this diminishes the sometimes almost mystical aura that surrounds the term ‘AI’ as being something unprecedented, there is no harm done if concrete problems of modern society are dealt with without mysticism.
[1] See blog 2 on the automated cars problem.
[2] Classic software programs written in computer language are not actual works but are treated as such by the legislator. To what extent AI applications can or should be equated with computer programs has not yet been sufficiently clarified.
[3] The term «tool» implies that its user creates a work of an individual character, i.e. it must be “his” work and not merely “a” work. For example, if someone translates a text by using an AI application and then revises the translation, this person creates its own individual text and acquires authorship thereof. However, if the person leaves the text unchanged, in order to acquire authorship proof of other factors will be required that give the translation an individual character (e.g. design and structure of the text, selection or configuration of the application).
Art. 2 Definition of works
1 Works are literary and artistic intellectual creations with individual character, irrespective of their value or purpose.
2 They include, in particular:
a. literary, scientific and other linguistic works;
b. musical works and other acoustic works;
c. works of art, in particular paintings, sculptures and graphic works;
d. works with scientific or technical content such as drawings, plans, maps or three-dimensional representations;
e. works of architecture;
f. works of applied art;
g. photographic, cinematographic and other visual or audio-visual works;
h. choreographic works and works of mime.
3 Computer programs are also works.
3bis Photographic depictions and depictions of three-dimensional objects produced by a process similar to that of photography are considered works, even if they do not have individual character.
4 Drafts, titles and parts of works, insofar as they are intellectual creations with an individual character, are also protected.
Art. 6 Definition
The author is the natural person who has created the work.
The blog posts in this series offer an interdisciplinary view of current AI developments from a technical and humanities perspective. They are the result of a recurring exchange and collaboration with Thomas Probst, Emeritus Professor of Law and Technology (UNIFR), and SATW member Roger Abächerli, Lecturer in Medical Technology (HSLU). With these monthly contributions, we endeavour to provide a factually neutral analysis of the key issues that arise in connection with the use of AI systems in various application areas. Our aim is to explain individual aspects of the AI topic in an understandable and technically sound manner without going into too much technical detail.
