Christie’s and Sotheby’s auction houses are accustomed to selling expensive and unique artwork.
In October, Sotheby’s sold a work by renowned street artist Banksy entitled Girl with Balloon for $US1.4 million ($A1.97 million), which the artist then proceeded to partially shred by remote device. The purchaser still proceeded with the sale and it is anticipated that the (partial) mechanical destruction of the work has substantially increased its value. It is thought to be the first time that an artwork has been intentionally altered (read: deliberately vandalised by machine intervention). Arguably that use of machine intervention has contributed substantially to the work.
Christie’s also created history in late October when it sold an artwork entitled Portrait of Edmond Belamy for $432,500 which was nearly 45 times its high estimate of $7,000 - $10,000.
The unusual aspect of the artwork is that Edmond Belamy is a fictitious person, with his likeness having been wholly computer generated and, specifically, created by an algorithm. The “artist’s signature” at the bottom right provides a clue as to the origin of the work. In cursive Gallic script, it displays an algorithm defined by algebraic formula.
The portrait is the first to be sold in that manner – having been created by an artificial intelligence (AI).
To achieve the result, the system was “fed” with a data set of 15,000 portraits painted between the 14th and 20th centuries. The system contained a “Generator”, which created the image based on the information contained in the data set, and a “Discriminator” which is designed to pick the difference between a human-made image and one created by the Generator. A successful new work is created when the Discriminator thinks that the new images are human generated (real-life) portraits, rather than being computer generated.
An earlier project had been undertaken by a group of researchers in the Netherlands in 2016, after analysing works by 17th century Dutch masters, to produce a work entitled The Next Rembrandt. Similar technology has also been used with software that can create musical works after listening to earlier recordings. Others have been used to write poems and even local news articles.
With the further development of AI, programs are being developed which allow the technology to learn from data input and evolve to make decisions that may be either directed or independent of human input. The technology is becoming far less reliant on input from a programmer.
One question which has been flagged by the development of the technology is to identify whether any intellectual property rights have been generated and, if so, who is entitled to them. The answer may well become less clear over time, as the distinction between human effort which contains a creative element and computer generated work which has ‘learned’ to incorporate its own inventive or artistic step becomes less distinct.
The starting point for many creators may not be too far removed from the questions which are currently asked at the start of new projects. Some thought may have to be given to whether there is a need to own the output/rights in the works from the outset. There may also be other usual considerations, such as identifying who owns (solely or jointly) the end result where there are a number of contributors to the work.
As with many new developments, in AI, the legislative protection has not yet advanced as quickly as the technology, which makes early and ongoing IP management particularly important.
To add to the complexity, although there are often similarities, different countries apply different rules across the globe in relation to the ownership of intellectual property rights. Most require human input (with an element of creativity) for a work to be capable of IP protection, although others have left the door ajar at this stage.
Examples of the forms of IP which may be relevant:
Anyone using AI tools should consider their freedom to operate to avoid encroaching on existing patents covering AI innovation. An assessment of earlier patents can help to understand the scope of third-party rights and mitigate risk.
Agreements should attempt to:
As with most IP, in the event that a work is copied/reproduced, there is a risk of infringement that may lead to an AI system acting (or even operating autonomously) in a manner that infringes third-party IP rights. If that occurs, then the owner, developer or operator may be responsible for the breach and liable for damages/loss.
Recent reports have indicated an overwhelming number of businesses (including professional services firms) intend to increase their investment in technology over the next five years. A significant proportion of that investment is said to be earmarked for AI development.
An IP strategy for AI systems will layer rights to protect different aspects of innovation and address issues relating to the use or development of innovation through data. Registration of rights should be considered and agreements on IP rights should be established to manage risk.
In the meantime, I’m off to find an easel and a shredder.
"The content of this publication is for reference purposes only. It is current at the date of publication. This content does not constitute legal advice and should not be relied upon as such. Legal advice about your specific circumstances should always be obtained before taking any action based on this publication."