While it is arguable that true AI is not currently in existence, there are many products and services that employ machine learning algorithms. These algorithms do not require precise programming for every eventuality but instead are given datasets to learn from. If the machine learning algorithm incorrectly identifies some data it is told so and applies weight so that it can identify that data. If large amounts of data are provided, then the machine learning algorithm becomes better at recognizing what data it is looking at and the result is impressively accurate facial recognition, speech recognition, and data prediction (such as autocomplete).
But AI can be used in many other ways and one area that scientists are looking into is creativity and design.
While not designed by AI, this antenna was designed using genetic algorithms. Image courtesy of NASA
One key aspect of intelligence in humans is the ability to not only recognize and learn from our environment but to also solve problems in many creative ways. It can be argued that almost all human invention is a result of our creativity.
But what if AI could be creative and solve problems? This brings up a world of possibilities, but also a Pandora's box of legal questions. One such question hitting the public sphere now is:
If an AI designs something new that can be patented, who owns that patent?
Meet DABUS, the Creative AI
Whether an AI can own its work is a complex issue, legally speaking.
This is a problem that a team of researchers has come across regarding the patent system. The research team, hailing from Imagination Engines and led by computer scientist Dr. Stephen Thaler, has created an AI known as DABUS ("device for the autonomous bootstrapping of unified Sentience").
In 2017, DABUS became somewhat famous for creating surreal art. Thaler and company accomplished this by injecting noise into DABUS's neural network to generate “thoughts”—but the key feature of DABUS is that it is not designed to solve any specific problem.
Unlike other AIs, DABUS can be used to develop new ideas.
The researchers have utilized this capability by having the AI develop a solution for food containers. DABUS's idea is arguably brilliant: a fractal-based design.
DABUS's fractal design. Image used courtesy of The Artificial Inventor Project
The idea is that the food containers use a fractal perimeter with parts that point outwards and identical patterned parts that point inward. This allows the containers to mesh together to improve stability of the containers but it also makes it easier for robotic systems to pick them up and move them around.
But the AI has also designed a beacon system that is significantly harder to ignore by pulsating a light based on fractal patterns to indicate neural activity.
The researchers took these ideas to be patented but, according to US patent law, only humans can be the owner of a patent. This presented a strange scenario in which researchers who had virtually no direct input into a design were able to take ownership of it.
What Does Inventorship Entail?
The first, and most problematic, issue is the legality of ownership and inventorship.
According to the US Patent and Trademark Office, the consolidated patent laws use the terms "inventor" and "person" but do not specify what those terms entail. That is, US patent law does not actively indicate that an "inventor" must be human. All the same, Thaler's team has been denied the ability to list DABUS as an inventor on patents.
Perhaps part of the puzzlement comes from the fact that Thaler has actually patented "Device and method for the autonomous bootstrapping of unified sentience", himself.
In any case, an AI presently does not possess rights under US law because it is not a living being. This is also a concept that patent-owners will be familiar with because a company or individual alike can own a patent—but when a patent is filed, the individual(s) who actually created the invention must be listed. This is why most patents filed by companies list the employees involved in the invention as they are the people who thought of and created the design.
In the case of DABUS, the AI cannot be listed as an inventor, nor can the company that created the AI. This means that the individual designers of DABUS would need to be listed, which is misleading as they had no direct input into DABUS's designs.
An additional quandary is that AI creations could increase GDP by as much as 10% but these creations would not be owned by those systems. Instead, owners of powerful computers could run AIs 24/7 and therefore be the owners of countless filed patents. This may make it harder for individuals to compete and keep up with large tech companies.
Thales, for his part, has recruited an international team of patent lawyers to apply for patents on behalf of DABUS which they call The Artificial Inventor Project. They have put together a legal case for recognizing the creativity of AI and its legal right to be acknowledged as an inventor in patents.
According to them, "No United States Law explicitly prohibits protection for autonomous machine inventions."
Now we'll see if their arguments in favor of allowing AI inventorship will be accepted.
It is clear that, as AI systems currently stand, they are nothing more than weighted functions that can alter themselves to better match and recognize data. However, some technologies such as spiking neural networks (SNNs) are starting to operate differently compared to traditional neural networks. Instead of weighted functions, they mimic how biological brains function with synapses, cells, and receptors that have activation levels.
While current systems are simple in design, what would happen if a system was given as many connections and neurons as the human brain? At this point, what would the AI be capable of?
To end this piece, here's a question I came across that made me sit down and think:
At what point will it be immoral to unplug an AI?