1

Can Computer Systems Using AI Patent Their Own Inventions

Increasingly, companies are using AI to invent new methods and products. But can such inventions be patented given the requirement that each patent application must identify at least one “inventor” which is defined as an “individual,” not a machine? (In re Application of Application No. 16/524,350).

The USPTO rejected a recent attempt to name an AI system as the inventor in two patent applications, holding that an “inventor” is limited to a natural person. Other major patent offices have also found that an AI cannot be named as a patent “inventor,” including the European, Japanese, Canadian, and Australian patent offices.

These decisions leave open the question of what, if any, legal protections are available for inventions and other works created solely by AI systems.

If an invention is invented by an AI system, does the law allow a company to put the name of a random person as the inventor even if that person didn't do anything to contribute to the invention? Or is choosing a random person considered a crime? How do we then determine who is eligible to be chosen as the inventor?

2
  • 2
    Surely humans are involved in posing the problem and setting up the AI program to attempt a solution. I don't think a company would need to put down the name of a random person. Jul 24 at 0:12
  • 1
    Computer programs don’t have property rights, so arguing whether or not a computer program “invented” something that could be patented puts the cart before the horse.
    – ColleenV
    Jul 24 at 10:10
2

The company wouldn't have to put the name of a random person. "AI" is a tool created with the expertise of a data scientist to provide output to a specific class of problems or operate in a specific environment, and further human actions are often required to place the output in context. "AI" is just advanced machine learning methods that humans can use as a tool (with the humans building and using the AI as a tool being the real inventors).

More (hopefully helpful) factual background following: This is not to say that machine learning algs are not impressive, or that the future might not hold something we cannot imagine. But machine learning algorithms are not created and do not provide output in a vaccuum. To create a machine learning algorithm, a data scientists has to a) frame the question or goal, b) decide what sort of algorithm to use, c) gather machine-parsable data, d) "train" the model by feeding it data so the algorithm adjusts to match the real world. Then, someone generally has to interpret the results and apply them.

Consider, as a mock example, that I take have access to 1 thousand different shapes of the outsides of an airplane wing from my competitors. I represent these designs as 3d vectors, and then I put each one into a machine learning algorithm called a neural network. The neural network updates and adjusts to fit the patterns with every design I put in, as I have programmed it to do. Then, I use my algorithm to output a new suggestion for the design. I have to look at this suggestion and determine if its trash, or how to actually create and test it. This is a silly example, but in it I have invented the new plane wing, not the algorithm I used to help me.

The results of machine learning algs can be very good at specific tasks, like playing chess (like Google's AlphaGo), but they lack expertise in broader scope. Even though you couldn't beat AlphaGo in chess, AlphaGo could never accomplish the countless tasks you do every day or recognize as many patterns in the broader world as you. (And maybe never will.) Machine learning algorithms are distinct from some other forms of computational assistance because a) they use statistics instead of intuition, so an explanation of why an alg decides on something is not always meaningful to a human, b) they can be really, really good, and c) some models learn and adjust to the world in a way more similar to humans and animals than a more vanilla problem solving algorithm. There is a lot of interesting research in things like unsupervised learning (where algorithms find patterns without being told exactly what to look for) and reinforment learning. But it is still inappropriate to think of AI as something that could "invent" something. (And I know that you framed the question as AI "inventing" something just to match the language used in the quote; its just a misleading way of describing the process.)

If you're really invested in this question, maybe try watching some YouTube vids about machine learning techniques, if you have not previously done so.

5
  • "further human interpretation is required to understand the output and apply it to something useful": I am not sure this is true. If we take the example of the Ramanujan Machine, it can come up with a conjectures for the derivation of fundamental constants from just their known values. Such an algorithm could have have commercial uses and be patentable.
    – Rod
    Jul 24 at 12:03
  • I'm not familiar with this. But looking back I could also see using the phrase 'understand' in the bit you quoted may be a poor choice. Will edit my answer :)
    – Hattie35
    Jul 25 at 15:34
  • Intuition is the “wing” that is the result that you feed your subconscious with an objective, and it outputs a result. The problem is that most of you who question the viability of AI achieving appear to overlook the fact that since René Decartes we haven’t had one theory about how our decision making is able to overcome our quantum physical, molecular and biological (“Material”) constraints; how we can change the course of Material events which determine us to certain action; what power metaphysical power we possessed that is not subject to our basic understanding of the action-reaction[...]
    – kisspuska
    Jul 25 at 15:49
  • [...]principle; where is energy coming from that is not subject to the deterministic nature of Material events; how does that not violate the law of conservation of energy? We don’t have answers just a presumption that we act “freely” and we are bestowed with “freedom of will”. This presumption is supported by a single theory: We all exist within a metaphysical omnipotence where we are bestowed by the power to process information “freely” — as a result of omnipotence — and that om’p. delivers our will into this physical reality. Believing in, to any extent, to be able to overcome Material[...
    – kisspuska
    Jul 25 at 16:01
  • [...]constraints is the equivalent of an implied belief of omnipotence regardless of asserting atheism or agnosticism as our theoretical physics stand. The intuition you refer to there is the same output of statistics on a level of information processing that we are not conscious of probably so as not to be distracted from the high level decision making we experience to conduct. Therefore, the problem is, indeed, still stands: Where to draw the line between computers inventing and humans inventing. And there is an ongoing debate over this, and some very recent int’l treaties were signed, too.
    – kisspuska
    Jul 25 at 16:05

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.