Machine learning has come a long way, from simple pattern recognition to beating the top professionals in the strategy board game Go. Machine learning has since seen its applications expand to various other fields such as in personal assistants like Siri, Alexa, Cortana and Google Now; in diagnosis of deadly diseases; self driving cars; developing pharmaceutical drugs; and many more. The technology is becoming ubiquitous due to its vast scope for applications which are being discovered every day. An emerging application of this technology is in the invention of newer technologies. Apart from the ethical and moral issues surrounding this technology, it is important to address the issue of the protection of inventions derived through this technology.

First let's understand what machine learning is, with the example of Go. Go is a strategy board like chess, but is far more complex due to its larger board size and the greater number of alternatives for each move. Go is sometimes considered the hardest and the most complex strategy board game due to the same. Professional Go players in many East Asian countries typically practice for nearly 12 hours a day. Professionals today have reached this level of expertise with knowledge accrued over thousands of years and the countless amounts of deliberate practice. Machine learning is similar to such kind of practice. In machine learning, the machine is basically given a set of algorithms which helps it determine what is good and what is bad, i.e. winning and losing in the case of Go. The machine then plays and analyzes an enormous number of data/games while applying those parameters. Just like how humans learn to find out patterns from a given set of data, the machine too churns the data provided to it and eventually derives patterns and formulates strategies. While traditional programming deals with a 'set of instruction' applied to 'inputs' to produce a desired 'output', the computer is left to analyze the preexisting 'output' to devise 'a set of instructions' to apply to the desired 'input' on its own in machine learning.

Since machines are not susceptible to fatigue, unlike humans, and also due to the difference in speed of electric signals transmitted through metal conductors and neurons in the human body, the machine can make up for countless hours accumulated by humans in a matters of weeks. Given enough data and time, a computer can surpass humans in any field through machine learning. It is pertinent to note that through machine learning, a computer can only attain expertise in one single field. For example, AlphaGo, the computer which defeated the best active player in Go, can only play Go and is not capable of performing any other task. However, neural networks, which are multiple layers of machine learning algorithms, are capable of performing multiple tasks. The applicability of neural networks only expands the applicability of this technology to many other tasks and fields. Neural networks are still under development since the number layers, at least today, is restricted due to the lack of computer power. However, this has not obstructed its applications in other fields.

The application of our interest is the use of this technology 'for' inventions. Machine learning and neural networks have become popular among scientists to create or assist in their inventions. This technology has especially become popular in the fields of medicine and pharmacology where new variety of diagnosis, treatments and drugs are being devised by machines or with the help of such machines. The technology is also being used in other fields such physics and engineering to devise new experiments and inventions. With the advancements in neural networks, the scope for inventions will only expand.

But, can these computer or computer assisted inventions be protected?

Machine learning algorithms themselves are per se not patentable by virtue of being mathematical methods. This means that inventions which merely use the algorithm to come up with a model or a combination of values for certain variables for an invention cannot be patented. On the other hand, machine learning algorithms which have a technical character and practical application can be patented. The algorithms which are a part of the invention, which again is industrially applicable, can be and have been patented in the past. For example IBM patented a machine learning models for pharmaceutical drug discovery.

However, this leaves one issue unanswered. Can inventions created by machine learning models be protected by the current intellectual property laws? If yes, who would claim its ownership?

This issue is still up for debate, though the law is clearly leaning towards the protection of intellectual property created by a 'person' or 'human'. There have been proponents, like Professor Ryan Abbott from University of Surrey, who support the idea of granting protection to such machines. However it has received significant criticism by many professionals and scholars. At present, the law is unclear and vague as to this aspect.

Machine learning is still in its infancy and is in no positions to replace inventors (yet!?). Still, these issues cannot go unaddressed and need urgent attention as the technology is burgeoning and expanding to various fields. It will be telling whether answers to such questions will be provided by laws over the coming years.

The content of this article is intended to provide a general guide to the subject matter. Specialist advice should be sought about your specific circumstances.