A Deal with the Devil? Insights on Google’s AI Patents [1]

The fast-paced development of AI and deep learning technologies have been overwhelming in the last few years, and open source softwares are beyond doubt one of the main reasons for such dramatic progress. Open source is believed to be the rocket fuel for innovation in the entire software industry, and companies such as Apple, Samsung, and Qualcomm, who traditionally have relied on the existing patent paradigm to protect their technology, are also actively participating to create and foster open source communities.

 

In the field of deep learning, it is hard to disregard the huge impact of Google's open source software “TensorFlow”. TensorFlow is a framework with codes for many deep learning models, so machine learning (ML) developers and engineers can easily implement these building blocks on their own deep learning models and as a basis for more advanced research.

 

Because open source softwares have acquired a central role in any software strategy, now companies contribute to the development of open source software and also benefit from the advantages that come with it, such as the opportunities for collaborative development. Hence, to a certain degree, patents seem to have become a relic of the past in the field of software.

 

Like with TensorFlow, Google open-sourced software developed by their AI researchers, and so did Facebook, Amazon, Microsoft, and even Apple, who have been encouraging openness as well. But at the same time, Google has been registering key patents to claim ownership over AI techniques, in particular related to deep learning. The fact that Google is patenting key machine learning techniques has become a concern for many in the open source community, as can be seen in various discussions on Reddit and Facebook.

 

Google has been steadily building an extensive portfolio of patents related to deep learning, including patents on essential technologies such as DQN, dropout, and batch normalization, which are commonly used by deep learning researchers and engineers. The range of Google's deep learning patent portfolio is diverse and covers fundamental algorithms in deep learning as well as applications employing deep learning.

Below some of the most significant ML-related patents owned by Google:

 

- Methods and apparatus for reinforcement learning (DQN)

 

 

- System and method for addressing overfitting in a neural network (dropout)

 

 

- Batch normalization layers (Batch normalization)

 

 

---

Open source licenses are licenses that comply with the Open Source Definition — in brief, they allow software to be freely used, modified, and shared.
https://opensource.org/licenses/

---

 

        Open source licenses, which are founded on the principles of sharing and freedom for the use and modification of technology, and patents, which grant a temporary monopoly over a certain type of technology, are concepts that have polarizing ideologies. What motivates Google to pursue two contradicting strategies at the same time? Will TensorFlow users be safe from future patent lawsuits filed by Google? Do open source licenses provide users a free pass to ignore patent infringement issues?

 

In the upcoming articles, we will attempt to answer these questions. In the next column, we will examine open source licenses and their relation to patents. We then consider a few hypothetical scenarios and try to identify Google’s real intentions. Finally, we will have a look at the impact and possible repercussions on any patents owned by users of TensorFlow. 

 

To read the next article, please go to this link