Cyber law or internet law concept with 3D rendering of AI robot and legal scales and judge with gavel
Getty
The Wild West of Intellectual Property in AI
In the past, intellectual property issues in AI were generally overlooked. Technology developed very rapidly, most systems published in the academic literature rarely progressed beyond proof of concept, and industrially powerful platforms led to paradigm shifts only in niche markets and rarely warranted the application of patents.
As someone who founded an AI company in 2014 at the height of the deep learning revolution, I still remember how it started. In 2014 and 2015, deep neural networks began to outperform humans in many tasks, including image recognition and basic computer games. And while much of this research has been publicly released, some of the companies have begun to build significant intellectual property portfolios. I remember after Google’s acquisition of DeepMind we were analyzing the IP landscape, many of the published concepts had patents pending. We decided to follow their path. At that time, in order for an AI company to justify valuation, venture capitalists recommended that we patent some of the underlying technologies, and we received numerous issued patents for transcriptomic, proteomic, and microbiomic predictors of biological age, many generative concepts in chemistry, biology, and even methods connecting the two fields. Today, even a high school student can replicate this work, but a decade ago, when deep learning scientists were very scarce and expensive, engaging in this type of research was very risky. However, when it came to enforcing these patents, it became clear that this was not common practice. Plus, violations are very difficult to prove in server industry systems. However, we continued to patent, even to show some of the partners that the software they were licensing was based on the original work. And so far we have not pursued the parties that are potential infringers. The market for AI in drug discovery is relatively small, and most of the value is not in the IP around AI, but in patents around specific therapeutic programs.
It’s likely that Google has also followed a similar train of thought when it comes to its many AI startups. Pursuing their AI patents just didn’t make much business sense, and building the AI ecosystem was a community effort. However, as ChatGPT, DALL-E, and other paradigm-shifting tools take the market by storm, challenging Google as the dominant player in AI and undermining its core business of search, Google’s attitude toward IP protection may change.
Google holds critical IP in systems using self-awareness
And when it comes to IP protection in AI, I don’t know of any company that is stronger than Google. I remember how our teams spent weeks making sure the architectures were unique enough to avoid potentially infringing on Google’s IP. Even DeepMind, a wholly owned subsidiary of Alphabet, has a very strong IP position in the space. A simple WIPO search returns over 800 results.
Google is also the original inventor of the basic self-awareness methods used in transformer architectures. It was the Google scientists who made the fundamental breakthroughs in transformer neural networks that paved the way for GPT-3. In 2017, at the Neural Information Processing System (NIPS, later renamed NeurIPS) Conference, Google scientists presented a seminal paper titled “Attention is all you need.” As of January 2023, this article has been cited over 62,000 times, making it one of the most cited articles in AI.
A screenshot of a seminal article on transformer neural networks
Dr. Alex Zhavovonkov
And a simple WIPO and Google patent search produces a very comprehensive and broad patent covering this powerful methodology.
A screenshot of a Google Patents search result for one of the major patents covering self-awareness… [+] based architectures
Dr. Alex Zhavoronkov
It is clear that the patent application was filed on June 28, 2018, approximately one year after the main article appeared on the preprint server on June 12, 2017, and the current status is that the patent has been granted.
OpenAI is known to avoid patenting
OpenAI, a well-known AI powerhouse since its inception, was originally established as a non-profit organization. I tried to find the patents filed under OpenAI and found it a challenging task. A WIPO search returned only a few results for OpenAI srl, which appears to be an unrelated company. A simple Google search didn’t turn up any significant results, so I asked ChatGPT itself.
Screenshot of ChatGPT results for OpenAI and DeepMind patent positions
Dr. Alex Zhavoronkov
It also correctly describes the patent situation in deep learning. Indeed, IBM, Google, Microsoft, Samsung and Baidu have significant IP portfolios.
ChatGPT on which company holds the most patents in deep learning
Dr. Alex Zhavoronkov
And according to ChatGPT, while GPT uses self-attention, it’s unclear whether Google’s patent will cover the use of self-attention in the GPT architecture.
Screenshot of ChatGPT output in response to a question about whether GPT is covered by a Google patent
Dr. Alex Zhavoronkov
Will Google try to test or enforce its patents?
Although Google has a very strong IP portfolio spanning many areas of AI and has a patent issued for its use of attention-based sequential transduction in neural networks, it is not known for flexing its IP muscles in the field of artificial intelligence. However, as generative AI continues to take over the market and Microsoft integrates OpenAI tools into its extensive product ecosystem, this position could quickly change. And ChatGPT has a perfect answer for it.
ChatGPT on whether Google will ever try to use its patents to challenge OpenAI
Dr. Alex Zhavoronkov
Don’t be mean
In 2014, Elon Musk, one of the co-founders of OpenAI, famously made all Tesla patents publicly available to prevent unfair competitive practices in the EV industry. And OpenAI’s decision not to waste time patenting core AI technologies likely stems from this philosophy.
Although Google pioneered key theoretical aspects of generative AI and certainly has priority in core neural network technologies using self-attention, it is unlikely to take legal action against OpenAI. Patents are a great way to demonstrate priority and protect against patent trolling, but suing other scientific groups in AI is not common practice. Especially considering the fact that OpenAI emerged from a non-profit organization run primarily by people who are dedicated to making this world a better place.
An image generated by Midjourney’s AI using the prompt “illustrates a patent war between Google represented by … [+] Godzilla and OpenAI presented by King Kong”
Alex Zhavoronkov, PhD and Midjourney
Add Comment