The Perspective of an AI Specialist Patent Attorney on AI Patents in the 'Scale is All You Need' Era
파인특허
July 26, 2024
.png)
The rapid advancements in generative AI technology have popularized the phrase 'Scale is All You Need' in the industry. This notion implies that with vast amounts of data and computing power, AI model performance can be significantly improved, suggesting that computational power is the key driver behind AI success. Consequently, companies worldwide are keenly focused on securing AI-related patents.
South Korea ranks third globally in AI patent applications over the past decade (2013-2023). However, a closer examination reveals some gaps. Korean companies primarily focus on applications using pre-trained foundation models, with fewer patents filed for fundamental AI architecture.
Conversations with AI researchers shed light on this trend. Since the emergence of models like ChatGPT, research has shown more tangible results through scale competition based on data and hardware rather than through architectural improvements. This area is also driven by economic incentives.
From last year to the present, the AI race has primarily been about expanding context windows, increasing parameters, and utilizing more training data, mainly among major tech companies in the U.S. and China. Academic research has received less attention, with the spotlight mostly on efforts to scale more efficiently, such as through selective attention mechanisms.
This raises questions about the effectiveness of patents on technologies using foundation models. The patent system aims to promote industrial development by granting exclusive rights in exchange for technology development and disclosure.
When comparing 'foundation models' with 'technologies using foundation models,' significant innovation and investment are focused on the foundation models themselves. Yet, large tech companies are reluctant to disclose these core model technologies, resulting in limited patent filings for the models themselves.
Companies owning foundation models hope their models will be widely used across various fields. Therefore, granting exclusive rights for 'technologies using foundation models' might not align with the patent system's purpose or effectively promote technological innovation.
Looking at OpenAI's patent filings provides some insight. OpenAI has focused on patents related to prompts, such as determining parameters through prompt input (US 11886826 B1) and deciding the tone, structure, and format of text based on user instructions (US 11983488 B1), rather than patenting the foundation models themselves.
Moreover, a recently granted patent (US 12039431 B1, dated 2024.07.16) describes an interface allowing users to input context prompts (e.g., clicks, annotations) to emphasize specific parts of an image. This approach seems aimed at protecting their service rather than the foundation model.
The patent system has evolved with advancements in technology. Early on, protection focused on basic technologies, but over time, it extended to various application technologies. This trend is evident in the histories of computer program inventions and smartphone applications.
Initially, software itself wasn't patentable, but later, various application technologies using software were protected. Similarly, complex user interfaces and diverse functions in smartphone applications now receive patent protection. This evolution shows the patent system's flexibility in responding to technological progress.
In conclusion, patents are crucial for fostering technological innovation. However, whether the current patent system effectively supports technological advancement in the 'Scale is All You Need' era requires ongoing review and discussion. Korean companies can maintain competitiveness in the global AI race by strategically filing and protecting both foundational and application technologies.
Pine Patent Law Firm prides itself on its expertise in AI patent applications. With dedicated patent attorneys from the AI Convergence Business Group, including specialists from Samsung Electronics AI, Hanwha Vision, Naver, and the Defense Science Research Institute, our firm leverages deep understanding and practical experience in AI technology to protect innovations and assist in successful patent filings. We welcome inquiries about AI patent applications at any time.
