AI가 변리사를 대체할 수 있을까? Can AI Replace Patent Attorneys?

 AI가 변리사를 대체할 수 있을까? Can AI Replace Patent Attorneys?

오늘 "튜링 테스트와 특허 세계: AI 생성 특허 출원의 평가"라는 흥미로운 글을 읽고 이를 공유하고자 글을 작성합니다.

Today, I came across an intriguing article titled "Turing’s Test in the Patent World: Evaluating AI-Generated Applications" and wanted to share my thoughts on it.

---

튜링 테스트는 컴퓨터 과학자 앨런 튜링(Alan Turing)이 1950년에 제안한 개념으로, 인공지능(AI)이 인간처럼 사고하고 대화할 수 있는지를 판별하기 위한 테스트입니다.

튜링은 기계가 인간과 구별되지 않을 정도로 지능적인 대화를 나눌 수 있다면, 그 기계는 "지능적"이라고 간주할 수 있다고 주장했습니다.

이 개념은 철학적인 질문인 "기계가 생각할 수 있는가?"를 실용적이고 검증 가능한 질문인 "기계가 인간처럼 행동할 수 있는가?"로 전환한 중요한 기여로 평가받고 있습니다.

The Turing Test, proposed in 1950 by computer scientist Alan Turing, is a concept designed to assess whether artificial intelligence (AI) can think and communicate like a human.

Turing argued that if a machine could engage in a conversation indistinguishable from that of a human, it could be considered "intelligent."

This notion shifted the philosophical question of “Can machines think?” to the more practical and testable query, “Can machines behave like humans?”

---

Ian Schick 박사는 이 튜링 테스트를 AI 생성 특허 출원과 인간 변리사가 작성한 출원을 비교하는 데 적용했습니다.

그는 특허 심사관이나 변리사가 AI가 작성한 출원과 인간이 작성한 출원을 구별하지 못한다면, 이는 AI가 고품질의 법률 문서를 작성할 수 있는 수준에 도달했음을 의미한다고 보았습니다.

Dr. Ian Schick applied this concept to the world of patent drafting, using the Turing Test framework to compare AI-generated patent applications with those drafted by human practitioners.

He posited that if patent examiners or attorneys could not distinguish between applications written by AI and those drafted by humans, it would signify that AI has reached a level capable of producing high-quality legal documents.

---

테스트 결과는 매우 흥미로웠습니다.

AI가 작성한 출원은 구조적으로 우수했고 오류가 없었지만, 인간의 글에서 발견되는 독특한 흔적을 찾을 수 없어서 인간이 작성한 출원 명세서와 구별할 수 있었다는 것입니다.

The findings were fascinating.

While AI-drafted applications were structurally sound and error-free, they could still be distinguished from human-drafted applications due to the absence of certain human characteristics.

좀 더 상세한 이유는 다음과 같습니다:

The reasons for this distinction are as follows:

1. 표현 스타일의 단조로움: AI는 동일한 기술을 설명할 때도 문구, 문장 구조, 강조점의 변화가 적어 개성이 부족했습니다.

   Monotony in Expression Style: AI tends to rely on uniform wording, sentence structures, and emphasis, resulting in a lack of individuality when describing the same technology.

2. 인간 특유의 변동성 부족: 인간 작성물은 경험과 직관에서 비롯된 사소한 실수나 비일관성을 포함하지만, AI는 이를 완벽히 제거했습니다. 이 완벽성은 오히려 "비자연적인 특징"으로 작용했습니다.

   Absence of Human Variability: Human drafts often include minor inconsistencies or intuitive nuances stemming from personal experience, whereas AI eliminates such variability entirely. This “perfection” can ironically appear unnatural.

3. 창의적 표현의 부재: AI는 기존 데이터를 기반으로 작동하므로 완전히 새로운 비유나 사례를 제시하는 데 한계가 있었습니다. 반면, 인간 작성물은 개인적 스타일과 독창성이 반영되어 있었습니다.

   Lack of Creative Expression: AI operates on existing data patterns, which limits its ability to introduce novel analogies or examples. In contrast, human drafts frequently reflect unique stylistic choices and creative approaches.

AI는 통계적으로 가장 적합하거나 일반적인 표현을 선호하기 때문에 모든 작성물이 구조적으로 비슷하고 단조로워 보일 가능성이 큽니다.

이는 AI 작성물이 인간 작성물과 구별될 수 있는 주요 이유로 작용했습니다.

AI inherently favors statistically optimal or conventional expressions, which can make its output appear overly uniform and predictable.

This uniformity serves as a key factor in distinguishing AI-drafted applications from those authored by humans.

---

Ian Schick 박사는 AI가 계속 발전하면서 AI 생성 특허 출원이 점차 보편화될 것으로 예상하고 있습니다.

Dr. Schick predicts that as AI technology continues to advance, AI-generated patent applications will become increasingly prevalent.

그는 특히 다음과 같은 전망을 제시했습니다:

He offers the following insights into the future of patent drafting:

1. 일상적인 특허 작성: 표준 기술이나 프로세스를 다루는 특허에서는 AI가 작성 과정을 간소화해 시간과 비용을 절감할 수 있을 것입니다.

   Routine Patent Drafting: For patents involving standard technologies or processes, AI can streamline drafting, reducing time and cost.

2. 복잡한 특허에 대한 인간의 역할 강화: 창의적인 법적 전략과 깊은 기술적 통찰이 필요한 복잡한 발명에 있어서는 여전히 인간 변리사가 필수적일 것입니다.

   Enhanced Role for Human Attorneys: For complex inventions requiring creative legal strategies and in-depth technical expertise, human practitioners will remain indispensable.

3. AI와 인간의 협력: 궁극적으로 AI는 인간 변리사를 대체하기보다는 그들의 능력을 보완하는 도구로 자리 잡을 것입니다.

   AI-Human Collaboration: Ultimately, AI is expected to complement, rather than replace, human patent attorneys by serving as a powerful tool that enhances their capabilities.

더 자세한 내용은 원문에서 확인할 수 있습니다:
For more details, you can refer to the original article:

Turing’s Test in the Patent World: Evaluating AI-Generated Applications

How AI and Strategic Drafting Can Reshape the Patent Landscape

How AI and Strategic Drafting Can Reshape the Patent Landscape

December 2, 2024 — In reviewing an insightful article by Dr. Ian Schick, I encountered a compelling argument about the transformative potential of artificial intelligence in the patent drafting process. Dr. Schick suggests that AI can streamline labor-intensive tasks—like preparing detailed descriptions and refining drawings—freeing patent professionals to devote their expertise to high-value matters such as claim drafting and inventor interviews. He anticipates that this shift could reduce the total drafting time per application from approximately 20 hours to around 6.7 hours, thereby lowering service fees to one-third of their current level.

While I hesitate to label detailed descriptions and drawings as “low-value,” there is no denying they consume a significant share of the preparation effort. Using AI to automate portions of these steps appears both logical and efficient, ultimately enhancing overall patent quality. Notably, Dr. Schick’s illustrative figures and tables break down the drafting process in a manner that highlights each task’s value—though I would have preferred to see prior art investigation more prominently included, as it is integral to most filings.

---

Why Detailed Descriptions and Drawings Are Crucial

In practice, the initial specification’s thoroughness often outweighs the importance of the original claim set. Rarely do claims issue without amendment; examiners commonly identify grounds for rejection that lead to strategic adjustments. When amending claims, an applicant must stay within what is explicitly or implicitly disclosed in the filed specification. Therefore, the application’s initial level of detail profoundly influences the ultimate patent scope.

Moreover, companies increasingly recognize the importance of accounting for European patent standards at the outset. Filings tailored to U.S. or Korean rules can falter under Europe’s more exacting scrutiny, where Article 84 objections or issues with non-allowable intermediate generalizations frequently arise. A robust drafting strategy should anticipate these hurdles by integrating European requirements from the beginning.

---

Drafting Strategies and the Role of AI

From a drafting standpoint, an effective approach is to conceptualize the overall inventive idea through schematic diagrams or “conceptual figures,” and then systematically introduce variations as they become clear in later product-development stages. Highlighting which features are essential, optional, or purely enhancements clarifies the invention’s scope. At the same time, such categorization—if done too rigidly—might inadvertently bolster an examiner’s view that a claimed invention is an obvious combination of known elements. Caution and balanced drafting are therefore prudent.

For patents serving as strategic “offensive weapons,” it is well worth investing time and resources in a detailed specification, high-quality figures, and a thorough review of relevant prior art. Here, AI offers considerable advantages: in my own practice, I have used ChatGPT-4 to review and refine sections of a recently disclosed specification on a paragraph-by-paragraph basis, finding it especially helpful in reducing preparation time without sacrificing quality.

---

The Cost Equation

It is instructive to consider cost benchmarks. Mid-level associates at major U.S. law firms often command hourly billing rates near $750. Based on international data comparing the United States and Korea, the U.S. purchasing power parity (PPP) is about 1.4 times higher, with average wages (in PPP terms) around 1.75 times higher. By that logic, a mid-level associate in Korea might have a reasonably set billing rate of roughly $429 (KRW 600,000) per hour. At 20 hours of drafting, an application would therefore cost about $8,579 (KRW 12 million). Yet, outside highly specialized areas such as pharmaceuticals, achieving such fee levels in Korea can be challenging.

---

The Strategic Value of In-House Patent Teams

Ultimately, these considerations highlight the pivotal role of corporate patent counsel. If in-house teams conduct inventor interviews, compile prior art research, and prepare a thorough factual record, outside counsel can operate more efficiently, reducing the need for extensive revisions. This streamlined process not only enables law firms to stay within a constrained budget but also boosts outcomes, giving in-house professionals a tangible incentive to excel.

If organizations recognize that such a collaborative framework yields better patent coverage and strategic positioning—while also controlling costs—then the contributions of in-house patent personnel become even more vital. AI is set to amplify these benefits, allowing companies to meet the demands of a competitive global patent landscape more effectively than ever.

Why AI Can Make Patent Attorneys Twice as Valuable

최근 판례를 통해 살펴본 영문라이센싱 계약 Draft 주의점

 

지난주 금요일 라이센싱 계약서를 검토하는 계약법무팀이나 라이센싱계약팀이 눈여겨봐야 하는 판결이 미국 연방순회법원에서 나왔다.

 

종종 라이센시에게 Non-Assert Provision, a Damages Provision, Most Favored Licensee Provision은 중요한 key term이 된다.

 

그러나 draft의 시제를 놓치거나 문장이 장황해지거나 조항간 일관성결여로 모호해지게 되면 의도하는 효과가 발생하지 않게되므로 세심한 주의가 필요하다.

 

Wi-LAN과 Ericsson간의 특허소송에 관한 것이었다. 쟁점은 Wi-LAN가 4건의 Wi-LAN특허를 Ericsson에게 라이센싱하면서 체결한 라이센싱계약에 포함된 Non-Assert Provision, a Damages Provision, Most Favored Licensee Provision 이 세가지 조항에 대한 것으로 Wi-LAN가 라이센싱 계약체결이후 취득한 특허를 가지고 추가로 특허공격할 수 있는지에 대한 것이었다.

 

문제가 된 조항은 아래와 같다.

The Non-Assert Provision provided that Wi-LAN would not bring an action against “[Ericsson] PRODUCTS which would, but for this Agreement, infringe any WI-LAN PATENTS.”

 

The Damages Provision provided that “With respect to patents other than the WI-LAN PATENTS … WI-LAN hereby agrees that no damages shall accrue against [Ericsson] … for infringement of any patents that, on or after the EFFECTIVE DATE, are owned or controlled by WI-LAN where liability results from…[Ericsson’s] UMTS/HSPA PRODUCTS….”

 

The Most Favored Licensee Provision provided that “In the event Wi-LAN owns or controls the licensing of patents not already addressed under this Agreement and which are infringed or alleged to be infringed by UMTS/HSPA PRODUCTS, … WI-LAN will grant to [Ericsson] … a license at most-favored licensee status.”

 

이에 대한 1심 Texas법원과 Florida법원은 서로 의견을 달리했으나 결국 연방순회법원은 Wi-LAN의 손을 들어주었다.

 

이와 관련된 Christopher의 글을 발췌하고 관련 링크를 올린다.

On appeal, the Federal Circuit agreed with the Texas court and concluded that Wi-LAN was not precluded from asserting the four newly acquired patents in the Texas action and was neither obligated to license the patents asserted in the Texas action nor the patents asserted in the Florida action. In evaluating the obligations and intent of the parties under the Agreement, the Federal Circuit compared the language of the Non-Assert Provision to the language of the Damages Provision. The court noted that while the Non-Assert Provision only stated that Wi-LAN would not assert an action against Ericsson’s UMTS/HSPA products that infringe the WI-LAN PATENTS, the Damages Provision permitted Wi-LAN to seek damages for the sale of UMTS/HSPA products that infringed patentsother than the WI-LAN PATENTS.

 

Based on the language used in those provisions, the Federal Circuit inferred that the Non-Assert Provision was only intended to apply to the WI-LAN PATENTS and not any other Wi-LAN patents. As to the Most Favored Licensee Provision, the court concluded that this provision did not apply to the patents asserted in either of the Texas or Florida actions because the provision were drafted in the present tense and was directed to patents that “Wi-LAN owned or controlled as of the effective date of the [Agreement].” As a result, the Federal Circuit concluded this provision would not extend to after arising or future Wi-LAN patents.

Copyright © CHINSU LEE, but may partially include other’s copyright, This article is for informational purposes and is not intended to constitute legal advice.