ChatGPT, based on artificial intelligence, has been a hot topic in the daily press for several months now. Thousands of people around the world are testing its capabilities, starting with the free version 3.5 and the paid version 4.0 (as of the date of this article's publication). The newer version 4.0, despite being paid, is a more advanced model than 3.5. GPT-4 can generate more precise and complex answers to user questions, has more parameters and better learning from available data, has a broader knowledge base because it was trained on more data, is more efficient, and supports a wider range of languages. You can read about the advantages of GPT-4 on the official OpenAI website [https://openai.com/product/gpt-4].
For some, chatGPT is a tool that streamlines work and provides a source of knowledge, while for others, it's a reason to worry about the future. Artificial intelligence is increasingly becoming a reality that many people, including entrepreneurs and universities, will have to navigate. The pace of change can be confusing, but it can also motivate people to learn the secrets of this revolutionary solution. Much has been written about AI and chatGPT, and I'll add my two cents on the subject.
Is the content generated by chatGPT accurate?
The use of a new AI-based tool in trade also raises legal concerns. ChatGPT's speed allows for increased productivity, but it also carries certain risks.
From the Terms of Use (Terms of Use) [https://openai.com/policies/terms-of-use] published by OpenAI, which provides chatGPT, shows that artificial intelligence and machine learning are still developing, OpenAI is working on improving services to make them more accurate, among other things. My observations show that GPT-4 is not yet performing well in legal matters. The GPT-4 training period ended in September 2021. Therefore, for this reason alone, not all data can remain current or keep up with events that took place in 2022 or at the present time. You also need to learn how to ask questions correctly to get the most accurate answers you want. It is true that GPT-4 "passed" the American Bar Exam [https://www.forbes.com/sites/johnkoetsier/2023/03/14/gpt-4-beats-90-of-lawyers-trying-to-pass-the-bar/] however, Polish legal provisions or rulings were not necessarily included in his training data set – therefore, in our conditions, his answers may be subject to greater uncertainty.
The Terms of Use clearly state that chatGPT may generate incorrect output, and the user should assess the accuracy and correctness of this data. This means that using output obtained through chatGPT without verification may be risky for use at work, university, or elsewhere. For example, the mayor of an Australian town threatened to sue OpenAI for damages to its reputation [https://www.rp.pl/sady-i-prokuratura/art38299381-tomasz-pietryga-prawne-zderzenie-ze-sztuczna-inteligencja].
Example differences in the obtained output data
Below I show examples of differences in the output data obtained from GPT-3.5 and GPT-4, in which I deliberately used a query that was not necessarily linguistically and grammatically correct.
I asked a simple question "Which courts in Poland deal with divorces?”
The GPT-3.5 response was as follows:
It would seem that the question was clear, but it was enough to change the query to "Please specify which courts deal with divorces in in Poland” and the result is completely different:
There is a significant difference 😊 so it is worth remembering that the content of the question may generate incorrect answers.
In turn, GPT-4's answer to the question "Which courts in Poland deal with divorces?” sounded like this:
Using the query "Please specify which courts deal with divorces in Poland.” GPT-4's response was as follows:
If you phrase your query correctly, you might even receive information from ChatGPT that, in theory, it shouldn’t provide based on its operating principles [see https://www.linkedin.com/posts/gptai_activity‑7054401412054581248‑6y0Q/?utm_source=share& utm_medium=member_ios] A big plus for ChatGPT is that it suggests to users that it’s worth consulting a lawyer 😊 engineering prompt [https://pl.wikipedia.org/wiki/In%C5%BCynieria_podpowiedzi].
Entering the query correctly may even result in receiving information from chatGPT that theoretically should not be provided according to its operating assumptions [see. https://www.linkedin.com/posts/gptai_activity-7054401412054581248-6y0Q/?utm_source=share&utm_medium=member_ios]
Entering the query correctly may even result in receiving information from chatGPT that theoretically should not be provided according to its operating assumptions [see.
Intellectual property law and chatGPT
Who owns the copyright?
Another issue related to the use of output data obtained from GPT chat is who created the content and who holds the copyright. Current Polish and EU legislation has not fully addressed the legal nature of content generated by chatGPT. Many uncertainties remain in this regard. Polish law stipulates that a work, within the meaning of copyright law, can only be created by a human, and chatGPT is not, after all, a human. In the case of chatGPT, output content is generated based on tasks or queries entered by a human.
OpenAI's Terms of Use stipulate that the user owns all input data (i.e., data entered by the user into chatGPT), while OpenAI transfers all rights to output data (i.e., data received via chatGPT) to the user. This stipulation in the Terms of Use should be taken with a grain of salt; no binding legal solutions have yet been developed, and each case should be interpreted individually. Furthermore, the user will ultimately be responsible for the content they obtain from chatGPT, particularly with regard to whether it infringes third-party rights, including intellectual property rights.
Interestingly, the US Copyright Office has published guidelines for registering copyrights for works containing materials generated by artificial intelligence [https://www.copyright.gov/ai/ai_policy_guidance.pdfThe issue of copyright regulation and other aspects related to the use of artificial intelligence is a topic for a separate post, which will be published soon. While many positions are currently being presented, there is no unified voice on this matter. Intellectual property and artificial intelligence are still a topic that requires in-depth study by lawyers.
Can plagiarism occur if chatGPT output is used?
In my opinion, it should also be taken into account that content obtained from chatGPT may pose a risk of plagiarism. Moreover, several people asking the exact same question may receive the same or similar answers. ChatGPT's knowledge is based, among other things, on resources from the internet and books, however, in version 4.0, ChatGPT does not currently provide the source on which it based a specific answer (this is different when using the Bing search engine Bing search engine provided by Microsoft in the Edge browser).). OpenAI's Terms of Use stipulate that intellectual property infringement may be reported by sending a notification. Infringing content may be removed, and accounts of users who repeatedly infringe may be terminated. Universities will also face challenges in using anti-plagiarism systems to detect AI-generated content.
What about trademarks?
When discussing intellectual property, we can point to another solution based on artificial intelligence – next to GPT-3.5 and GPT-4. This is a tool Midjourney, which allows you to create images from short text descriptions entered by the user. The effects of some images can be impressive. This is probably one of the tools that can improve the work of graphic designers, including those who provide custom logos for a given company. Who wouldn't want an eye-catching logo for their business that can be created quickly? A logo can often be a trademark subject to protection. In the case of Midjourney It also remains unclear who will own the copyright to the logo created using it. This is an important issue, as the holder of an earlier personal or property right may file an objection to the trademark application or file a motion to invalidate an already granted trademark protection right.
Returning to chatGPT, it's possible that it could be used to create a slogan or wordmark, which would then be registered with the patent office as a trademark. Using appropriate chatGPT plugins would also allow for the creation of images, which could constitute a distinctive mark. In this case, problems could arise related to the potential use of a work to which someone else holds personal or property rights.
Trade secrets, know-how, confidential data and personal data and chatGPT
The ease of access to chatGPT also tempts people to test its capabilities in their daily work. Many employees and collaborators possess information that may constitute trade secrets, know-how, or confidential information as defined in a confidentiality agreement.
Pursuant to the Act on Combating Unfair Competition, a trade secret is understood as technical, technological, organisational information of an enterprise or other information of economic value which, as a whole or in a specific combination and set of its elements, is not generally known to persons usually dealing with this type of information or is not easily accessible to such persons, provided that the person authorised to use or dispose of the information has taken, with due diligence, steps to keep it confidential.
Please note that, as stated in the Terms of Use, Open AI may use content to provide and maintain services. ChatGPT is constantly learning as it responds, so it processes user-entered data. Therefore, when using chatGPT to streamline daily work, you should be aware of the risk of disclosing confidential or trade secret information. The same applies to professional privileges, such as legal counsel, attorney, or detective privilege. This may involve professional liability and breaches of professional ethics.
What could be the consequences of disclosing trade secrets or confidential information?
It is worth remembering that disclosing business secrets or confidential information may result in the termination of the employment contract or cooperation agreement [more: https://www.prawo.pl/biznes/zwolnienie-z-pracy-za-rozmowe-z-chat-gpt-odpowiedzialnosc,520711.html]. Using inaccurate or outdated data generated through chatGPT may also cause problems, possibly resulting in termination of employment or a breach of professional conduct. The user is responsible for the use of data obtained through a query submitted to chatGPT. Finally, using incorrect data may result in liability for damages.
Personal data
Moreover, some queries and content entered in the GPT chat may contain personal, private, or sensitive data. It's important to remember that chatGPT is a solution provided by a Californian company headquartered in the US, meaning it's outside the GDPR (GDPR) area. This also means that personal data entered into chatGPT will be processed outside the European Union. Therefore, we must not overlook this aspect. Furthermore, users voluntarily provide their personal data to use chatGPT. They are therefore consenting to the collection and processing of their personal data by an entity outside the European Union.
Summary
In the near future, the use of artificial intelligence solutions in the daily work of businesses and universities will require the development of appropriate regulations and policies for the use of this type of technology. Training employees and collaborators on the daily use of chatGPT and the associated responsibilities will also be a good solution. EU and Polish law will undoubtedly need to adapt to the reality that, in addition to traditional works, content generated using artificial intelligence is also emerging.
To sum up, the following legal risks can be mentioned when using chatGPT:
- possible infringement of intellectual property rights and the essence of industrial property solutions
- possible disclosure of business secrets, know-how, confidential information and personal data
- disclosure of legally protected secrets
- transfer of personal data outside the European Union
- violation of professional ethics, behavior inconsistent with the principles of social coexistence
- liability for damages related to infringements or resulting from the use of output data without checking its correctness.
What are the other risks of using chatGPT? Share your experiences and leave your answers in the comments.
Please feel free to contact me law firm.
The main photo at the beginning by Markus Winkler on Unsplash
