Artificial intelligence has undergone a tremendous transformation, evolving from a neglected field to a domain that is now filled with both excitement and apprehension. The regulations surrounding these intelligent systems, which have surpassed human capabilities in certain areas, are still uncertain. In order to ensure that the benefits of AI, such as advancements in science, medicine, and overall improvement in quality of life, outweigh the fears associated with it, it is essential to make the right decisions regarding the protection and control of this technology.
Over the past year, the introduction of AI chatbots like OpenAI’s ChatGPT has sparked significant concern. Prominent figures from various fields have issued warnings about the far-reaching changes AI may bring, from transforming workplaces and classrooms to potentially shaping our entire lives. The magnitude of these concerns is exemplified by Russian president Vladimir Putin’s claim that the leader in AI will become the ruler of the world. Industry leaders have also emphasized the importance of addressing the risks of unconstrained AI. Consequently, legislative efforts are already underway to tackle these issues.
The European Parliament recently approved the new Artificial Intelligence Act, which requires AI systems like ChatGPT to implement safeguards and disclosures. These safeguards include avoiding the use of subliminal techniques beyond human consciousness and not exploiting specific groups of people due to vulnerabilities related to age, physical or mental disability. Furthermore, the act emphasizes the need to mitigate foreseeable risks to health, safety, fundamental rights, the environment, democracy, and the rule of law.
A significant question that arises globally is whether authors or performers should provide consent for the use of their works as training data for AI systems. Some governments have introduced text and data mining exceptions to copyright laws, allowing AI systems to train on information owned by others. However, these exceptions have faced opposition, particularly from copyright owners and critics who aim to limit or diminish these services. Furthermore, concerns have been raised about AI’s potential biases, social manipulation, income and employment losses, disinformation, fraud, and even the prospect of catastrophic consequences for humanity.
Multiple countries have adopted diverse approaches to intellectual property rights associated with AI training data. The U.S., for instance, is currently grappling with lawsuits to determine the extent to which fair use exceptions apply to AI training data. In Europe, a 2019 directive on copyright in the digital single market introduced exceptions for text and data mining, while allowing copyright owners to prevent their works from being used in commercial services. Singapore also created an exception in its copyright law specifically for computational data analysis.
China has indicated its intention to exclude from training data any content that infringes on intellectual property rights. However, determining the copyright status of the vast amount of data commonly used for training AI systems, often obtained at massive scales from various online sources, can be challenging. Many countries have not yet taken a clear stance on exceptions for text and data mining within their copyright laws.
While some countries, like India, are not yet ready to regulate AI, they nevertheless aim to support their domestic industries. It is crucial to avoid a one-size-fits-all approach when crafting laws and regulations in this field. Such an approach could inadvertently impede scientific research and development if the same rules governing recorded music or artwork were applied to scientific papers and medical data.
The massive scale of data involved in training large language models poses its own challenges. The latest version, GPT-4, used an undisclosed dataset for training, while previous versions like GPT-3 were trained on 45 terabytes of data. Clearing copyrights for training data, especially in larger projects, can be difficult and identifying the rights holders can be nearly impossible. Consent requirements for copyrighted work in scientific research could provide publishers with significant control over data usage.
Beyond consent, the challenges of credit and compensation are also evident. The cost of litigation for copyright or patent infringements is high, and managing these issues can be complex. However, a well-managed AI program has the potential to implement benefit-sharing in certain domains, such as proposing an open-source dividend for the successful development of biomedical products. Decentralizing data used for AI training and implementing privacy protections and data governance can contribute to addressing some of these challenges.
It is essential to recognize that intellectual property rights assigned to training data are primarily governed by national regulations, while the race for AI development is global. AI programs can be executed anywhere with electricity and internet access, without requiring extensive resources. Therefore, companies operating in countries with onerous data acquisition and usage obligations may face competitive disadvantages compared to those in freer environments. This raises consequential questions about the future of AI and the global competition surrounding it.
In conclusion, as laws and regulations pertaining to AI continue to emerge, it is crucial to approach the topic with caution and avoid a one-size-fits-all approach. Balancing the risks and benefits associated with AI technology requires thoughtful considerations of issues like consent, copyright, credit, and compensation. By addressing these concerns appropriately, we can harness the potential of AI while minimizing the negative consequences and ensuring a fair and equitable future.
Note: The views expressed in this article are solely those of the author and do not necessarily reflect the opinions of Scientific American.
Author Information: James Love is the director of Knowledge Ecology International, a non-profit organization focused on financing, intellectual property, prices, and access to medical advances.
Disclaimer: This is an opinion and analysis article. The information presented here is for general informational purposes only and should not be relied upon as legal, financial, or professional advice.
References: This article provides insights and analysis based on various sources, news articles, and publications related to artificial intelligence, copyright law, and intellectual property rights.