Artificial Intelligence (AI) has brought about new legal and ethical challenges. The EU’s AI Act, proposed in April 2021, seeks to establish a robust framework for regulating AI, ensuring its ethical usage while promoting trust and transparency. This guideline aims to help businesses grasp the AI Act's main points, focusing particularly on large language models (LLMs) and their potential impact.
1. Overview of the AI Act
Purpose of the Act:
Establish a regulatory framework for AI systems.
Promote ethical usage, safeguarding fundamental rights, and enhancing trust and transparency.
Timeline: Finalisation is expected around mid-2024, followed by an 18-month transposition period.
Definition of AI: The Act adopts a broad definition, aiming to be as "technology neutral and future proof" as possible.
Risk Categories in the Act:
Unacceptable risk: AI systems posing serious threats to safety, rights, and livelihoods, e.g., critical infrastructure, law enforcement.
High risk: AI systems potentially causing significant harm or infringing upon fundamental rights, e.g., AI-based hiring tools, facial recognition.
Limited risk: Systems with potential risks but not as severe as the above two categories.
Minimal or No Risk: Systems that do not fit the above categories and have no compliance requirements.
2. Understanding Generative AI and LLMs
Generative AI:
AI systems capable of generating human-like content, such as images, music, or text.
They analyze patterns in vast data to produce content, sometimes surpassing human capabilities.
LLMs:
A subset of generative AI models focusing on natural language processing.
Can generate text, answer questions, and engage in conversations.
Examples include OpenAI's GPT-3/4.
3. Implications of the AI Act on LLMs
Risk Category: Generative AI, including LLMs, falls under the "Limited Risk" category.
Key Provisions:
Data Governance: Ensure diverse and high-quality training data to avoid bias.
Transparency: Inform users of the AI system's artificial nature. For instance, a letter generated through AI might need a disclosure.
Accuracy and Reliability: Regularly monitor and test AI outputs, ensuring clear accountability for misleading or incorrect content.
Penalties: Companies face fines of up to €30 million or 6% of global income for non-compliance. Misleading documentation can also result in fines.
4. Key Considerations for Businesses
Intellectual Property (IP):
The Act doesn't address ownership of outputs from generative AI.
Businesses must evaluate IP rights when using AI tools and mitigate risks, e.g., by documenting creative processes and including AI provisions in contracts.
Be cautious of breaching third-party IP rights when using AI trained on copyrighted material.
Commercial Contracts:
Businesses should address AI-related matters, such as the use of AI, AI-generated output ownership, and AI-generated content liability.
AI Policies:
Implement policies governing AI usage, addressing AI's application areas, protecting IP/personal data, and managing associated risks.
Audit and Risk Assessment:
Determine the risk associated with AI systems, especially in areas like recruitment that fall under the high-risk category.
Data Protection:
Comply with data protection frameworks like GDPR, especially concerning automated decision-making.
Conclusion
The EU's AI Act seeks a delicate balance between safeguarding rights and fostering AI innovation. For businesses, understanding and adhering to this regulatory landscape is crucial. The proactive approach involves risk management strategies and continuous education about AI's evolving capabilities.
Note: The information provided is not legal, tax, investment, or accounting advice and should not be used as such. It is for discussion purposes only. Seek guidance from your own legal counsel and advisors on any matters. The views presented are those of the author and not any other individual or organization. Some parts of the text may be automatically generated. The author of this material makes no guarantees or warranties about the accuracy or completeness of the information.
Comments