What do transparency obligations under the EU AI Act mean for AI systems?

Transparency, in the context of AI systems, refers to the ability to understand how they arrive at their decisions. This includes knowing what data they're trained on, how they process that data, and the rationale behind their outputs. 

Transparency is very important because of the following reasons:

• Building trust - When users lack understanding of how AI systems work, they're less likely to trust their outputs. Transparency fosters trust by demystifying the decision-making process and allowing users to understand the basis for AI-driven actions.
• Ensuring accountability - Clearer understanding of AI behavior allows for better accountability. This is especially important for high-risk AI systems that can have significant impacts on people's lives (e.g., credit scoring, recruitment tools). Transparency allows for tracing decisions back to the underlying data and algorithms, enabling identification of potential issues or biases.
• Mitigating bias  - AI systems can inherit biases from the data they're trained on, leading to discriminatory or unfair outcomes. Transparency helps to identify and mitigate potential biases in AI decision-making. By understanding the data used and the algorithms employed, developers and regulators can take steps to ensure fairness and inclusivity in AI development and deployment.

The EU AI Act categorizes AI systems based on their risk level (high-risk, low-risk, minimal risk) as outlined in our article.  The transparency requirements become more stringent as the risk level increases, and logically, the strictest requirements have high-risk AI systems.

These systems, like facial recognition software or credit scoring algorithms, require a level of transparency that allows providers and users to "reasonably understand the system's functioning." This might involve providing information on:

• The type of data used to train the AI model.
• The specific algorithms employed and their decision-making logic.
• The potential limitations and uncertainties associated with the model's outputs.

It's important to know the challenges of achieving transparency in complex AI models.  These models can be complex networks of algorithms, making it difficult to fully explain every step of the decision-making process. 

Even if possible, balancing business secrets with transparency is tough, as no one likes to reveal how their algorithms function.

However, the EU AI Act encourages developers to strive for a level of transparency that is "sufficient" for understanding the system's core functionalities and potential impacts.

Here are some specific examples of transparency obligations under the EU AI Act:

• Disclosing interaction with an AI system - users interacting with AI systems, like chatbots or virtual assistants, should be clearly informed that they're not communicating with a human. This ensures informed consent and allows users to adjust their expectations accordingly.
• Labeling AI-generated content - AI-generated content, such as text, audio, or video (including "deepfakes"), needs to be clearly labeled as artificial to avoid misleading users. This protects consumers from misinformation and ensures responsible use of AI-generated content.

The transparency obligations outlined in the EU AI Act will undoubtedly have a significant impact on businesses developing and deploying AI systems within the European market. 

We’ve prepared a breakdown of some key implications to consider:

• Shifting development processes - To comply with the EU AI Act, AI businesses may need to adjust their AI development processes to incorporate explainability and transparency from the outset. This might involve using more interpretable algorithms, documenting training data and model behavior, and developing methods for communicating these details to users
• Focus on human-centered design - The emphasis on transparency necessitates a shift towards human-centered AI design. Developers will need to consider how users will interact with the AI system and ensure that the level of transparency provided is meaningful and easy to understand for the intended audience.
• Potential benefits - While complying with these regulations might require initial investment, there are also numerous benefits. Increased transparency can foster trust in AI systems, leading to wider adoption and user confidence. Additionally, a focus on explainability can improve model development itself by helping identify and address potential biases or limitations.

Looking ahead, the EU AI Act represents a significant step towards responsible AI development.  But it’s important to note that EU AI Act is just the first legal framework that regulates AI, and governments worldwide are proactively working to release their own frameworks that will regulate development, deployment, and use of AI.

By embracing transparency and prioritizing human-centered design, businesses can ensure their AI systems not only comply with regulations but also contribute to building trust and shaping a positive future for AI in Europe. We’ve already covered this topic in our article Building trust and credibility: How AI compliance elevates your market position?.

Business customers are seeking AI systems and applications they can trust. Your adherence to the EU AI Act will help distinguish your AI system or application amidst the vast array of others. However, the initial step is understanding the level of risk associated with your AI system as defined by the EU AI Act. To simplify this for you, we've prepared a free self-assessment tool. This tool will aid you in comprehending the risk level of your AI system and guide you on the subsequent steps to comply with the new regulations ahead of your competitors.

Taking this assessment should not require more than 4 minutes of your time, yet it will offer valuable insights on positioning yourself in this new era of regulated AI. You can take the assessment here. Should you have any questions, feel free to reach out to us.