AI Act is here – there is time to explain, get into the car.
Artificial Intelligence (AI) has become an integral part of our lives, from virtual assistants to self-driving cars. As AI systems become more sophisticated, so do the challenges associated with their deployment. In response, the European Union (EU) has introduced the AI Act, a comprehensive regulatory framework aimed at ensuring the safe and ethical use of AI technologies.
The EU AI Act
The AI Act is a groundbreaking proposal by the European Commission to create a legal framework for AI governance. It categorizes AI systems based on their risk to society, from unacceptable to minimal, and sets out corresponding requirements.
High-risk AI systems, such as those used in education or transportation (critical infrastructures), are subject to stringent obligations. These include accuracy, cybersecurity, and explainability. The Act recognizes that without understanding how AI decisions are made, we cannot fully trust or control these technologies.
The case of Explainable AI
Explainable AI (XAI) refers to the methods and techniques that make the outputs of AI systems understandable to humans. It’s not just a technical necessity; it’s a societal one. Here’s why:
- Transparency: XAI demystifies AI decisions, showing the ‘why’ behind the ‘what.’
- Trust: When people understand AI decisions, they are more likely to trust and accept them.
- Accountability: If an AI system causes harm, XAI helps trace the root cause, ensuring accountability.
- Regulatory Compliance: The AI Act mandates explainability for high-risk systems, making XAI a legal requirement.
This is not new demand, as this challenge has been recognised long before EU AI act, and generated number of approaches. These are some of the most popular explainability methods:
- LIME (Local Interpretable Model-Agnostic Explanations) is a technique that helps us understand and interpret the predictions made by machine learning models. It does this by creating a simple model around the prediction we want to explain, which is easier to understand than the original complex model. LIME perturbs the input data and observes the corresponding changes in the output, thereby learning which features significantly influence the prediction.
- SHAP (SHapley Additive exPlanations) is based on the concept of game theory and provides a way to measure the contribution of each feature in the prediction made by a model. It assigns a value to each feature indicating its importance. The SHAP value is the average marginal contribution of a feature across all possible combinations.
- Attention Mechanisms in neural networks allow the model to focus on different parts of the input for different tasks. They are particularly useful in tasks like translation, where the model needs to pay more attention to specific words at different times. Attention mechanisms weigh the importance of different inputs and decide where to focus at each step of the task.
These methods are powerful tools in the field of explainable AI, providing transparency into the decision-making process of complex models. They help us trust the predictions made by AI systems by giving us insights into their reasoning.
Bridging the Gap
The AI Act and XAI are two sides of the same coin. The Act mandates explainability, and that high-risk AI systems must be transparent and explainable, where XAI provides the means to achieve it – some of them being LIME, SHAP, and attention mechanisms. But it’s not without challenges. Achieving explainability without sacrificing performance is a delicate balance – there is no sense of explaining poor performing model. On the other hand, there is the needed for human understanding, where for complex models like Neural Networks (black boxes) we still don’t have universal way of explaining them. Yet, explainability and understanding model decisions are a non-negotiable aspects of ethical AI development.
Conclusion
The AI Act is a significant step towards responsible AI. It acknowledges the power of AI and the potential perils it poses. Explainable models are not just a regulatory checkbox but a cornerstone of ethical AI. As we move forward, the synergy between the AI Act and XAI will shape the future of AI, making it as accountable as it is advanced.
Author: Marek Brynda
Photo source: Unsplash
