Modern Artificial Intelligence (AI) is more than just a clever program. It creates increasingly complex models of our world. These models are so complicated that even experts can no longer fully understand them. Such opaque systems are the basis of many new applications. They raise important questions.
This is important because these opaque systems become "black boxes." A black box is a system whose operation is not understood. If even the developers don't know why an AI makes a decision, it's hard to find errors. It's also hard to detect biases. No one can then take responsibility for the consequences. This destroys trust and carries great risks. These risks grow the more power we give to the systems.
The well-known MIT Technology Review warns in its latest issue. It states that the complexity of today's AI models exceeds human understanding. These models shape our lives in many areas. However, it is becoming increasingly difficult to understand their exact workings. Their decision-making processes are also difficult to comprehend. This applies to large AI language models. It also applies to specialized systems in medicine or finance.
For you as a user, this means: You have to rely on AI systems. No one can oversee their internal processes anymore. If your smartphone assistant reacts unexpectedly, the reason is hard to understand. The same applies if an AI-powered recommendation system makes strange suggestions. For example, a recommendation system suggests products to you. A dependency on technology arises. We can only control this technology very little. This affects jobs, decisions, and the flood of information you encounter.
Companies face even greater problems. Firms use AI models in important processes. Examples include fraud detection, personnel selection, or product development. If the AI's decisions are no longer comprehensible, major problems arise. These include difficulties with compliance and liability issues. It is difficult to conduct internal audits. It is also difficult to explain to customers why an AI made a certain choice. The risk of uncontrollable errors and reputational damage increases. Therefore, more and more human approvals are needed before the AI acts. This is called 'Human-in-the-Loop' approvals.
Despite the risks, complexity also creates new opportunities. These highly developed models can recognize patterns. They can solve problems that would be too difficult for humans alone. They promote research. They enable innovations in areas like materials science or drug development. In doing so, they find solutions that we had not previously considered. The ability to replicate complex systems can help us tackle bigger challenges.
