1. The article discusses the issue of "understandability" in modern AI systems that are based on complex neural networks. Understandability refers to the ability of humans to understand how the AI system arrives at its decisions and predictions. This is important because as AI systems become increasingly sophisticated and pervasive, their decision-making processes can have significant impacts on human lives and society as a whole. If the inner workings of these systems are not transparent and understandable, it can be difficult to determine why certain decisions were made, and to identify and address any biases or errors that may be present in the system. The lack of understandability also raises concerns about accountability and the potential for misuse or abuse of AI systems.

2. The article explains that classical AI, which is based on rule-based systems, is much less likely to run into issues surrounding understandability compared to AI based on neural networks. The key difference between these two approaches is that classical AI systems rely on explicit rules and logic to arrive at their decisions, which can be easily understood and explained by humans. In contrast, neural network-based AI systems operate through complex, interconnected layers of artificial neurons, and the decision-making processes are not always transparent or easily explained. While neural networks can be highly effective in certain applications, their inner workings are often considered a "black box," which makes it difficult to understand why they arrive at certain decisions. This lack of transparency and understandability is a key challenge in the development and deployment of AI systems.

3. According to the article, the emergence of large data sets played a crucial role in making neural networks mainstream. This is because neural networks require large amounts of data to be trained effectively. The more data that is fed into a neural network, the better it can identify patterns and relationships between inputs and outputs, and the more accurate its predictions and decisions will become. In other words, large data sets enable neural networks to learn and improve over time, and to become more sophisticated and effective at a wide range of tasks, from image and speech recognition to natural language processing and decision-making. The availability of large data sets has therefore been a key driver in the development and success of neural network-based AI systems.

4a. The author of the article alludes to the fact that full understandability is not just lacking in modern AI, but also in biological intelligence in the following quotes:

• • "Even the people building these systems cannot fully explain their behavior. Neural networks, for instance, are composed of layers of interconnected nodes that weigh the input they receive and pass signals to other nodes, until they generate an output that solves a problem—like recognizing a face or translating a language. But the internal workings of large neural networks are a mystery. Even their creators do not know why some nodes and connections emerge as critical to the solution of a problem, while others seem irrelevant."
  • "This opacity is not unique to artificial intelligence. Natural organisms like our brains are also vastly complicated systems whose workings are not completely understood."
  • "Indeed, AI’s lack of transparency is often held up as its Achilles’ heel, the thing that makes people distrustful or even fearful of it. But it is worth remembering that humans are also deeply opaque systems."

4b. There may be a fundamental difference between biological and artificial intelligence when it comes to limits on understandability and ways in which society can deal with those limits. While both types of intelligence can be opaque and difficult to fully understand, biological systems have evolved over millions of years through natural selection, and their complexity is the result of a long and complex evolutionary process. On the other hand, artificial intelligence is designed by humans, and we have the ability to modify and control its behavior to a greater extent than we can with biological systems. As a result, society may have more control over the development and deployment of AI systems, and may be able to take steps to ensure that they are transparent and accountable, even if full understandability is not possible. Nonetheless, it is important to recognize that there are limits to our understanding of both types of intelligence, and we need to continue to explore ways to make these systems more transparent and accountable.

5. Dennett's suggestion that the creation of systems capable of performing tasks their creators do not know how to do is a natural part of the evolution of intelligence itself could be interpreted as a reflection of the potential of AI to transcend human intelligence. In this view, an AI system that can perform tasks that its creators do not know how to do could be seen as evidence of true artificial intelligence, in which the system is not simply a tool but has its own internal logic and decision-making processes.

5a.Yes, I agree that it is a sign that we have built a truly intelligent AI system when we don't know how to do the things the AI does. However, the suggestion also raises concerns about the limitations of human understanding and the need for transparency and accountability in the development and deployment of AI systems. If we create AI systems that we do not fully understand, there is a risk that they may produce outcomes that we cannot anticipate or control, potentially leading to unintended consequences or negative impacts on society. Therefore, it is important to balance the potential benefits of AI with the need for transparency and accountability in its development and use.