Imagine an AI that doesn't just play games or write stories, but actually makes scientific discoveries and explains them to us. This isn't science fiction; it's the ambitious goal explored in Antonio Norelli's PhD thesis, "Artificial Scientific Discovery." Norelli's research journey begins with Olivaw, an AI trained to master the game of Othello, much like the famous AlphaGo. Olivaw achieved world-class level play using minimal resources, demonstrating that AI can learn complex strategies from scratch. But a crucial question arose: while Olivaw could *play* Othello brilliantly, it couldn't *explain* its strategies. This realization led to the development of "Explanatory Learning" (EL), a framework that emphasizes the importance of language and interpretation in scientific discovery. A true artificial scientist, Norelli argues, must be able to understand and generate explanations, just like human researchers. To tackle this, Norelli and his colleagues created "Critical Rationalist Networks" (CRNs). CRNs generate hypotheses ("conjectures") about how a phenomenon works and then test those hypotheses against data, mirroring the scientific method. Tested on a custom-designed game called Odeen, CRNs were significantly better at discovering the underlying rules of the game than standard AI models, showing the power of the EL approach. Going deeper into how AI understands language, Norelli's thesis explores how to create common representations for different kinds of information, like images and text. He introduces ASIF, a technique that aligns pre-trained image and text models without any further training, creating a surprisingly effective multimodal model. Finally, the thesis examines the limitations of current Large Language Models (LLMs) like GPT. While impressive in many ways, LLMs struggle with key aspects of scientific reasoning, such as understanding cause and effect, recognizing their own limitations, and critically evaluating information. A custom-designed task within the Big-Bench project highlighted this gap, with LLMs performing no better than random chance while humans excelled. Norelli's work presents a compelling vision for the future of AI, one where machines don't just mimic human intelligence, but participate in the process of scientific discovery, pushing the boundaries of human knowledge alongside us. It underscores the need for AI systems that can not only reason and learn, but also explain, interpret, and actively engage with the world around them.