How can artificial minds help us understand our own minds? Blank will discuss two aspects of this question in the domain of language, using both simple and complex artificial intelligence (AI) systems. First, AI systems trained on highly constrained input can help disentangle representational formats that are difficult to differentiate in human minds. As an example, Blank will ask what kinds of world knowledge (e.g., about dogs) are embedded in our knowledge of language (e.g., how the word “dog” is used). Blank will demonstrate that a simple AI system (a word embedding), which is only exposed to patterns of word co-occurrences, acquires complex common-sense knowledge, which is represented in an intuitive and structured way. Because humans also track word co-occurrences, such world knowledge is also represented in our mental lexicon. Second, Blank will emphasize that such inferential leaps hinge on how closely an AI system represents language “like us”. This resemblance is easier to establish for word embeddings, but harder for complex Large Language Models (LLMs, e.g., GPT models). Two studies comparing LLMs to humans will illustrate this point. One study asks whether semantic information can “penetrate” and influence syntactic processing in LLMs—like it does in humans—or whether some syntactic processing stages in LLMs are “encapsulated” from meaning. The second study asks whether LLMs represent a fundamental aspect of linguistic meaning: distinguishing between agents and patients in sentences. These studies reveal both similarities and differences between LLMs and humans.