Johannes Brinz

The philosophy of AI has a long-standing tradition of discussing brain duplicates and brain simulations as well as a tendency to blur the lines between the two. The distinction between simulating and duplicating the brain has become increasingly important with the emergence of “ʼneuromorphic computers”, hardware operating with bio-inspired mechanisms and interconnecting artificial neurons and synapses. This paper explores what it means to duplicate the brain rather than merely simulate it. I claim that while simulations share only a mathematical structure with their targets, duplicates have the same relevant measurable properties and are governed by the same relevant causal processes as the target system. I propose six criteria for brain duplication that are often not met by simulations, thereby offering a clearer understanding of the complexities involved in achieving brain duplication in artificial systems. Furthermore, this paper explores whether neuromorphic computers can duplicate neural computations by exploring the notion of models of computation. I submit that analog neuromorphic computers that use memristive technology are candidates for duplicating neural structures with respect to simple models of neural computation. Finally, I discuss five possible objections to my view.

Does the Chinese Room Argument (CRA) apply to large language models (LLMs)? The thought experiment at the center of the CRA is tailored to Good Old-Fashioned Artificial Intelligence (GOFAI) systems. However, natural language processing has made significant progress, especially with the emergence of LLMs in recent years. LLMs differ from GOFAI systems in their design; they operate on vectors rather than symbols and do not follow a program but instead learn to map inputs to outputs. Consequently, some have suggested that the CRA is no longer relevant in discussions surrounding artificial language understanding. Contrary to these authors, I argue that if the CRA successfully demonstrates that implementing a symbolic computation is not sufficient for language understanding, then it also shows that implementing an LLM is not sufficient for language understanding. At the core of my argument lies a thought experiment called “the flashcard sorter”.

The media often discuss artificial neural networks like ChatGPT or Amazon's Alexa, and policymakers grapple with regulating emerging technologies. However, the precise nature of "artificial neurons" remains ambiguous. Is this term to be understood merely metaphorically or does it refer to physical entities resembling biological neurons? While commonly understood as mathematical nodes in AI, the discussion extends deeper, particularly with the advent of neuromorphic engineering. This paper discusses whether artificial neurons are indeed neurons and what the potential implications are. Specifically, the paper addresses the question: I discuss three typesAre artificial neurons functional analogs of biological neurons? of artificial neurons against the background of the two main theories of biological function, viz. Causal Role and Selected Effects Theory. My thesis contends that while artificial neurons lack the morphological structure of biological neurons, certain types, especially those in neural circuits, exhibit functional parity, and thus can be considered functional analogs.