Conclusion

Errant Intelligence aims to promote a broader and more open understanding of AI and machine learning, moving away from its biological underpinnings and towards a media-theoretical view. This also implies a change of perspective in how we “imagine” an artificial intelligence. While the current discourse is stuck in the imaginary of the brain, thereby preventing an examination of the symbolic preconditions of intelligent machinery, the history of computing offers a wide field of inspiration for moving beyond such a connectionist paradigm. According to Alan Turing, all computation can be performed by a universal machine, which embodies the accumulation of logical thought. Therefore, computation and logics are in fact two sides of the same coin. The Turing machine not only inherits the paradoxes of logical reasoning, particularly with regard to the inherent incompleteness of axiomatic systems and thus of all knowledge, but is also constituted by them. And since AI systems run on computer algorithms, which in turn represent logical operations, machine learning in its current form is still bound to its symbolic prehistory. Neural networks, deep learning, and LLMs need to be thought with and through the “logico-mathematical” paradigm ¹ because logic gates and Boolean algebra provide the computational basis of these systems. Despite, or precisely because of, recent claims that machine learning algorithms are “the automated discovery of abstraction,” ² we need to recall the intellectual history of abstraction. Indeed, the very idea of comparing neural systems to microprocessors stems from the abstraction that neurons operate like logic gates. ³ So rather than biologising machine learning, the aim of this book is to highlight the symbolic in all intelligence. This does not mean that we simply return to symbolic AI. Rather, as has been argued, we must work through the abstract forms of machine intelligence in order to move beyond its current imaginary as a biological entity. ⁴ We need to acknowledge the alien, paradoxical, and often strange aspects of computation, so that we can truly learn something new. For if machine learning is transforming the production of knowledge, as many claim, then that knowledge could also be used to transform our understanding of machine learning. ⁵