On 8 June 2023, Mireille Hildebrandt will deliver a keynote at the Ninth International MBR Conference in Rome (7-9 June, Sapienza University) on Model-Based Reasoning, Abductive Cognition, Creativity (Inferences & Models in Science, Logic, Language, and Technology).
The topic of the keynote is "Abduction in LLMs, the Issue of Real World Navigation" and the abstract reads as follows:
"Are LLMs merely stochastic parrots that may infer reasoning patterns from our datatified language behaviours or should we expect them to develop reliable reasoning based on logic and causality? Taking C.S. Peirce seriously, I will raise the question of how reasoning relates to abduction and how language models relate to real world meaning. This will involve a third perspective to assess both reasoning and meaning, next to logic and causality. This third perspective is the performative effect of speech acts, closely related to the consequentialist dimension of pragmatism. I will argue that the utilitarian understanding of consequentialism pervades Anglo-American discourse, whereas the pragmatist understanding is better aligned with real world navigation. For LLMs to develop abduction they would need to develop what roboticists Pfeifer and Bongard called ‘complete agency’, meaning they would need to be built to survive in the real world without being under external human control. This, in turn, raises the issues of the intentionality (Brentano) of these systems and their ‘conatus’ (Spinoza), which cannot be assumed and seems highly unlikely. I will conclude that LLMs may become useful tools to the extent that their integration is not made dependent on manipulative business models, while we should not expect them to arrive at abduction, artificial general intelligence or complete agency. In the meantime it is key to acknowledge the abductive creativity of those who develop and deploy these systems and the way that LLMs may trigger and challenge such creative abduction."
More information about the event & the keynote can be found here.