Should we use AI and LLMs for Christian Apologetics?
The article argues that open-source large language models are fundamentally unreliable for generating factual truth due to their inherent architectural design. Unlike traditional open-source software, these models consist of opaque neural weights that cannot be audited or fixed. Their training processes are inaccessible, and their core function prioritizes plausible text generation over accuracy, meaning they inevitably hallucinate errors regardless of their openness or the data used. This technical limitation poses a significant risk to open data integrity, particularly in specialized fields like theology or apologetics. The author demonstrates through testing that even capable models fabricate details, such as quoting incorrect Greek texts and inventing provenance stories, while presenting false information with unwarranted confidence. This behavior undermines trust, as users may accept these plausible but false outputs as valid, thereby spreading misinformation rather than providing accurate, verifiable data. Furthermore, the reliance on such chatbots degrades the quality of information ecosystems by displacing rigorous research with convenient, yet erroneous, summaries. Unlike human experts who can admit ignorance or correct mistakes based on moral accountability, AI systems lack the capacity for truth-seeking or improvement. Consequently, integrating these tools into open knowledge platforms creates a liability, eroding trust in digital resources and promoting a culture where plausible fiction is mistaken for verified fact, which is counter to the goals of transparent and accurate open data initiatives.
Source: lukeplant.me.ukPublished on 2024-09-19