Rushdie over AI, for the time being
Salman Rushdie weighing in on AI's capability to write quality fiction is like Garry Kasparov sharing his views on the technology's ability to play chess — which the grandmaster did with fury after being beaten by IBM supercomputer Deep Blue in 1997: it wasn't the machine that was great, but he who had cracked under pressure. Rushdie's take this week in an article for a French literary journal was more measured, but equally derisive. Testing OpenAI's ChatGPT, the veritable writer described the LLM having 'no originality' and 'completely devoid of any sense of humour', a key yardstick of quality in 'serious' literature.
While a Rushdie-ChatGPT contest is not the same as a Kasparov-Deep Blue one — no one wins or loses in something as subjective as literature — the writer is both right and wrong.
Right, because LLMs have not developed sophistication and nuance — yet. AI is yet to figure out when to 'hallucinate' — be creative — and when to 'drop back' to plain reasoning. Most importantly, it hasn't yet figured out how to recognise this key essence of quality literature that the good writer knows through intuition — a product of talent and time.
For those wowed by AI's ability to produce writing that reads like Rushdie's The Moor's Last Sigh or a Tennyson poem, it is the novelty of mimicry that impresses.
But Rushdie is also wrong — or will be. Currently, LLMs work with extensive training data sets, much like writers do over decades of reading other writers and their styles. What great writers produce is a distilled product of their readings and own life experiences, and understanding their craft.
