Hi ,
Most of what I know about British history, I learned from
Shakespeare, Jane Austen and Blackadder.
And this knowledge has served me surprisingly well across multiple quiz nights and games of Trivial Pursuit over the years.
But I wouldn't rely on this information for anything more important - lest I accidentally claim Edmund Blackadder to be an ancestor of King Charles III.
That's what encyclopedias are for.
Yet, to claim that Shakespeare, Austen or Blackadder *failed* for taking liberties with historical truths would be absurd. It would make as much sense as claiming Wikipedia *failed* by not being an engaging bedtime read.
Both types of works have completely different use cases, and each places different levels of emphasis on creativity and fact, accordingly.
Here's the
thing...
Tech experts are now starting to believe that the problem of AI "hallucinations" - that is, generative AI tools, such as ChatGPT, producing spontaneous lies - may never go away. Which means the outputs of such tools may never be 100% reliable.
Yet, expecting generative AI tools to be completely reliable shows a
misunderstanding of their fundamental use case.
The internet is full of websites containing facts, which can easily be accessed via your favourite search engine.
What sets ChatGPT apart from the Google search engine, though, is its ability to take those facts and mash them up in often unusual ways - allowing people to consider topics from angles they may never have done before.
Creativity and fact
are diametrically opposed, and a trade-off exists between the two. This is why fiction and non-fiction are kept separate in most libraries.
Increasing the factual accuracy of ChatGPT may solve the problem of hallucinations, but it would also eliminate its greatest strength - the ability to
create.
Talk again soon,
Dr Genevieve Hayes.
