Shallow insight
Last Monday, billionaire surgeon, biotech director and LA Lakers shareholder Patrick Soon-Shiong announced that his paper, the LA Times, would incorporate a new AI tool on its website. Called ‘Insights’, the tool analyses content labelled as ‘Voices’ – including opinion, commentary, editorials and more – ‘to identify where the views expressed may fall on the political spectrum’. It also provides an ‘annotated summary’ of the article including ‘different views on the topic from a variety of sources’.
It quickly caused controversy. A Bluesky post by New York Times journalist Ryan Mac drew attention to annotations the tool made to a column by Anaheim journalist Gustavo Arellano. The column made the case that Anaheim should not forget the role of the Ku Klux Klan in the city’s history. Mac said the tool appeared to ‘well, actually’ the KKK in appending the following comment: ‘Local historical accounts occasionally frame the 1920s Klan as a product of “white Protestant culture” responding to societal changes rather than an explicitly hate-driven movement, minimizing its ideological threat’. Journalistic uproar ensued as the annotation was interpreted as sympathising with the KKK. Some cancelled their subscriptions to the paper. The tool was quickly removed from the column.
On Friday Arellano weighed in, arguing that the tool had not downplayed the KKK at all. Rather, it noted how some local historical accounts downplayed the KKK’s history in Anaheim, which was actually the point of his column. Arellano was miffed – suggesting carelessness on the part of those with their knickers in a knot. He also observed that it’s up to the reader to use the tool or not – you have to press a button at the top of the page for it to work its magic: ‘Like the comments section, you can engage with it or not. You can choose just to read what the humans have to say — and criticize or laud them. Why, if you ignore the AI pendejada enough, it could very well pick up its digital football and go home.’
There’s a lot to be said for Arellano’s nonchalance, assuming Insight’s insights are accurate, if shallow. But as Arellano observed himself, the other two notes the tool left on his column were ‘wildly out of context’. And the categorisation of articles on a political spectrum seems to open a different can of worms. Here we’re not talking about using technology to add contextual information but to superimpose an ideological judgement that can only ever over-simplify journalists’ work. Classifying a person’s political beliefs on a two-dimensional spectrum is a fool’s game, to say nothing of trying to tag individual articles. When I activated the tool on another column, it unsurprisingly revealed that the article ‘generally aligns with a Center Left point of view’. What are readers to do with this information? Those on the centre left might feel themselves in fellow company, while others will likely not be there in the first place. If they are, the tool is more likely to chase them away than tell them anything useful. Transparency – if you can use that word for something delivered by AI without human input – can sometimes increase trust, but sometimes not, and mostly amongst those who already trust you, perhaps because you’re on their side. For the rest, no such luck.
Michael Davis, CMT Research Fellow
