AI and LGB Erasure

AI has been making waves recently, and has been met with everything from bemused interest at some of its more bizarre hallucinations and inaccuracies, to dire prognostications for the future of work.

We use Hootsuite for publishing most of our social media posts, and Hootsuite recently announced a new AI-powered assistant. In theory this allows users to take old posts and reword them quickly, or create automated summaries of blog posts and news articles. So we thought we would give it a quick try. 

The results were… revealing.

Taking a recent post regarding IPSO’s consultation on sex and gender identity, the new AI tool produced semi-reasonable summaries of the content, but in the process “corrected” the name of our organisation to “LGBT Alliance”.

Another test run, on our intervention with the European Court of Justice in favour of same-sex marriage – a specifically LGB issue – the group that the AI decided should be celebrated with a flag emoji is transgender people:

This is not really Hootsuite’s fault, but a result of the way that current Large Language Models are being trained on enormous amounts of data filled with cultural bias. 

The AI is merely reflecting the statistical biases of what is celebrated in wider culture – and overwhelmingly, LGB interests have become swallowed up by transgender inclusion and celebration at all costs. The idea that widespread institutional biases like this will average out given large enough amounts of data is clearly naive.

There are huge ethical concerns around the training of AI on data that contains hidden assumptions that negatively affect minorities like lesbians, gay men and bisexuals, especially given the use AI is increasingly being put to: automatically moderating, generating and censoring content.

At this rate, the assumption that LGB & T must always go together, or that we must be represented by a continuously expanding initialism is only going to become more deeply embedded and harder to speak out against.

We all need to be drawing attention to issues like this now.