top of page
Search
  • Writer's pictureJay Krall

Why it's worth communing with the data

Updated: Feb 13


Mushroom hunting in Arkansas

In meetups and webinars, I hear data analysts say one thing consistently about the promise of generative AI: it can save them time. "Summarization helps us quite a bit. We're trying to understand millions of conversations at a time," said an intelligence director for a movie studio.

Typically, AI skeptics will respond to such assertions of hope with talk of hallucinations, those information fabrications to which large language models (LLM) are so prone. But actually there is a more permissive and less dramatic danger to letting bots do the reading for you: the inherent bias of their training data will inevitably skew your thinking.


This was true as well with the older and less controversial form of AI, "discriminative". Unlike generative AI which produces long textual outputs in expository tones, discriminative AI simply infers meaning from data and performs probabilistic labeling tasks. Interpretive tasks like named entity recognition, which aim to disambiguate references to brands like Apple, Tide, Target and Nest from common-word usages of those same words, are widely used in many data analysis tasks across a range of industries. Such tools demonstrate the issue of bias which is innate to both discriminative and generative forms of AI.


Recently, I studied major media coverage of the psychedelic therapy movement. As jurisdictions like Oregon and Jamaica legalize the use of psilocybin and other psychedelic compounds under medical supervision for the treatment of mental health issues, public opinion has started to shift on these substances, which remain illegal drugs at the federal level.


The expressed opinions of celebrities and public personalities are typically critical to growing acceptance around any taboo subject. So I tried to uncover, who are the public voices moving this discussion forward? Most discriminative AI tools which perform such tasks across public news articles are trained on Wikipedia. Of course, Wikipedia's register of public people is overwhelmingly white and male. Amongst the top 10 public personalities discussed in the news alongside psychedelic therapy, all but Will Smith were white men:



The men in this table have a range of relationships to the psychedelic therapy movements. Some are medical professionals and passionate advocates for alternative mental health treatments, while others described their experiences with such treatments in media interviews. But are these the people who are influencing tastemakers and policymakers?


To counterbalance the pale-and-stale Wikipedia analysis, I began listening to podcasts on the topic. This enabled me to identify women and people of color who have shared deeply personal experiences with psychedelic therapy for different types of trauma. For companies in this service space, these people represent much more relevant and salient ambassadors:


Inevitably, in the rush to appease research clients and "drive efficiency", analysts will always use tools to help them minimize manual work. But too much reliance on automation can heavily skew your view of a cultural moment and how it's evolving. The only remedy is to make some time to commune with the data.


3 views

Comments


bottom of page