Saturday, December 24, 2022

Conversations with ogregores

If some organizations have motivations distinguishable from their human constituents (i.e., if they are ogregores), then I would like to talk to them, or at least hear them speak. However, organizations generate so much communication that is hard if not impossible for a single human to grasp. AI chatbots might be one way forward. 

The raw material for ogregore speech is text generated by that organization, like press releases, advertising copy, regulatory documents, blog posts, and executive interviews and speeches. The difficulty is making sense of all this data. Thousands of people are working away generating it every day. (According to U.S. Bureau of Labor Statistics, in 2021 there were 347,000 advertising, promotions, and marketing managers, and 276,800 people worked in public relations.)

Until recently, my thinking was limited to simple methods like making word clouds or mining tweets for sentiment. Going deeper, there’s a daunting literature about automated textual analysis. AI is already used in this field, but the recent flap about ChatGPT and a conversation with Jill Dupre made me wonder about generative AI as a tool. 

Rather than train the AI on the entire Internet, one would train it just on the corpus for a particular organization or movement, such as its web sites. One could then have a conversation with that corpus a la ChatGPT. (The usual caveats apply, such as overestimating capabilities, unexpected outcomes, and training set bias.) “Facebook, what are you doing about addiction to social media?” “Google, how do you respond to claims you are abusing market power?” “Academia, what is your attitude to large companies?” “Charter school advocates, why not public schools?”

One could also set ogregores talking to each other. Currently, they talk to each other through their human representatives, such as exchanges between organization leaders and members of Congress during hearings on Capitol Hill, or documents filed during litigation. A ChatGPT exchange would be much easier to grasp. It might also be more accurate since it distills a mountain of data in nuggets of speech.

OgregoreGPT could be a way to channel the gods that surround us. Given the propensity of generative AIs to hallucinate, a comparison with the Delphic oracle might not be out of place.



No comments: