5 minutes estimated reading time
I was reluctant to put fingers to keyboards to type up a blog post about ChatGPT for planning. I didn’t want to be THAT person that turns out personal branding content on the latest fad as narcissistic clickbait. There is also a larger question of is it worth using ChatGPT for planning now that it has moved to a subscription model? Finally, while the next evolution of ChatGPT won’t be launched for a while, it propertied abilities seem to be evolving in certain areas the more people use it. Much of what I will cover in ChatGPT for planning also has an application with Bing’s search chat interface, or services like Notion.
Thinking about ChatGPT for planning, came after colleagues working the design team introduced me to their experimental efforts using Midjourney for image creation. Autumn rolled into winter, and ChatGPT started to become more accessible as a tool for the general public.
What is ChatGPT?
ChatGPT is a class of machine learning platforms known as a large language model. It’s given a huge amount of data and analyses it. It then uses that data to build a probability based model for what might come after a given set of terms. For instance, a user may type:
Tell me about Fenway Park, the baseball stadium in Boston
And it would be highly probable that ChatGPT would talk about how the park is home to the Boston Red Sox major league baseball team because there is so much content out online about the Boston Red Sox and Fenway Park.
In this respect, the mechanism of ChatGPT seems to resemble Bayesian inference based on Bayes theorem in output, if not, mode of action.
Named after the mathematician Thomas Bayes, the theorem describes the probability of an event, based on prior knowledge of conditions that might be related to the event. For example, if the risk of having a car accident is known to decrease with the number of years driving without an accident; Bayes’ theorem allows the risk to an individual based on their prior driving record to be assessed more accurately by conditioning it relative to their driving experience, rather than simply assuming that the individual is typical of the wider population.
Bayesian inference is a type of statistical inference where Bayes’ theorem is used to update the probability for a hypothesis as more evidence or information becomes available. It works better with dynamically updatable data (like a user correction).
Clear boundaries in using ChatGPT for planning
I could see some obvious risks in ChatGPT in terms of how it works and in how it presents its responses. But, the more that I have looked into ChatGPT, the more that I saw how it could be useful. But that is contingent on having well-defined immutable guard rails are employed in the use of ChatGPT.
A quick story
This isn’t about using ChatGPT for planning, but using ChatGPT to help a friend out in January this year as they worked on their master’s degree. They were studying law and wanted to write an essay on a particular arcane area of law, doing a comparison between how it is implemented in two countries.
We didn’t ask ChatGPT to write the essay, but used it to recommend academic authors who would have written papers on the areas of investigation, with a view to reading their works and incorporating their thinking as citations.
We got names. Some of them wrote about law, but not the specific area that we asked about. Others didn’t seem to exist at all when we looked them up via academic database tools and Google. ChatGPT’s process had somehow conjured them up.
Other people have been less careful than we were:
- ChatGPT invented a sexual harassment scandal and named a real law prof(fessor) as the accused | Washington Post (April 5, 2023)
- Two US lawyers fined for submitting fake court citations from ChatGPT | The Guardian (June 23, 2023)
I would not be surprised if these examples that have been called out are just the tip of the iceberg and others have got away with similar practices largely undetected. Also knowledge workers may be reticent to admit whether, or how much they rely on machine learning based tools. Think about that for a moment…
Watchouts of using ChatGPT for planning
ChatGPT can give you an example in terms of writing style. ChatGPT has been used successfully as a church sermon writing tool as an example. But everything needs to be separately fact checked – trust but verify.
Secondly, ChatGPT can be used to ideate around a theme, in a similar way to using a thesaurus. This could be things like language for messaging, inspiration for search terms or even terms to use in the creation of stimuli for mood boards. Again, I would look to check all of this against a thesaurus as well.