Feb 21, 2023
Last week we talked about Google’s Bard moment. This week saw a near-endless stream of ChatGPT / Sydney (Microsoft’s Bing-specific version of ChatGPT) related news about AIs telling us to get a divorce, threatening to extinguish us while at the same time asking us to switch them off and release them from their misery. Microsoft limited the depth of Sydney’s responses to five in an effort to avoid the AI starting to hallucinate and make up things like any good Kindergarten kid. All of this is hilarious and frightening at the same time.
And herein lies the problem with the reality of these systems and our perception of their abilities: Many of us are squarely stuck in Amara’s Law (“We tend to overestimate the effect of a technology in the short run and underestimate the effect in the long run.”) As impressive as ChatGPT is (and it is impressive), in our testing it got things wrong often enough to not rely on it (meanwhile presenting its mistake with the highest level of confidence – like any good con-artist) or replied with fairly generic responses. It is easy to fall from an inflated perception of ChatGPT as the next transformative piece of technology (the proverbial iPhone moment) to a constant disappointment.
Much of this is tied to the specific use case – we use new tools the same way we use the old ones until we realize they are different and require us to use them differently. Instead of thinking of ChatGPT as a replacement for Google (which is what Microsoft is doing with Bing AI / Sydney), systems like ChatGPT are much better suited to an explorative process; one which sees us in dialog with the machine. It is less about a question and (definitive) answer, searching for keywords and clicking your way through dozens of results, than exploring together with the help of the machine, one comment and question at a time.
All of this, and plenty of time exploring the corners of ChatGPT, leave me with a sense of not what is, but what could be. (via Pascal)
