Last week, I received the email. "You're invited to create with DALL·E," OpenAI declared in big letters.
An excited, "eeep!" involuntarily escaped my body, and I got to work immediately, rapid-firing various concepts that have been floating around my mind, just waiting to erupt in this moment. I was now an artiste with a vision and I was sent into a flow state, experimenting with submitting descriptions to DALL·E. I tried varying levels of vagueness and abstractness, which had different effects (only a few examples are posted here, but I will post more art from DALL·E in the future).
Sometimes, the art that was generated was exactly what I was going for:
Other times, it was slightly off, but still had cool results that were close to what I was looking for:
...and other times were complete laugh-out-loud-smack-your-forehead failures.
As I was submitting descriptions to DALL·E, I realized that going through this process could be a beneficial exercise in communication.
The process of creating AI generated art goes something like this: I provide a description to DALL·E, a few seconds pass as it is processing and generating, and then DALL·E responds with four images. In doing so, DALL·E is basically saying, "So, you meant this?" Sometimes the images are exactly what I meant to convey, and sometimes it is not at all what I intended to communicate. When the images are not what I wanted, I then have the opportunity to clarify and edit my description to attempt to arrive more closely to my intended artistic goal. The stakes are fairly low here.
The same thing happens when communicating with my dog, Maple. I may prompt him to sit by saying, "sit," and he may respond by laying down. That was not my intended goal, but I have a quick response from him that allows me to see that there was a miscommunication. At that point, I can go back and edit my prompt to him, by saying "sit" and adding the hand signal that he associates with sitting down. Then he will sit (and get a treat, of course).
Communicating with humans can be more difficult or inconsistent when the quick feedback loop is not the norm (I recognize that this is more of a norm among certain communities, but it is not the broader standard norm). Whether it is in a discussion, a lecture, in writing, or on social media, there is room for misinterpretation in communication, and for us not to see that our words may have been misconstrued. In social media, there may be feedback in the form of comments, but often those are aimed at responding rather than clarifying.
Imagine DALL·E functioned the same way that human communication often functions. What if I submitted descriptions to DALL·E but could not see the images that came from the descriptions? Or what if the images arrived as physical copies in the mail, weeks after I submitted the descriptions? What if I were making t-shirts to give away for an important cause, but could not see the art that DALL·E created until all the t-shirts were made?
The stakes seem higher, and the likelihood of the printed images or t-shirts coming out the way I intended them to look decreases, as there is not room to clarify, or add more context, or elaborate on what was meant to be conveyed. Yet, this is often how human communication works.
In typical discussions, people converse back and forth, each trying to communicate something, while listening to the other person speaking. I believe that many people in these conversations have good intentions, and likely believe that they are actively listening, understanding what the other person is saying, and being understood by the other person. There may be cues such as head nods that indicate that a person understands what the other person is trying to convey, but after parting ways, each of them may be left with a construct of what the other person meant, which may be correct or it may be the sort of smack-your-forehead failure that occurs with DALL·E at times.
This is something that I have been improving on recently. I am fairly comfortable asking questions and listening to what people say, but I am less practiced at echoing back what people say to me. This involves a more analytical, deeper sort of listening, which has a goal of not just listening to the words that people say, but making sure to verify that the interpretation of those words are correct. When I am actively making this mental move, I say phrases such as, "What I heard you say is..." then continue on recounting and summarizing what the other person said. If there is anything that is not quite right or parts that are fuzzy, the other person has an opportunity to clarify or elaborate until my account of what was said and its meaning aligns with what the other person was intending to convey.
This takes work. This takes time, and energy. However, it is an investment, as putting in the emotional labor now can be costly, but not as costly as actions that may be taken from misunderstood messages.
DALL·E provides a quick feedback loop that allows us to see whether our intended message or goal was understood. In human conversations, we can practice echoing back or recounting what a conversation partner has said and check in to make sure our interpretation matches the intended message in order to have that same effect.
I'm excited to improve my communication skills with DALL·E, and I'll post more results here on MindCambio along the way. May we all improve our communication with DALL·E to generate epic art, and improve our communication with each other to collaborate better as a whole.