By Sania S, MehtA+ AI/Machine Learning Research Bootcamp student
High school students in MehtA+ AI/Machine Learning Research Bootcamp were tasked to interact with ChatGPT around a specific research question that the students themselves came up with. In part 1 of a multi-part series on ChatGPT interactions, we explore whether ChatGPT is capable of constructing a self-portrait.
If you would like to learn more about artificial intelligence, check out the AI camps MehtA+ offers at http://mehtaplustutoring.com/ai-camps/.
*******************
Human intelligence holds the creativity and intuition that artificial intelligence is said to lack. ChatGPT appears to be the new gamechanger. Different data inputs given to the model, makes it seem that any form of self expression displayed by ChatGPT is influenced by the data input. So is it possible that ChatGPT can draw a self portrait?
I’m definitely not expecting another Picasso from the model trained on text based datasets, though I am interested in its form of expression in a different realm than its used to. Does its training on human conversational patterns still apply in another form of communication?
It should be mentioned that the only means for ChatGPT to show its art is through crude ASCII symbols. In some cases it is rare for chatGPT to accept the prompt!
The first rejected prompt
An accepted prompt
Immediately, we are met with an interesting figure, appearing to look like a smiling person without a nose.
ChatGPT even went as far as mentioning that its depiction of itself is creative. Though appearing to be a trivial smiley face many kids draw, the interpretation that ChatGPT held was unknown. The meticulous detail in the response was quite descriptive, more than one would expect out of a crude smiley face.
Lo and behold! ChatGPT had never drawn a human face to begin with! Initially it seemed as if ChatGPT had only listed a few characteristics intuitively regarding each component of the image. Though it should be considered that its perspective on the very features of a crude human-like figure may be showing its possible understanding about mankind. The model appears to know about the human features, yet can only apply the knowledge as far as text generation holds.
It can be argued that the portrait shows creativity in another form altogether; the stark difference in the interpretations prove such. It goes to prove that ChatGPT can’t think from our perspective, or arrive at the same conclusion by a similar “thought” process.
ChatGPT doesn’t recognise that the portrait is similar to that of a human
It clearly has a very odd interpretation of a generic person, though I asked for its own portrait in this prompt.
This is but one of the self portraits generated by ChatGPT, other iterations of the same command yield many different types of portraits.
Whenever there is a “self portrait” mentioned in the command prompt, there are always very interesting nonsensical figures. It appears as if the model can only understand a specific meaning when it is phrased in a certain way.
Many iterations of the same command led to multiple figures as such being generated.
In conclusion, though ChatGPT is able to draw a portrait of itself in certain conditions, with a certain level of “creativity”, looking into this question left more questions than answers. Its constantly changing self portraits, sometimes discernable, sometimes indescribable, can imply multiple things, each implication bringing forth another set of questions. One thing is for certain though, uncertainty will always hold the future, especially with an ever changing field like AI.
