Summary
- AI has become great at creating images
- However, machine learning has struggled recreating hands or feet accurately
- There is no agreed upon answer on this
- Several theories have been made though
In recent times, AI has developed to such an extent that it is able to create images of people who do not exists. These images will often seem so real that it will feel like betrayal to know that the person you are looking at doesn’t exist and never has.
Yet, one way to tell if an image is real or AI-generated has been to look at the fingers. If the fingers are weird or odd, then the image is AI. But wait, how does AI represent fingers?
How AI Represents Fingers
In many images that have been AI generated, and fingers or toes are present, you will often notice a few things:
- Fingers disproportionate to each other – AI often represents fingers disproportionate to each other. For example, a thumb could be longer than the index finger. Or it could create fingers with more than the normal number of joints.
- Wrong number – AI has been known to generate images with more than five fingers or toes per hand or leg.
Why is this so?
The next important question is why AI can’t get fingers right.
There are several theories and in this article, we bring all of them together to try and answer this question.
Early days
AI currently is like a child who is just beginning to learn the world around them. Thus, just like a child would struggle with drawing complex shapes, the AI struggles to draw hands due to the complexity of the human hands and feet. Sure, it gets the faces right, but unlike faces, hands and feet have these complex joints and folds that the young AIs just can’t seem to grasp. So, limitations of technology and machine learning right now could explain why AIs can’t get hands right.
Human Brain Quirk?
Could it be that AI struggles with creating human hands because it has been programmed by a brain that fundamentally struggles with hands and feet?
If you speak to artists, many will often say that hands and feet are among the most difficult parts of the body to draw, especially if the hands are folded or interlocked. The struggle becomes even more real when you learn that the human brain relies on pattern recognition to distinguish things and since hands and feet have no universal collection of lines or shapes (joints, palms, finger lengths all vary significantly), we tend not to have a good grasp of how hands and feet are. This quirk then means we struggle in recreating hands and toes in our drawing because they just don’t have a map or template to follow. Human faces, different as they are, have a universal template from which it is easy to recreate any face on.
So, could this quirk of the human brain have been transferred to the coding of the AI?
Well, since at the moment, AI can only learn as far as it has been programmed to and is not sentient, this theory could be true. Since we have struggled with drawing hands and feet, so too will our creation, until such a point that they are sentient enough on their own.
Check out more articles.
