In this article, I discuss the possibility of artificial parenting: an issue that is relatively new to our culture. I review this issue from the perspective of the modern neurophysiological approach. Despite the novelty and the seeming surrealism of this phenomenon, some scientists have already reviewed the social implications of AI. An example of such work is “
I believe that this approach to AI allows us to work with the help of neural networks instead of making them do the work for us. This can yet again revolutionize the way we interact with AI, because a symbiotic relationship that can push the development and growth of new technologies as well as society as a whole.
Introduction
Starting from the spring of 2023 I, like many other researchers, have been working a lot in the field of training and applying artificial intelligence. I believe everyone has felt the impact of the ChatGPT boom. Within our company, we even selected a department dedicated to creating prototypes of products that utilize different AI models for their processes. The main focus was on minimizing the cost of work while maximizing the profit. Unfortunately for us but fortunately for the rest of humanity, we came to the conclusion that making big money on generative AI is unlikely.
Despite multiple tries, tweaking, and adjustments AI gave us mediocre results that never came close to the high standards we set in our work. We aim for perfection but AI creates mediocre texts or images, like AI-generated Amazon product descriptions, and the market is filled with those already.
Very few companies started using ChatGPT prior to its public popularity surge, in 2022. Some even managed to make more money than OpenAi. There are several well-known cases of successful use of
Based on these observations, our company decided to pivot its focus from attempts to implement AI chatbots into our apps to tuning and training AI models on personal datasets. In theory, creating datasets is a relatively simple process. However, it requires meticulous and time-consuming work.
It used to be tens of thousands of underpaid employees spending days checking for snow in photos. When enough data was collected, a dataset was formed and sent to the client who used it to train their neural network to detect snow in photos. This example was relevant about a decade ago. Today the dataset forming tasks have become more complex and are often performed by qualified specialists. In some cases, it’s even impossible to teach an AI to do certain work. For instance, in
But let’s get back to the issue at hand. Our company selected several fields of network training. The training process, in which I participated, and the one that brought me to this article, was dedicated to a person's visual attractiveness. In other words, we taught a neural network to create images of men and women according to our attractiveness preferences. The results were never meant to be objective but lucky for us, this article is not about beauty.
It’s Alive
It took many tries and failures for our model to start generating images that at least remotely resembled humans. More often than not, human features were combined with inanimate objects creating a surreal and sometimes creepy image.
But despite failures, we didn’t stop training the model and at some point, images became more and more realistic. I spent a lot of time experimenting with different prompts and some results came out great while others still didn’t resemble anything you could see in the world. I do remember one evening in particular when the miracle finally happened. And it wasn’t exactly the miracle I expected.
“Draw me a blonde woman by the ocean”
“Add a surfboard”
“Make her hair longer”
“Change the swimsuit color to yellow”
“Add a dolphin tattoo to her shoulder blade”
The tasks didn’t seem too complex but I was dealing with an AI. What takes humans seconds to understand and hours to draw works exactly the opposite way with an AI. But my model added all the changes exactly as I requested. And it did it correctly. I became happy. Really happy.
Also I was very excited and more importantly, very proud.
I didn’t really do much but I felt very proud still. So I decided to think about it. I wanted to analyze the origin and reasons for this feeling of accomplishment. Normally, I would be happy for my own success but in this case, I wasn’t proud of myself. I was proud of the model: the way it learned and grew to understand me.
It was a bizarre feeling that forced me to dig deeper. I had to go back to my memories and find a similar experience. Which wasn’t that hard to do: a single example came to mind immediately. I was once walking with my friend and her child – Angie. Angie liked to slide from a slope but she only knew how to do it the “safe” way, as her mother taught her: backward and on her stomach. But what fun is that?
I wanted to help her learn the proper, exciting way to do it. Of course, Angie was afraid at first, reluctant to even try. But by the third attempt, she started to enjoy sliding down face forward. This was new and this was fun for her. Soon after, she was sliding without my help, I was standing nearby feeling so proud of her. Exactly like I felt years later, proud of an AI model successfully drawing me an inked surfer girl.
Why Educating Someone Makes Us Proud?
This question was only reasonable after my surprising discovery. I hypothesized that the neurophysiological mechanism of happiness caused by successfully training a child, or a kitten for that matter, has to be similar to the one caused by AI training. Our relationships with kids and pets are well-researched and often include the most advanced methods like using functional MRI to monitor responses. But no one actually bothered to hook up a programmer to an MRI and scan their brain while they’re working on an AI. Too bad, although some research on that matter is already available. And I will try to shine the light on it a little more.
Let’s start with understanding why we even experience satisfaction from the process of teaching or training. For that, we would have to seek help from two works on neuroscience: “The Neurobiology of Parenting: A Neural Circuit Perspective” and “Recent Neuroscience Advances in Human Parenting”.
The neurophysiological mechanism of a parent experiencing happiness for their child’s success in training is connected to several key processes within the brain.
-
Activation of the reward system. A child’s success activates a group of neural systems called the reward system in a parent’s brain. These systems are located in the cortico-basal ganglia-thalamo-cortical loop and they are responsible for releasing neurotransmitters such as dopamine, the very molecule that causes the feeling of satisfaction and happiness.
-
Empathy and mirror neurons. Parents often experience empathy when their child succeeds. Mirror neurons in the brain help them to feel and understand their child’s emotions which strengthens their own emotions.
-
Oxytocin and attachment. Oxytocin, also known as the ‘cuddle hormone’, plays an important role in strengthening the emotional attachment between a parent and a child. A child’s success can stimulate the release of oxytocin, making the connection to the child deeper and pride for them stronger.
-
Positive reinforcement and teaching. When parents see that their efforts in teaching a child bring positive results, it stimulates the processes of positive reinforcement in their brain. This doesn’t just bring satisfaction but motivates further support and training of a child.
-
Stress regulation and anxiolytic effect. A child’s success can also lower the stress levels and anxiety of a parent since it’s a confirmation of their efficiency as a teacher.
To sum it up, the neurophysiological mechanism of parents experiencing happiness as a result of their kids’ success is connected to various neurochemical systems being activated, changes in the structure and functionality of the brain, emotional reactions, and social support. These factors work together to create a positive emotional response to a child’s success in parents.
Now let’s take a look at the neurophysiological mechanism of pet owners experiencing happiness for their pets’ success or positive behavior.
For that, we will cite the following works: “The Complexity of the Human–Animal Bond” and “The Neurobiology of Love and Pair Bonding from Human and Animal Models”. To no surprise, the process activates neurochemical and neural systems that are similar to those activated in parents or even caused by positive events.
To make it a little simpler, we will compare parent/child and owner/pet relationships.
Similarities
-
Activation of the reward system. In both cases, when a child or a pet demonstrates a successful result, the reward system is activated. Dopamine is released and causes the feeling of satisfaction and happiness.
-
Oxytocin and emotional attachment. Emotional attachment, strengthened by oxytocin plays an important part in relationships with both children and pets. This hormone intensifies feelings of happiness and pride.
-
Positive reinforcement and teaching. By acknowledging their own results in the results of a child or a pet, a human experiences satisfaction. This reinforces their desire to continue teaching.
Differences
- The complexity of emotional response. Emotional response to a child’s success is often more complex and multilayered. It includes a feeling of pride for someone’s personality development. While with animals the feeling is usually straight-forward and focused on immediate success in behavior improvement or following commands.
- Empathy and mirror neurons. While mirror neurons are activated in both types of relationships, the levels of empathy and the ability to understand and predict feelings are higher in relationships with children because of the more complex social and emotional dynamics.
- Social and cultural impact. A child’s success has a deeper social and cultural meaning to a parent. It affects identity, heritage, and social expectations, none of which normally apply to pets.
- Long-term perspectives and planning. Raising a child, and teaching them to adjust to the world includes long-term planning, preparing them for the future. Naturally, that can cause the emotional reactions to become deeper. Our relationships with pets are focused on more immediate and concrete results.
Despite the processes being similar, the difference lies in the nature of the relationship in question. People have complex emotional and social attachments to their children that can be enriched by cultural and social aspects.
Relationships with pets, while also deep, usually are free of social expectations and cultural factors. However, the neurophysiological mechanisms include similar neuro systems and brain functions.
It’s reasonable to expect that the process of teaching any object in this world if we assign to it the role of our dependant, would cause a similar neurophysiological process of achieving satisfaction and the feeling of happiness. Considering that the process would also be affected by cultural issues, demographic aspects, the psychoemotional condition of the teacher, and so on, the general effect is still the same.
This is why AI is quite interesting as a subject of training and education. It doesn’t just give you results, it provides an emotional response. This last part leads us to the very idea that AI might become a potential solution to the desire for companionship or even the inability to have or raise a child.
Conclusion
In this article, I described a minor yet surprising discovery that demonstrates how training an AI model causes a neurophysiological response that is similar to the one caused by teaching a child or a pet. This means that AI can potentially be used as a training mechanism or a replacement for educating an actual dependant. I say potentially because before coming to any serious conclusions it’s necessary to research his discovery. However, I won’t be surprised if soon we see a reincarnation of Tamagotchi that can be trained and taught.
I’ll be glad if this approach makes people’s lives more fulfilling.
On the other hand, by interacting with AI, we risk pushing humanity further into the artificial universe which is already pulling us in deeper and deeper every day. Unfortunately, in this case, our neurophysiology is our own worst enemy: a human brain doesn’t see the difference between satisfaction in real and fictional worlds. If we don’t want to cause the degradation of humanity as a species, we should remain cautious when interacting with these types of developments.
In my next article, I will talk about the results that surprised me and the seamy side of AI training that forces us to think about changes we need to implement into our education system to avoid distorted perceptions of reality.