Artificial Creatures
Homeostasis
If an elevator was conscious of its own needs, how would it work? In this section, I'll be applying the homeostasis model to an elevator. The homeostasis model is based on homeostatic emotions, which are feelings like hunger or tiredness that drive behavior like eating or sleeping. There's a certain tendency to achieve and maintain an internal balance in order to feel good as people. So what if we applied this model to an elevator?
First we need to state the elevator's primary function and ideal internal balance. I would say that the primary function of an elevator is to transport people from point A to point B. This means that the ideal internal balance for an elevator is one where it can perform this task most easily. In order to perform this task, the elevator needs to be able to do a few things.
- The elevator needs to be able to hold people
- The elevator needs to be able to be called to the point of departure
- The elevator needs to be able to accurately travel to a destination
After this we can easily consider which actions influence this internal state.
- The need to be able to hold people is influenced by how many people are inside the elevator. If the elevator is empty, it can hold a lot of people. If the elevator is full, it can't hold any more people than it's currently holding.
- The need to be able to be called to the point of departure is influenced by the buttons on every floor which call the elevator to that floor. If ten different people on different floors all call the elevator to their floor, the elevator will only be able to travel to one floor at a time, thus influencing the ability to be able to be called.
- The need to be able to accurately travel to a destination is influenced by the people inside the elevator pushing the buttons. If one person in an elevator were to push every button on the elevator without actually wanting to go there, it would impede the elevator's ability to accurately travel to the destination of other passengers.
Some example scenarios:
- If there are no people inside and nobody called the elevator, it would do nothing. It has fulfilled its primary purpose and doesn't show any behavior.
- If the elevator is full of people with different destinations, it would want to get rid of those people as quickly and efficiently as possible in order to return to that internal balance. So, it goes to every floor and drops the people off until that balance has been restored.
- If there are no people inside but there are ten different people on different floors who need to get inside the elvator, the elevator will move to each floor to pick up and drop off the passengers. If possible, it will drop off some passengers on the same floor where other people would like to get in the elevator.
Research
Snake AI by CodeBullet https://www.youtube.com/watch?v=3bhP7zulFfY
One of my favorite channels pertaining to artificial creatures and AI in general is CodeBullet. In this video here he's made the snake from Snake an artificial creature. The concept is simple and can also be explained using the homeostasis model. The creature's primary need is to get a high score, and it does this by getting food and avoiding death by touching the wall or itself. This need to get a high score drives the behavior, but what's interesting to me is the way that the creature learns behavior. CodeBullet uses a kind of generational machine learning system where the creature pretty much learns by trial and error, learning from mistakes and improving strategies as the generations go along. In the first few generations the snake doesn't really know how to avoid walls yet but it figures this out pretty quickly. More difficult tasks like looking around for food while avoiding touching itself come later on. At the end of this learning process the snake is able to get a score of over 100.
If you're interesting in this type of generational learning, I definitely recommend CodeBullet. The whole channel's content is similar to John Conway's Game of Life, except he uses machine learning for a bunch of different things like playing chess, solving Rubik's Cubes or playing Flappy Bird.
Reflection
I think artificial creatures is a very interesting topic. It doesn't seem immediately apparent, but the value an artificial creature can have can be incredible. This value can present itself in many different forms, from a robot that vacuums the house for you to an artificial companion to help combat loneliness with elderly people in a retirement home. This topic is honestly way broader than I initially realized. An artificial creature can be as simple as a couple of pixels on a screen and as complicated as the robots made by Boston Dynamics. Even things like chess computers can be considered artificial creatures!
When talking about the future of artificial creatures, a lot of people will get to a point about how AI is heading to a point where it's going to take over the world in some kind of violent dystopian robot revolution. However, I don't think the point at which AI becomes truly sentient is anywhere in the near future. Computers are definitely getting smarter and smarter, but most of that work is thanks to human innovation. I think that if at one point an artificial intelligence achieves true consciousness, it won't be because it gained that on its own, but because people found out how to program that in somehow.