Abstract
This document attempts to define consciousness. It also suggests a model for consciousness and explores ways to implement it.
Defining consciousness
Here is my attempt at defining consciousness through its features:
-
It is the ability for a system to simulate its universe, including itself and other systems (conscious or not). A conscious system can navigate in four dimensions (space and time) in its simulated universe.
-
It is subjective, every conscious system generates its simulation with itself as a referencial. What is ‘itself’? Here we define individuality as a part of the simulated space that is constant over time and space.
-
It can be measured, typically not being able to assess the consequences of one’s action is deemed as being “unconscious”. On the other hand, a superior form of consciousness is a system that understands and predicts outcomes with impressive accuracy.
-
It is recursive, it can represent itself representing itself etc… This is limited to the system's computing power.
-
It is used by our brain to accurately predict outcomes and drastically increase our survival rate.
-
It cannot be split apart from objective, otherwise the simulated universe would have no meaning. Since the simulation is a simplified representation of reality, it is optimized relatively to its objective, furthermore, the objective is a part of the self, a thing that remains somewhat constant over time and space.
Feelings
The brain and to an extent the human body is fully connected, therefore most of its systems have the ability to affect each other in very (very) complex ways. Naturally, consciousness is also affected by our body’s chemical balance. A lack of serotonin can alter the consciousness’ ability to properly categorize the past experiences or predict an outcome, most of its experiences will be viewed in a negative way by the person. Love on the other hand, has the opposite effect, we can often be blind to obvious situations because the chemical balance in our brain is so disrupted by our need to procreate. What we call feelings could be our consciousness’ representation of our objective function. A way for our system to attach labels or ‘meaning’ to our past experiences. The brain also uses emotions to optimize data, strong emotions can remain longer in memory while most every day events are not stored.
An algorithmic model
Examples of reinforcement learning based probabilistic models have shown some ability to display human like behaviours, but they require astronomical amounts of data and computing power on simplified systems. Why is that? Searching the entire spectrum of the universe for a solution is bound to have a very high cost in computing power. Convolutions have shown that injecting hard coded logic to reduce the dimensionality of the problem can lead to tremendous results. Nature itself used a similar approach on our brains: our cortex, is a ‘simpler’ system that includes basic pattern matching systems (fear, hunger, motor skills etc…). Then we have the neocortex, a more complex system that includes our ability to reason.
Here are a few components of a conscious system:
- A core made of hardcoded rules. An amount of handmade rules and biases about our world will be injected to our system. They will then be fine-tuned using data from our world.
- A perception system. Raw data need to be turned into an input for the core. This actually is the hardest part.
- A high level layer. This part can be optional, this allows your system to learn and improve with experience. With this step, our system gets the ability to create models.
- An objective function. The lower and higher layers serve as input for your objective function, it looks for patterns in the raw input and the simulator system’s predictions outputs to classify them.