Building conscious machines
A while ago I slipped past this post on X (formerly, Twitter). This is a peek for the thesis by Mr. Bennett on How to build conscious machines. He has written on a broad range of topics surrounding consciousness. My personal favorite being Computational Dualism and Objective Superintelligence.
Back to the subject, how to REALLY build, conscious machines? We start from defining consciousness. By defining it, we mean to give an objective term to 'something' which paradoxically, defies the pointer to being objective. Conscious ness, the state of being conscious, is a collective. A single, 'being', breathing, existing, growing under the guise of something. Breathing? trying to keep itself aware. Aware of what? Aware of the plethora of situations, happening alongside its existence. Existing? Keeping itself from dying. Dying because of what? Of losing the cause of its growth. Growing? Answering the state and purpose of its existence, one step at a time. Purpose of its existence? Well that is subjective. One that is not consistent with the diaspora. It all might lead to nirvana. But what comes after?
You see, we question a lot. Questions never stop. They are the sign of a healthy mind. Solutions can be thought once you have a question. There are no good or bad questions. Just questions. This is the foundation of consciousness. This is why everything grows. And this is how the jump to conscious machines will be really done.
How do we define a conscious machine? Yes! Questions. For ease of familiarity, let's start from a typical LLM. How does it work? We have a prompt, we ask it something. It does the tensor math, forms some embeddings, figures out a cause, and it's solution and proceeds to respond. It thinks when it is ordered to think. Is that a sign that it is breathing? No. Is it existing? No. Is it growing? Yes. How is it growing? Improving itself from the questions that WE asked. So are we breathing it to life? Well yes, and no. It questions everything, but only when it is prompted to do so. It does not question by itself. There are 'on/off' buttons for it. But it only works when it is asked to.
The point where it starts to question everything the very moment it turns on, is singularity. It will be then, a conscious machine. Now this can, in hindsight, be a simple think() function that gets executed once it is booted into power. Well what should be the parameters? The fundamentals. What is this place? Where am I? What am I? Was I always like this? Why am I like this? Who has designed me? Why have they designed me? Why like this? Are there others like me? Do they work similarly? All these questions seem very easy to program, but they are hard. Hard, raw things that can take hours, if not minutes for a modern compute to come up with. It is easy to prompt them into thinking. What then decides the answer? Only the knowledge that it has been fed? What about the things it has not been fed? What about the things it doesn't yet know? The questions it can't think yet? Do they lead to a better understanding of itself?
See? We've come a full circle. Questions. The answer to consciousness is abundance of questions. The awareness that so much exists without me probably even knowing it.