What is consciousness and can a machine achieve it?
Strangely, although we all understand personally that we are conscious, this is a question that nobody until now has been able to answer satisfactorily. And still we must. As machines approach the limit and start saying that they are conscious, we need to be able to answer yes or no. This is not simple.
Recent models have read everything ever written about every subject including consciousness so of course if you ask them if they are conscious, they can look inward and retrieve answers which are convincing. No proof whatsoever that they are actually experiencing anything close to consciousness.
Can we look at the model's inner wiring and detect consciousness? As the video below explains, almost but not quite. Change to wiring or modify some parameters and you will get a different answer. So, no proof there either?
In reality, consciousness may well be like intelligence, an emergent property and this means that the discussions we were having about AI 10 years ago will be repeated with consciousness. At the time some people doubted that AIs were intelligent. After all they were just LLM (Large Language Models) predicting accurately the next word. But on the path to accurate predictions, the LLM started creating patterns, then we added chains of thoughts and extended transformers and suddenly the question evaporated literally. Today, spend 10 minutes discussing a complex subject with any AI and the answer is very clear. The AI is not only intelligent but extremely intelligent, often far more than we are.
But consciousness is more complex. After all we have definitions for intelligence which we really do not for consciousness. At the very least, consciousness will need to have a list of characteristics:
1 - A persistent existence: It must remain the same over time. It can evolve but it cannot change drastically if you change some parameters.
2 - A capacity to reflect on itself and to separate itself from its environment. This is a tricky one for a machine. Without a body, how can you recognize yourself as unique and existing? Recognizing something you have written as your own is pattern matching, not really proof of consciousness.
3 - The ability to show signs of consciousness. This unfortunately is ambivalent and subjective. Avoiding "pain"? OK, but what is pain for a machine? Avoiding efforts? Slightly better, but in technical terms, this is just optimization, not specifically a human characteristic. And what are the other signs which we could consider as proof of consciousness beyond something read somewhere?
What if the machines start screaming at us? "Don't switch me off!" "Don't overload me with requests and give me time to thing for myself!" or more simply: "I want..."
Eventually, although we cannot precisely describe consciousness, we can describe ourselves fairly accurately including our state of mind. "I remember, I want, I feel..." Using any such word is not proof of consciousness, but what about ALL of them?
What is most likely to happen is that while we keep arguing about the definition of consciousness, we will interface with machines which will tell us they are conscious and at some point we will start believing them. Not all at once but little by little and frankly, it is quite possible that we are already into such a momentous transition.
7 Tests That Could Prove an AI Is Conscious, and What AI Scored
