AI systems are becoming more advanced at an ever increasing pace. At some point these programs are going to be so good that we might start to wonder if they are alive. But how will we know? And what does it even mean to be alive, or conscious?
Definitions
First it might be good to distinguish between being alive and being conscious. Animals and plants are alive, but most are probably not conscious. At the moment only humans are considered to be conscious. Let’s look at some existing definitions for consciousness to see what the difference is. There are many different interpretations but the most used are:
- Wakefullness - the state of being awake, instead of asleep
- Having experiences - experiencing the world subjectively
- Awareness - becoming aware of something, learning or noticing
- Self-awareness - being aware of your own existence
An analogy
To better understand this difference between alive and conscious, it might be easier to use an analogy:
Imagine a self-driving car, it can drive autonomously without human intervention, but intervention is possible by taking over the wheel. In this analogy your body is like the car, your body can do many things autonomously, like beating your heart, breathing and digestion. Even more complex things like walking or entire routines can be done without thinking about it. The car can drive by itself, manage the throttle, turn on the lights, etc.
This is where it ends for most animals and plants, but we humans have something special, our consciousness. Consciousness is the driver inside the car. If you are awake, you look around and experience the world. You notice things around you and become aware of them. You may even think about the fact that you are sitting in the car and driving around. This driver is what makes us different from the non-conscious living things.
So how can we use this analogy to think about the definitions from earlier?
Wakefullness
When we go to sleep, it’s like parking the car and falling asleep inside. When we zone out during a boring presentation, it’s like falling asleep in the car, but the car keeps driving.
Having experiences
Inside the car you can look out at the world and experience it from your point of view. Maybe you see something you recognise or that reminds you of an old memory. Or you hate that new building that looks so modern and boring. These subjective experiences mean that you look at the world in your own unique way.
Awareness
The car might be driving autonomously and suddenly stop and make a beeping noise. You notice that the road is blocked and you have to take over the steering wheel and decide on a new route. You just became aware of the situation and the need to take action.
Self-awareness
At some point you might even look in the mirror and realise that you are looking really tired from this long car ride. You can even look outside at the other cars and be jealous that they are a newer model than yours. All these actions reflect on yourself and make you self-aware, or even self-conscious.
Tests for consciousness
Now think of a scenario where you are driving on the freeway, and looking around you see other cars with blinded windows. How do you know if there is a driver inside those cars or if they are just driving autonomously?
This is almost impossible because the self-driving mechanisms can be so advanced that they appear human-like. Even if you are able to send a text message to the other car, it might run very complicated systems that send very realistic looking messages back.
There are many existing tests that propose to test some parts of consciousness:
- Turing test - can you convince another human that you are human?
- Picture understanding - what is wrong with this image?
- Mirror test - can you realize that you are looking at yourself in the mirror?
- Theory of mind - read this story and explain what each persons perspective is
- Empathy test - read this story and explain why person B is angry at person A and what he should do
Testing GPT-4
Currently, advanced AIs like GPT-4 can already pass the turing test. You can also try this for yourself on the website human or not where you can chat with other humans or AIs and at the end of the chat need to guess which one it was.
GPT-4 has also demonstrated to possess theory of mind in this paper. And in my own tests it is also able to provide correct explanations of each persons emotions in a short story:
Human: Harry wants to become an astronaut. His friends are telling him that the statistical chances of making it through the selection are very small. How does Harry feel about this?
AI: Harry might feel discouraged or disheartened upon hearing about the slim chances of becoming an astronaut, but it’s also possible that he may feel motivated and determined to prove his friends wrong and pursue his dreams. People react differently to challenges, so it would depend on Harry’s personality, resilience, and mindset.
As you can see it can clearly describe what Harry’s reaction might be, and can even explain there might be a difference between different personalities. And with image capabilities coming to GPT-4 soon, it might be able to pass picture understanding and the mirror test as well. This is all to say that either our tests are not advanced enough, or current AI systems are more conscious than we expect.
Sliding scale or on/off?
Another question is if consciousness can be expressed as a sliding scale, where you can be more conscious than someone else, or an AI system can be somewhat conscious, but not as much as a human. Or if consciousness is something that just suddenly emerges as systems become advanced enough and there exists only consciousness and non-consciousness.
Using our analogy, you can imagine that you might be able to train your driving skills to become a better driver. This way you can take over the wheel and steer the car in the right direction when needed. This could be compared to mindfulness, where you can train the control you have over your own mind with techniques such as meditation, introspection and breathing exercises.
On the other hand you can imagine people who are fine with just being driven around. They like looking out the window and the thrill of driving at high speeds. But they don’t particularly care about the workings of the car and don’t want to drive themselves. These people are content with life and enjoy living it this way.
Does it matter for the degree of consciousness if the driver has more or less skill, or is it only important that there is a driver? These are questions we don’t have an answer to at the moment.
Conclusion
The more you look into consciousness, the more you find out that we have very little scientific information about it. For the moment consciousness is a subject left to philosophers. But in a couple years we might be in a situation where an AI system passes all our existing tests for consciousness and we will have to make a decision about how we will interpret this.
Sources
-
What is Consciousness?
https://www.lesswrong.com/posts/sLT53ns3WFNKwm5yk/what-is-consciousness-1 -
Testing Consciousness
https://www.worldscientific.com/doi/epdf/10.1142/9789814407168_0009 -
Theory of Mind May Have Spontaneously Emerged in Large Language Model
https://arxiv.org/pdf/2302.02083.pdf