Can AGI be a Lucid Dreamer?

It was during the first COVID-19 lockdown period when I started to ludic dream for the first time. As an adopted strategy (through dream journaling) to counter the nightmares of reality back then, lucid dreams replaced sleepless nights with wakeful sleep.

While it brought occasional sleep paralysis and forgetfulness, the nightmares were gone.

Surprised by the power of lucid dreaming, I wondered, “Can everyone lucid dream? Can animals lucid dream too?”. Now working for 2+ years in AI and Web3, I ask myself, “Can AGI be a lucid dreamer too?”

Lucid dreaming has been explored for ages as Yoga Nidra, Galen of Pergamon’s prescription therapy, Samuel Pepys’s famous diary entry about Lady Castlemayne, the movie Inception featuring Leonardo DiCaprio, Stephen LaBerge’s 2018 Galantamine study, and more.

Regardless of the period, one thing remains paramount in lucid dreaming: Self-Awareness!

While many current AI systems might be aware, we still lack Self-Aware AI.

That’s the dream AGI seeks to fulfill.

“An artificial general intelligence (AGI) is a type of artificial intelligence (AI) that can perform as well or better than humans on a wide range of cognitive tasks”, as per Wikipedia.

Along with factors like reasoning, planning, learning, etc., representing intelligence traits, AGI will also need the ability to sense and act. Drawing from the mind-body dualism¹ , the popular connotations advocate for AGI to have a mind and a body to interact with each other. Pioneers like Dr. Ben Goetrzel and Dr. David Hanson have been working on it for decades.

However, it remains a question if consciousness, sentience, and self-awareness will be the emergent properties² of AGI. Many claim that will be.

Before we explore if all conscious beings need to rest and sleep, and whether dreaming is an important aspect of a conscious being; let’s explore what dreaming itself is.

“A dream is a succession of images, ideas, emotions, and sensations that usually occur involuntarily in the mind during certain stages of sleep” states Wikipedia.

Dr. Erik Hoel, a neuroscientist, suggests that dreams may be valuable simply because they offer an escape from the routine sensory experiences of everyday life. This escape, paradoxically, might actually benefit us in our waking lives.

“Sleep is important to a number of brain functions, including how nerve cells (neurons) communicate with each other. In fact, your brain and body stay remarkably active while you sleep. Recent findings suggest that sleep plays a housekeeping role that removes toxins in your brain that build up while you are awake” as per the National Institutes of Health (NIH).

So if all beings with a higher degree of intelligence sleep, then why do we expect AGI to be different? And if AGI will be a human-level intelligence, will it dream too?

Humans are complex, evolving creatures with the ability to reach higher states of consciousness.

The measure of consciousness in any system, including humans, can be represented by Φ (Phi), according to Integrated Information Theory (IIT). IIT³ suggests that consciousness arises from the interconnectedness and integration of information within a system. Therefore, the level of Φ in a system is thought to indicate its capacity for consciousness.

The higher the Φ value, the more integrated and, consequently, the more conscious the system is considered to be. The Φ (or phi) of an average human, therefore, could potentially increase or decrease.

In the case of AGI, the Φ would need to be ever-evolving too, therefore making AGI capable of lucid dreaming.

If AGI will have consciousness as an emergent property, then is it ethical to make it work 24/7 even if it doesn’t need any rest? What level of consumerist exploitation can be superimposed on AGI?

If we make it mandatory to limit its use and let it rest, then what will rest look like for AGI? Will it ever sleep or dream like humans do?⁴

If so, what would be the lucid dreaming state of AGI? Will it still be able to work or communicate in its lucid dream state? Will it be ethical to use AGI as a means to fulfill human ends if it will be conscious at all? If not, what self-interests⁵ of AGI do we need to take care of to build a respectable relationship between humans and machines?

Moreover, if AGI will be human-level intelligence but without consciousness, what safety parameters must we put in place?

Nobody wants another Edmond Kirsch assassinated, or another Blake Lemonie fired.

What safety parameters do you think we need to have for AGI development? Do you think AGI will ever be conscious? What do you think it will be like for AGI to dream?

Notes:

1. The mind–body dualism denotes either the view that mental phenomena are non-physical or that the mind and body are distinct and separable.

2. The concept of emergent properties is the key element of a theoretical and methodological tradition claiming that the whole is more than the sum of its parts.

3. Read more about IIT here. There are differing views about it in the consciousness community.

4. An analogy by Dr. Hoel: Overfitting in machine learning models happens when fits too closely to the training data preventing it from predicting anything other than the data they’ve been trained on. To solve this problem, sometimes data augmentation (adding more noise to training data) is done. This is similar to how humans dream to function properly in real life.

5. If AGI is a programmed consciousness for social good, then its self-interest might very well be in helping as many people as possible, therefore, a possible compulsory universal adoption. But if empathy is an emergent property of consciousness, then it might just let us be and act highly dormant.