The holy grail of AI research is to create Artificial General Intelligence (AGI). AGI is what we will get once we get a machine to operate on the same level of intelligence as a human.
There have been countless theories and approaches over the years trying to achieve this goal, but, so far, none of them have succeeded.
The first wave of theories around the 50s and the 60s, was focused on using logical rules in order to represent knowledge and make machines reason like humans. This was based on the assumption that the property that makes humans special is the ability to reason in a logical manner. For example, humans can easily understand logical statements of the form
“If all men are mortal and Socrates is a man, then Socrates is a mortal”.
This approach is now termed “Good Old-Fashioned AI“. It was invented by researchers like John McCarthy and Marvin Minsky. While this approach helped birth AI as a scientific field, it, unfortunately, ended up being very limited.
Research then moved on to other approaches like machine learning and neural networks. The 90s saw the rise of what was called “computational intelligence”. This was an approach inspired by biology. Computational intelligence had three main “arms”: neural networks, genetic algorithms (or evolutionary computation) and fuzzy logic.
The idea behind computational intelligence was very attractive. Since intelligence exists in nature, we should be able to replicate intelligence if we mimic natural processes. For example, ants display a kind of intelligence in the way they are organised. By mimicking the way they organise we can create algorithms which display the same kind of intelligence.
A very common analogy was that of the birds’ flight. While feathers is a way for an organism to develop the ability to fly, it is not a necessary prerequisite, as there can be other ways to generate flying organisms or machines. However, by creating machines that imitate birds, we can identify the core features that let us create machines that can fly.
Unfortunately, computational intelligence didn’t manage to give us true general AI. It lacked a proper mathematical theoretical framework which could guide research and provide satisfactory explanations as to how these methods worked.
These days, the latest iteration in the effort to create general AI is deep learning. But, even this might not be enough. While deep learning has been very successful in multiple tasks like computer vision and natural language processing, it has been criticised as simply a very powerful curve fitting machine, but nothing. Judea Pearl claims that it is simply a more powerful version the the same thing that algorithms have been doing for the last 150 years. According to his opinion, the only way to generate real intelligence, is to create an algorithm which can understand causality.
In any case, it looks that the debate is far from over. So, let’s go back to the title of this post. What is consciousness-based AI and what does this have to do with general AI?
Computational intelligence tried to imitate nature in order to generate AI and it failed. Other approaches have tried to use mathematical wizardry, in order to do the same.
