AI Development: A closer look at Narrow, General, and Super-Intelligent Artificial Intelligence

 

The Development of Artificial Intelligence

If you're like me, then you kind of just assumed that AI is AI is AI, right? Well, actually there are three different types of artificial intelligence that we need to be aware of. Each represents an evolution in technological advancement and intelligence. Narrow AI is what we all know and use daily. This is the only type of AI that is available in production and for consumer use. General and Super-Intelligent AI are both theoretical, at least for now. This article is going to take a closer look at each of the three types of AI, describing them in more detail and highlighting some advantages and disadvantages, or general concerns that I have with the technology.

Thanks for reading!

Narrow AI

Narrow AI, also known as weak AI, is what we all interact with today. This is the only type of artificial intelligence that is in production and available for consumer use. These systems are designed and developed to perform specific tasks. Think about the AI interactions that you have; each model that you interact with performs a specific task, or perhaps several specific tasks. For example, you may use GPT 4o (like me) to serve as an assistant with any number of information related requests. As mind blowing as this technology can be (I'm projecting my personal opinion here), the reality is that large language models (LLMs) is designed to provide information based on input requests using human language, and then deliver that information in a nice, easy to read package. It does this very well! If you asked it to bake you a lemon meringue pie, however, GPT 4o cannot do that.

I am most definitely projecting. I love pie!

Don't confuse narrow, or weak, with unintelligent and incapable. The American Academy of Ophthalmology conducted a study, positioning OpenAI's GPT 4.0 (not 4o) versus humans. Out of 1,023 questions, GPT 4.0 answered 82.4% correct, while the humans answered 75.7% correct. (1) Narrow AI systems are becoming more intelligent by the day. The key, however, is to remember that they are still trained for a specific task or series of tasks.

What are some of the advantages with narrow AI? Undoubtedly, the models that are available today help improve efficiency. They allow for the automation of repetitive or mundane tasks. They can also help to make informed decisions. Thinking back to the ophthalmology example, there are models in existence today that help ophthalmologists to recognize and diagnose diseases and sometimes even perform surgery. Narrow AI models can also serve as personal assistants. For example, I use AI to help teach me Python, to help with my garden, with project management and more!

Of course, there are also disadvantages to narrow AI. They are limited in scope and tend to lack the ability to "think outside the box" like humans do. The text to image and text to video generators are clumsy and often don't provide the output that we are hoping for (more projections). They are also dependent on the data that was used to train them. Garbage in equals garbage out. The most concerning of disadvantages, in my opinion, are the negative impacts to humanity. I'm not going to go into detail about this just yet, because I want to dedicate a lot more time to this subject. But already, we are seeing people lose their jobs due to the increased efficiency and reduced cost that artificial intelligence provides. I maintain that people who know how to use AI are the ones who are taking people's jobs, for now.

General AI

Moving from practical to theoretical, we focus our attention on Artificial General Intelligence (AGI), or General AI. AGI is a field of theoretical research that attempts to create software with human-like intelligence and the ability to teach itself. The goal with AGI is to create machines that are able to learn and perform tasks that it was not specifically trained for. I think Amazon does a good job explaining the difference between narrow AI and general AI. "Current AI technologies all function within a set of pre-determined parameters. For example, AI models trained in image recognition and generation cannot build websites. AGI is a theoretical pursuit to develop AI systems that possess autonomous self-control, a reasonable degree of self-understanding, and the ability to learn new skills. It can solve complex problems in settings and contexts that were not taught to it at the time of its creation." (2). 

There are several theoretical approaches that facilitate general intelligence research. The "symbolic approach" assumes that computer systems can develop AGI by representing human thoughts with expanding logic networks based on physical objects. The symbolic approach cannot replicate perception, like humans can, but uses an if-else logic that allows it to interpret ideas at a higher thinking level. The "connectionist approach" focuses on replicating the human brain structure with neural-network architecture. In other words, scientists hope to replicate human brain behavior to demonstrate low-level cognition, like perception. The "universalist approach" focus their research at the calculation level, attempting to create formulas that will be repurposed into practical AGI systems. Other than a hybrid approach, which use multiple approaches for their research, the final approach is called "whole organism architecture". This approach integrates AI models with a physical representation of the human body. Researchers who use this approach believe that AGI is "only achievable when the system learns from physical interactions". While I don't know for certain, I'm reminded of the robots that Boston Dynamics have been creating and question how close they have been to achieving artificial general intelligence.

Super-Intelligent AI

The final evolution of our AI journey leaves us with artificial superintelligence (ASI). This is where things get real interesting, and possibly scary. IBM defines ASI as "a hypothetical software-based artificial intelligence (AI) system with an intellectual scope beyond human intelligence. At the most fundamental level, this superintelligent AI has cutting-edge cognitive functions and highly developed thinking skills more advanced than any human."(3)

Woah, what does that even mean? Well, that means that machines using ASI will be the most intelligent thing that we've ever seen, barring any religious beliefs. These machines would be vastly more intelligent and capable that humans could ever be. If ASI becomes possible, the rate in which they would continue to advance and evolve would be extraordinarily fast.

What implications would ASI have on humanity, on our planet, on our solar system, galaxy or universe? I have no idea. No one does! If artificial superintelligence ever becomes a reality, I pray that will not lead to the demise of humankind!

Conclusion

I cannot help but think about movies and shows such as the Matrix, Star Wars, Star Trek and more. Will we end up living side by side with robots like R2D2 and C3PO? Will we limit AI development out of fear of the unknown? On the one hand, artificial intelligence is magical in all of the best ways. The narrow AI systems that we are using today have helped so many people become more efficient, learn new skills, and improve healthcare. As we humans tend to do, it is also being used for manipulation, for committing crimes, and other sinister purposes. What happens when artificial intelligence evolves into AGI and ASI? No longer will they require human input. In fact, human intervention will just slow them down. No one can know for certain what the future will hold, but there is no doubt in my mind that we are amidst a technological revolution. Will you take the red pill or the blue pill?

How I Used AI in this Article

• Dall-E created the image at the beginning of this article.

About the Author

Dan Doerr has dedicated his life to defense and security. He was introduced, first hand, to national defense and security while serving in the U.S. Army. Since then, he has achieved his Master's Degree from the Josef Korbel School of International Studies, focusing his research on International Security. Dan holds his CompTIA Security+ Certification and is an active volunteer for the Denver Chapter of the Information Systems Security Association (ISSA). He loves technology and when he isn't diving head first into the tech world, he likes to do the opposite by getting out into nature.  

Sources

1. Taloni, A., Borselli, M., Scarsi, V. et al. "Comparative Performance of Humans versus GPT-4.0 and GPT-3.5 in the Self-Assessment Program of American Academy of Ophthalmology." Scientific Reports 13 (2023): 18562. https://doi.org/10.1038/s41598-023-45837-2. 
2. IBM. "Artificial Superintelligence." IBM. Accessed June 4, 2024. https://www.ibm.com/topics/artifical-superintelligence.
3. University of Wolverhampton. "Narrow Artificial Intelligence: Advantages, Disadvantages, and the Future of AI." University of Wolverhampton. Accessed June 4, 2024. https://online.wlv.ac.uk/narrow-artificial-intelligence-advantages-disadvantages-and-the-future-of-ai/.