What are the three levels of AI capabilities?
In the previous article in this series on Artificial General Intelligence (AGI), we began our journey on The Road to AGI. I provided an overview of the topics we’ll encounter on our journey from the limited AI we have today to the highly capable AI we’ll have in the future.
However, to get started on our journey, we need a road map. So, in this article, you’ll learn about the key milestones on our journey — the three levels of AI.
When you think of “Artificial Intelligence”, you might imagine modern AI tools like ChatGPT, Amazon Alexa, or Tesla’s self-driving car. We refer to this type of AI as Artificial Narrow Intelligence (ANI) or “Narrow AI”.
ANI is a type of AI that is focused on performing a narrow set of specific tasks. For example, Amazon Alexa listens to your voice commands and performs an action. ANI is the type of AI that exists today — which is currently revolutionizing our world.
ANI is a machine that takes data as input and produces data as output. The input can be tabular data, text, images, audio, video, etc. The output can be a decision, a prediction, an action, text, images, audio, or video.
On the other hand, when you think of AI, you might imagine sentient robots like R2-D2, Data from Star Trek, or HAL 9000. We refer to this type of AI as Artificial General Intelligence (AGI), or “General AI”.
AGI can perform a wide range of general-purpose tasks. For example, an android like Data can walk, talk, use tools, and play music. AGI is a type of futuristic AI that doesn’t currently exist today. However, it is the end goal of many AI research projects and startups.
We define AGI as a machine that can perform a wide array of economically valuable tasks better than an average unskilled human worker. Essentially, it’s an AI that’s as good as most humans at doing work that humans do every day.
When you think of AI, you might also imagine super-intelligent superheroes and supervillains like Vision from The Avengers, VIKI from the movie iRobot, or Skynet from The Terminator franchise. We refer to this type of AI as Artificial Super Intelligence (ASI) or “Super AI”.
ASI can perform every task that a human can perform better than any human. For example, vision is smarter, stronger, faster, and more powerful than any human. ASI doesn’t currently exist, and we don’t know if or when it might come into existence.
However, once we achieve AGI, it’s likely that ASI will emerge rapidly. An AGI as capable as an average AI researcher will be able to recursively self-improve. A million of these AI research agents continuously running experiments to incrementally improve themselves could quickly lead to ASI.
Understanding the three types of AI is the first step in understanding how we will get from ANI to AGI and, eventually, ASI. We’re ready to begin learning about the problems in AI research we solved over the past few decades. This will provide a foundation for understanding where we are and where we’re going.
To learn more, please check out the next article in this series: AI of the Past.