AI, ANI, AGI, ASI...What?!

Disclaimer: Views expressed are mine and mine alone, and do not necessarily represent those of my firm or my colleagues. Portions of this article were generated by two of the leading GenAI tools

The world woke up to multiple provocative headlines over the past 24-48 hours about how the leaders of one of the hottest AI startups is seeking additional funding to build AI "superintelligence". As with anything related to #AI these days, there is the reality and there is the narrative. FWIW, he's talking about "General Intelligence" not "Super Intelligence"; at least for now.

As I was discussing this with my family, the inevitable question came up: what's the difference? There is no universally agreed-to definition of AI, so I set about asking our future assistants (or overlords) not only about the various types of AI but also to help me explain that in plainspeak. Below is an adaptation of what two of the leading LLMs told me in reponse to various prompts. I also asked them to cite their sources and I got a reference to a number of scientific papers. Though a simple Google search came up with this 4 year old article which did answer my original Q. So...search > genAI?

Anywhoo, hope you find this of use.

PS: I am not going to talk about #ML, #DL, #GenAI. Plenty's been written on those :).

Classification based on capability: ANI vs. AGI vs ASI

1. Artificial Narrow Intelligence (ANI), also known as weak AI, is the type of AI that we are most familiar with today. It is designed to perform specific tasks, such as playing chess, driving a car, or recognizing faces. Some commong applications of ANI include:(a) Chatbots, such as Siri and Alexa, use natural language processing to interact with humans, (b) Self-driving cars use computer vision and machine learning to navigate roads and avoid obstacles, (c) Fraud detection systems use machine learning to analyze data and identify fraudulent activity.?
2. Artificial General Intelligence (AGI), also known as strong AI, is the type of AI that we often see in science fiction. It is capable of understanding and learning new things in the same way that humans can. It does not exist for now (although many many companies, including the aforementioned startup, want to be the first to get there). Expected applications of AGI include: (a) solving complex societal problems that are currently beyond the capabilities of humans and their ANI machines such as reversing climate change and solving world hunger, (b) Scientific and technological discoveries such as solving the Grand Unified Theory (c) Creating new art by merging or fusing various schools and traditions.?
3. Artificial Superintelligence (ASI)?is a hypothetical (for now) type of strong AI that is even more intelligent than humans. Expectation is that ASI will be capable of solving complex problems and making advancements that are beyond human comprehension. Expected (desired) applications of ASI include:(a) Solving some of the most difficult scientific problems facing humanity that don't seem to have an answer in the near future such as curing cancer and human space travel (b) Creating new technologies that are far beyond our current capabilities, such as teleportation (c) Making decisions that affect the future of humanity such as whether or not to colonize other planets.

Classification based on awareness: Reactive vs. Limited Memory vs. ToM vs. Self-aware:

1. Reactive machines: Reactive machines are the simplest type of AI. They can only react to their current environment and do not have any memory or knowledge of the past. Examples of reactive machines include chess-playing computers and self-driving cars.
2. Limited memory: Limited memory machines are a more advanced type of AI that can store information about the past. This allows them to learn and make better decisions over time. Examples of limited memory machines include recommendation systems and spam filters.
3. Theory of mind: Theory of mind AI is a hypothetical type of AI that would be able to understand the thoughts and emotions of others. This would allow them to interact with people in a more natural and intuitive way.
4. Self-aware: Self-aware AI is a hypothetical type of AI that would be aware of its own existence and place in the world. This would allow them to make decisions that are in their own best interest, rather than simply following the instructions of humans.

(non LLM output) Classification based on type of interation: Automated vs. Assisted vs. Augmented vs. Autonomous (full credit to Dr. Anand Rao, PhD, MBA for this one)

1. Automated: Where machines / algorithms do pre-programmed tasks with pre-defined parameters and limited new data in a predictable manner without need for human intervention BUT no feedback loop. Humans have the final decision and evaluation rights and "evolution" needs to be explicitly programmed
2. Assisted: Where machines / algorithms assist human actions and decision making by processing existing and new data to predict a number of outcomes / actions which the humans can then pick & choose from. Again, feedback loop does not exist and "evolution" needs to be explicity programmed. While there is a lot of attention, hype and $$s being put into the next two categories, for a lot of the industries and humankind there is still tremendous amount of productivity unlocks from categories 1 and 2.
3. Augmented: Where machines / algorithms augment human actions, capabilities and decisioning by processing data to prescribe a number of outcomes / actions which the humans can pick & choose from or delgate the decisioning to the algorithm. In this category, the algorithm is capable of learning decisions taken vs. not taken to improve its capabilities and (i.e., the algorithm can be trained or pre-trained, and a feedback loop exists). Majority of Weak AI (ML / DL / GenAI / RL) would all fall in this category.
4. Autonomous: Where machines / algorithms substitute and/or supercede human actions, capabilities and decisioning. A full circle to automation in some ways (humans are not in the loop). Strong AI would definitely fall into this category, though many could argue that many Weak AI applications (level 5 driving) would be in this category as well.