This week is AI week, both to date and going forward. Some don't understand the difference between AI and AGI, the latter being artificial general intelligence, denoting a system that behaves like a human being in capabilities. So, for your amusement I've put together, with credit to the YouTube channel AI Uncovered, a description of the seven levels of AI.
- The first of these are rule-based AI, or single task systems. These use a predefined set of rules or an algorithm provided by a computer programmer. An example of this are the early chess programs. These were based on a set of decision points based on the play to date, plus rules and info stored. There is no adaptation, context or learning, just you against the programmer and the amount of data they had entered into the system. This approach is good for problems with clearly defined rules like processing a form or an automated basic phone answering system, or chatbot, before being passed onto a human if you're lucky.
- Stage 2 are context awareness and retention systems. These systems remember what has occurred before and use that as the context for the next interactions. Phone systems like Siri or Google assistant are examples of this where you ask a question and then ask a follow-up without directly referring to the last one. The system can infer the link and respond based on that. This is a step up from the stage one interactions. ChatGPT is an example of this stage. There is no independent thinking but context and previous interactions inform the results of the next one. Yes, ChatGPT is still only at Stage 2 of the AI scale.
- Stage 3 are the domain specific mastery systems. This takes Stage 2 and applies it to a single field exceeding the capability of a human in that field, or the expert level within the given domain. IBM Watson is one example of this, beating out humans in a quiz show, or Deep Mind AlphaGo that can beat the best human at the boardgame Go. These systems have a deeper understanding of the area than any human can. They analyse data, identify patterns, learn from their mistakes and make informed decisions in their area of expertise. This is the adult level of AI in a specific field.
- Stage 4 represents thinking and reasoning AI systems. This level mimics human behaviour. Complex concepts are able to be understood, creative ideas can be generated and unfamiliar problems can be solved. Using machine and deep learning the process improves over time. Such a system could read a book, understand it and even work out the motives of the characters. It could look at the stock market and make investment recommendations. Our current level of AI is somewhere in the middle of this stage. This is not yet the equivalent of the human mind but is specialised in reasoning and thinking tasks.
- Stage 5 is artificial general intelligence (AGI). This is the generation of a new mind that is similar to a human in intelligence. Currently, at least as far as I'm aware, this is still theoretical. Like us, it makes decisions based on understanding, learning, adaptation and the implementation of knowledge. It is something that will either just appear on its own or be trained through the stages of childhood until it reaches the same kind of understanding we do as adults. Unlike stages 3 and 4, it will have a general coverage of multiple areas like most humans do. It will be conscious and self-aware in the same way we are. There are some arguments around if it will also suffer from things like depression, PTSD, addiction and so on. This will in all likelihood be a brief and transitory stage in AI development and is currently a controversial subject area.
- Stage 6 is artificial super intelligence (ASI). This gets harder to define. Take Stage 5 and speed it up by orders of magnitude and make it more intelligent. ASI will solve problems we can't. It will be faster, more accurate, more creative and any number of other mores. It could solve problems like what is gravity and things we haven't even thought about. Given the resources, AGI would evolve into ASI quite rapidly.
- Stage 7 is what people refer to as the AI Singularity. This is past the point where we could even understand what an AI has become. Those that have seen the movie Lucy will have some possible frame of reference for this. In the intelligence realm we would be less than ants by comparison. AI self-improvement would be exponential and potentially uncontrolled. The date Aug 16, 2036, keeps appearing in AI results, perhaps this is when the singularity occurs or perhaps that idea is rubbish. It will not be the date of the heat death of the universe.
- There are six levels beyond this you can find on AI Uncovered. Now you know.
James Hein is an IT professional with over 30 years' standing. You can contact him at jclhein@gmail.com.
