My only worry is that if many areas of current AI might not be at all relevant to AGI research (for instance, you yourselves say that AI and AGI are not the same thing on the link - ai.stackexchange.com/questions/6267/… ) and I might be wasting time learning them.

@nbro The book I have in mind dates from the mid-1990s and covers genetic algorithms. Not exactly career advice today, I would say ;-) What I wanted to know is what CS concepts, level of mathematics and possibly programming language are required. Since the book is old, there is no preview on Amazon.

@user56417 So, AI is not the same thing as AGI. AGI stands for artificial general intelligence, which is an artificial intelligence that can be applied to many problems, like us humans - we can solve so many problems that we take for granted. Now, a program that is intelligent but solves only 1 problem is not general, it's narrow, so it's not AGI. A famous example is a program that plays chess or go. It cannot solve any other problem, so it's not AGI.

Now, this doesn't mean that concepts that you learn e.g. in ML are not also helpful to understand Hutter's work, which is actually AIXI, as far as I know, he's not done much more than research on AIXI

Many concepts that you learn e.g. in a Machine Learning course, like Reinforcement Learning, will be very useful to understand AIXI. In fact, AIXI is a reinforcement learning agent

Concepts like planning and search are also useful to understand the ideas behind how AIXI works

But if you want to get into the details of AIXI, you will need to understand other more technical ideas, like Turing machines

Other things like generative AI or neural networks are not particularly useful to understand AIXI or Hutter's work

So, the first thing I recommend you study is reinforcement learning at least the first chapters from Sutton and Barto's book. If you don't understand those chapters, you should learn the things that you don't understand

Anyway, if you have time, following a full ML course can be helpful because you willl learn about prediction, induction and models, which are also concepts used in AIXI

No need to spend time on courses that focus on neural networks or generative AI. Just a general ML course is OK

@Tsundoku To understand genetic algorithms, you don't need much. They are based on very simple ideas. You need to understand what optimisation and numerical algorithms are. You need to understand what an algorithm is for sure. Data structures are also important, but only the basic stuff, matrices, vectors. This is really the basics of Computer Science.

Any programming language is fine. I'd recommend Python, because it's widely used in AI nowadays. There's this library github.com/DEAP/deap I used in the past

Anyway:

@nbro Thanks for the help.

@nbro Sorry, I wanted to clarify something. Correct me if I wrong, but I think you have misunderstood my question. I am not looking to get into Marcus' work including AIXI (I might look into it in the future, if it is still relevant at that time).

I mentioned Marcus in the context of his book recommendations on his webpage. My question was that learning from those books seems daunting to me, so, would it be a good idea to learn the concepts of current AI to ease into this field and only later look into AGI (and his book recommendations).

For instance, you mention don't look into neural networks (probably because you thought I want to focus on Marcus' work) but Marcus mentions it's importance on his webpage. To sum it up, I want to do novel research in AGI and would I be wasting my time if I learn current AI concepts first?

@user56417 No, you would not waste your time if you learned the current AI concepts, if your goal is to do research in AGI. But there might be concepts that you may not need to focus too much on. But learning the basics will not be a waste of time

@Tsundoku To get started with genetic algorithms, you don't need to know much about optimisation and numerical algorithms, just the basics, like you need to know what an objective function is, what it means to find the minimum/maximum of a function (basic calculus), etc, because genetic basically are optimisation algorithms with objective functions called fitness functions