How the Mind Solves Problems and its Implications in Creating AGI
Human creativity as fractals of conjectures and refutation loops
If we accept the idea that all intellectual pursuits arise from problems, that creativity is necessary for solving these problems, and that problem-solving involves the generation and testing of hypotheses, then we can arrive at a solid theory and definition of creativity. By understanding creativity in this way, we can see how it plays a crucial role in all aspects of intellectual endeavors and problem-solving.
How does the mind solve its problems?
Problem-solving is a fractal process of continuous conjecture and refutation loops. Fractal because at a given moment we are focused on solving a portion of the problem that would get us to the next step - this creates a conjecture-&-refutation (C&R) micro-loop, let us call it micro-CR. We either resolve the loop by finding a tentative solution for the current step and move on to the next step or jump to another micro-CR in search of the solution. Arriving at the next step is itself a process of micro-CR. Thus we move from micro-CR to micro-CR, picking up bits and pieces of knowledge gained along the way. These bits and pieces of knowledge are stowed away in our stash of knowledge knapsack, mostly unconsciously. When we encounter a new problem, we dive into this knowledge knapsack. This ability to think abstractly and break out of repetitive loops is what sets humans apart from animals, according to Douglas Hofstadter in his book Fluid Concept Creative Analogies. This way of human problem-solving differs from animals when they encounter a problem. Animals get stuck in a loop of repetitive actions as they can not necessarily think to get out of the loop. Humans can not only take a step back and decide to jump out, but they can also carry these knowledge nuggets with them and use them in any future problem scenarios. Animals can not do this - or if they do it is in a limited capacity.
The metaphor of an explorer moving from problem to problem, using their knowledge knapsack is a fitting way to describe the creative process of problem-solving. Creativity involves continually testing hypotheses and incorporating new knowledge and insights gained through these micro-loops. We solve small problems, integrating their solutions into existing knowledge to these micro problems into the current set of knowledge, and move to the next micro problem - building on it, until the macro problem is solved.
This process of micro-conjecture-refutation is always limited by the intelligent entity’s current state of knowledge. If one doesn’t have knowledge about the planet’s rotational patterns and the theory about how it changes seasons on earth, one is limited to solving problems that do not rely on that knowledge. Either one discovers it on their own or learns by reading from others.
One of the things that separate an intelligent entity’s creative processes vs. any current “AI” system is the ability to adapt and incorporate newly learned knowledge/discoveries into its existing pool. An intelligent entity is truly creative because it can go back to this ever-growing edifice of knowledge knapsack, and pick likely relevant items in the current quest to solve the current parochial problem. The once-thought-to-be suitable item may turn out to be not so - after the current micro loop of conjecture-refutation is resolved. The explorer has many items in their backpack of knowledge, and they can pick a next one and next one and next one until the loop is satisfied.
Bob Ross standing in front of his half-finished painting, considering his next brushstroke, is a perfect example of creativity at work. Where does that stroke come from?
It must be the result of his brain unconsciously drawing upon his stored knowledge and guessing at the best solution. He is in a micro-CR loop, testing out different ideas and seeing how they fit into the bigger picture of the painting. If a stroke doesn't work, he'll try to fix it and see if that solution is successful. If it is, he moves on to the next stroke. When we zoom out and look at the finished work of art, it's the result of countless micro-CR loops and the integration of all the successful solutions.
How can we make AGIs?
How does the current AI system need to change to be general? The current systems are given a large amount of data, and patterns are drawn from the limited data to output an inference about the new input. This is limiting and there is no adaptability in the system, therefore just adding more data will never get us a general solution. What needs to happen is to import the creativity module into the system. It is easier said than done, as no one has programmed this module yet, but understanding what creativity means is the first step toward it.
In The Beginning of Infinity, David Deutsch writes about creating AGI:
The analogue of the idea that AI could be achieved by an accumulation of chatbot tricks is Lamarckism, the theory that new adaptations could be explained by changes that are in reality just a manifestation of existing knowledge - Artificial Creativity
Taking examples from human learning, each new piece of knowledge that one has learned from birth up until this point has been through a loop of conjecture and refutation. It has thereafter been stored as knowledge once it has shown to be successful. It has been acquired anew for each human being by creatively conjecturing and refuting ideas in their minds. In other words, no one has ever learned a task by transmutation of a piece of knowledge from a book or another human.
It may seem impossible to mechanize the creative process of problem-solving, as it involves a lot of uncertainty and randomness in the way that we generate and test hypotheses and integrate new knowledge. However, we can begin by replicating the overarching process of entering into micro-CR loops, storing tentative solutions as knowledge, and adapting to changing circumstances. This would involve the ability to draw upon previously held knowledge in a flexible and adaptable way and to move on to the next problem if the solution is successful. While it may not be possible to create a recipe for creativity, we can still attempt to replicate the key elements of the process in a mechanical way.
Contrast the way that a chatbot like GPT works with the way that a human learns and uses knowledge. While GPT can generate seemingly interesting and accurate responses to input, it lacks the ability to store and retrieve this output or to determine whether it was useful to the user. In contrast, humans learn by generating guesses or hypotheses, testing them, and adjusting their understanding based on the results. In order for a chatbot to become more sophisticated and more similar to human problem-solving, it would need to be able to interact with users in a more flexible and interactive way, asking relevant questions and using the interactions to its advantage. It would also need to be able to store and retrieve these interactions efficiently.
While progress in the field of artificial intelligence has been impressive, we are still far from achieving true AGI. There are likely many pieces that we have yet to even consider that are necessary for the creation of a truly intelligent machine. While tools like large language models and chatbots have given us a taste of what is possible, we should not be fooled into thinking that we are close to achieving AGI. It is important to continue thinking critically and deeply about the challenges and potential solutions to creating AGI. This blog post aims to provide some food for thought for those who are serious about making progress in this field.
Our ability to self criticise, is key to our freewill/creativity,now all we need to do is break it down into logical steps 😜
Breaking it down to understand and build from first principles is the engineering approach needed at some point in building agi.
We need to bolster our philosophy about this first. 😌