It's been a lifelong dream of mine to connect ideas from all across human knowledge in meaningful ways depending on the context and I would have liked to train a program / bot / AI / something to do it for me in order to get through more knowledege than I could in one lifetime.
I wrote a small rambling about this here https://stirringsteer.blogspot.com/2014/11/skeletons-of-ideas-or-cross-field-is.html
- my own intuition, introspections and self reflections
- Douglas Hofstadter's Fluid Analogies, particularly Sequence and Copycat
- Melanie Mitchell's work on AI analogies
- Wikipedia seems to be the best source of concepts and building blocks for all human knowledge.
- That and books. Definitely books from the public domain. Textbooks, too and other paid books as long as someone purchases them for the Cross-Field Ideas AI.
- source of knowledge: wikipedia, public domain digital books. Format: English language text
- parse source text from the English language into concepts, building approrpiate links between concepts based on the source text and in the appropriate contexts. This basically builds the AI's knowledge.
- querying the knowledge. There are multiple applications to this.
- helping people learn by asking questions and exploring, going on serendipituous journeys
- finding outside of the box solutions to problems
- innovating by finding isomorphisms between fields of konwledge that to humans seem disconnected. innovating intentionally instead of waiting for accidents to happen and for the right people to meet and have the right conversation
- expanding human knowledge based on existing knowledge (this won't work with concepts that don't exist yet since the AI doesn't have access to reality to run its own experiments)
- finding gaps in knowledge and patterns
Purpose and actions can be deduced from an objects attributes. Example: earth shovel and snow shovel. // to be written
What is an apple?
What is an apple (fruit) to me as a human? Let's say I've never seen this fruit before. It's a kind of fruit so it's edible. I can eat it. But that implies I have a way to eat the apple. What if I didn't know what a fruit was? It's something you can eat, it can have different colors, it grows on trees, it tastes sweet and sour, in English it's called "apple". I think we as humans learn new concepts by experiencing them with our body's senses and attaching words in languages. So to a human, a concept is a collection of body expereince memories, words in languages referring to the concept and stories about and involving the concept heard from other people (other humans' body experiences about the concept). But an AI doesn't have such senses. All it has is the word "apple" to identify this concept by. And for my purposes with cross-field ideas I think that is good enough because I want to find isomorphisms between collections of concepts in different subjects. Being able to answer questions such as what do origami and optics have in common? What do economics and gastronomy have in common? Taking this even further, what are the applications of apples? You can turn them into a pie so that links it to cooking. You can drink the juice. But it doesn't taste good if you mix apple and flour and try to eat it as is. Humans make new knowledge by trying new things and experiencing it in reality and perceiving the result with their bodies. An AI stuck in a computer without outside sensors can't do that. So then we would need to pass it the knowledge that we perceive with our bodies and have it generate new ideas for us to try in reality. But the AI alone cannot know what the result of a new combination is. Unless it has a kind of body of its own and it is programmed to use it to check results and learn from its actions.