The power of knowledge graphs
We explored a lot of options, from graph databases to scraping the USPTO and running machine learning models over it.
Finally, we concluded that in order to meaningfully generate new ideas, we needed computers to understand the map of existing ideas first.
The general idea was as follows:
For a computer to understand that, you would need to map all humankind’s knowledge in a graph. There have been some attempts at doing so, and we even tried to leverage on those for Unpatent:
But the issue is always language. Language is subjective by design. Words don’t have formal specs and implementations, like software does.
Language is wet code, not dry code. This makes it so hard for computers to enter our world, understand it and build on their own.
Today I won’t go into the AI rabbit hole, but rather on how we can improve our systems today using tools to reduce the gap between language and code.
Currently, Wikipedia defines motorbike as:
A motorcycle, often called a bike, motorbike, or cycle, is a two- or three-wheeled motor vehicle
This doesn’t tell us much, specially about what it needs to be in place in order for something to be a motorbike. In an idyllic world, a word would take meaning because of its place in the knowledge graph.
For example, motorbike would be roughly defined as follows:
Of course this would be inside an infinite graph, so Motorbike would be related to Transportation somehow and Battery would be indirectly related to Wind or Solar.
Also, the above could be expressed as code:
This gives us some interesting properties:
This can now sound very abstract. What are the use cases, apart from AI?
I’ll enumerate a few:
I just discovered Mermaid, which is a tool that allows to quickly chart by using code. I have been looking for something like this for a long time.
Now I plan to start writing code for multiple processes I want to formalize, and also for my own knowledge and mental models I have been accumulating from life and reading.
I’m genuinely amused by knowledge graphs, and I think they could help reduce misinterpretations and improve accuracy in how humans and organizations communicate their view of the world.