When I read the article "Knowledge Graphs: Opportunities and Challenges" this morning, I thought again about the relationship between symbolic AI, say, in the form of the Semantic Web, i.e. RDF and OWL, and statistical AI in the form of neural networks, language models and transformers.
ChatGPT shows how far statistical AI has come, here we can clearly see approaches to general intelligence, AGI.
Where is the sense in dealing with symbolic AI, with formal logic? There is no sign of anything comparable happening here.
Perhaps logic is simply the wrong way to understand the world.
And yet, humans have been researching formal logic for millennia. Fretting over their own imperfections, their mistakes, the inaccuracies in human reasoning.
And - funnily enough - we have now developed algorithms that do the same here: very powerful, but full of mistakes, inconsistencies, and plain lies, oh no, "hallucinations."
Can we trust ChatGPT with our accounting? Certainly not today, but will we ever be able to?
Because our respective tax authorities would like reliable results, and I guess GPT could get more creative here than the tax investigators would like.
So maybe logic, and symbolic AI, is more about the search for truth and a basic understanding of the world, and maybe ChatGPT can help us here as little as we can help ourselves.
Will Knowledge Graph, which maps the various terms and phrases used in product development and regulatory agencies, for example, and which can use classical logic to translate one into the other, be superseded by a learning system like GPT?
It will be exciting to watch the further development. I don't think at the moment we can overlook even in the slightest what will result from this development, what the split between humans and algorithms will look like, where the split between formal understanding and finding one's way in the world will go.
Thrilling.
And this article obviously needs urgent revision, I seem to be failing to write a lot of what I think. But my thoughts on this are just not organized yet, I can't write what I think because I don't know what I know - I wonder if GPT has something like that? :)
Add new comment