Ask box: Are you still related to ConceptNet?
A comment in the ask box asks:
Are you still related to ConceptNet?
Kinda? I don't do much with it anymore.
ConceptNet is a crowdsourced knowledge graph, intended as a basis for natural language processing. It's made of facts that, in one way or another, people actually intended to be used by computers.
I started working on it in 2005 and was its designer and lead developer until 2021, when I left Luminoso (the company that supported its development) for better opportunities such as biking to Vermont. It's now kind of frozen in time. The database and infrastructure are still hosted at Luminoso, and I don't have access to the servers.
It's still serving its website and API with only basic maintenance, and I'm proud of the database and site I designed that make that possible. I can't really provide support for it, except maybe to answer basic API questions.
I see ConceptNet as kind of a throwback to an era when it was possible to be optimistic about AI without buying into exploitative hype, and when there was a goal that an AI system could "know" facts as opposed to just predicting words to sound like it does. I don't think it's being used much these days, now that all the funding has lined up behind spicy autocomplete as a substitute for basic knowledge. But maybe I'm glad that, as far as I know, ConceptNet isn't being used in the big fake hype machine that's ruining everything.
Some of the things that were most satisfying about ConceptNet:
- seeing projects people built on its free API, like the game Robot Mind Meld, where you and the computer system try to converge on saying the same word
- hearing about how ConceptNet was helpful in Portuguese NLP, when most other NLP tools only supported English (and often didn't even specify, just saying "natural language" and meaning English)
- the fact that a couple of state-of-the-art results gave me a platform to talk about the widespread biases that were appearing in NLP, as computers learned to mimic human prejudices