I'd really rather prefer there was some online repository of facts, constantly updated via Machine Learning over publicly available information, which might give up to the second responses of the kind 'Our current understanding is ...'
Is there one already? Wolfram Alpha is close, I guess. Wikipedia is not quite what I'm thinking of. Google is working on something, not sure if it's public though. Their search engine.. also not quite what I'm thinking of.
This is an example of precisely the kind of false-positive that you would expect to have emerge when you leave it up to machine learning over publicly available information to try to figure out the answer to a question.
I kind of agree TBH, although I would say their search engine (as it is today) and a 'fact-repo' are quite different things. I'd expect a higher level of rigour in the fact-repo ML checking processes.
I remember Freebase, I even recall fact-checking on it a while ago.. I'd be interested to see how their current efforts are shaping up - if it's all ML or requires human assistance.
They use WikiData now to fulfil the same basic function as Freebase, so it does seem that they require some human assistance. Don't know if they've ported everything over from Freebase yet, will have to check up on that.
Comments
I'd really rather prefer there was some online repository of facts, constantly updated via Machine Learning over publicly available information, which might give up to the second responses of the kind 'Our current understanding is ...'
Is there one already? Wolfram Alpha is close, I guess. Wikipedia is not quite what I'm thinking of. Google is working on something, not sure if it's public though. Their search engine.. also not quite what I'm thinking of.
What/where have I missed?
This is an example of precisely the kind of false-positive that you would expect to have emerge when you leave it up to machine learning over publicly available information to try to figure out the answer to a question.
I kind of agree TBH, although I would say their search engine (as it is today) and a 'fact-repo' are quite different things. I'd expect a higher level of rigour in the fact-repo ML checking processes.
The Google one is based on an older, public one: http://www.freebase.com/
I remember Freebase, I even recall fact-checking on it a while ago.. I'd be interested to see how their current efforts are shaping up - if it's all ML or requires human assistance.
They use WikiData now to fulfil the same basic function as Freebase, so it does seem that they require some human assistance. Don't know if they've ported everything over from Freebase yet, will have to check up on that.