top | item 12631405

(no title)

cftm | 9 years ago

You do realize that we're still dealing with the world of models that need be trained. Models are only as good as the data that is given to them, and data is only as good as those who collect it. Neural Networks aren't even close to real AI, they're black-box models - believe them blindly and catastrophe will follow. An excellent example I recently read pertains to a hospital attempting to use a neural network in determining whether to send pnemonia patients home or have them stay in hospital for treatment. The model told them to send asthmatic patients home. Why you may ask? Well, asthmatic patients were always triaged to the ICU, thus the results told the hospital to send this group home. We don't even understand cognition in the human brain, how can we expect to stumble upon it with digital systems?

discuss

order

No comments yet.