No Stefan, we know humans are really good predictors, the actual test is to
make a really good predictor for text so that your compression is extremely
high and so that it usually has a high probability for the next letter or bit
to predict that it should predict during decompression in a lossless generator
and hence the error it stores is very small ex. 10MB only.
Compression/decompression is done as a chain of predictions. If you can
compress the 100MB enwiki8 to 10MB and still regenerate the 100MB back
losslessly perfectly then this means your prediction accuracy is extremely
high. Now if you take probabilities for the next letter to predict during
decompression and let it randomly choose any of the top 10 letters or bits, it
can regenerate not the enwiki8, but related data you'd want too. And that's how
you make a lossless text predictor able to return us new data we never had
directly.
------------------------------------------
Artificial General Intelligence List: AGI
Permalink:
https://agi.topicbox.com/groups/agi/T409fc28ec41e6e3a-Meb07a0ed0463c724dd035b55
Delivery options: https://agi.topicbox.com/groups/agi/subscription
