Read on Twitter
Lately I've been musing on types of patterns/intelligence/knowledge that are impossible for statistics to discover
What kinds of truths can only be learned through grid search or memorisation?
And if these truths exist - what does it mean if we're less likely to find them?
What kinds of truths can only be learned through grid search or memorisation?
And if these truths exist - what does it mean if we're less likely to find them?
Cryptography - for example - is a great example of intentionally encoding knowledge in a way that statistics is useless to find it.
And what does it mean if there are other things in our own life and society which are presented in this way.
No real answers yet - but I think it's an interesting question - particularly for checking our assumptions on modern techniques for ML.
I also think it's an interesting existential question - given that statistics is increasingly viewed as the primary/only source of truth.
If you're tempted to think this way - fwiw - I think it's a surprisingly new phenomenon and is quite shortsighted.
If you're tempted to think this way - fwiw - I think it's a surprisingly new phenomenon and is quite shortsighted.
For example - consider what evidence you would need to believe that we're living in a simulation. By very definition - you would need to observe something that 100% defied the patterns statistics indicates exist
But if the source of truth is just statistics - this is unlearnable
But if the source of truth is just statistics - this is unlearnable
Also - healthy psychology sortof works this way. Some things you can only "know" by experiencing it yourself.
Meaning there's no experimentation externally to validate something that is totally subjective.
But subjective things can still be just as true.
Meaning there's no experimentation externally to validate something that is totally subjective.
But subjective things can still be just as true.
So in this way - if you were to eliminate anything but statistics (possibly including the scientific method) as your "source of truth" - there are whole categories of truths which you could never discover - even if they were true.
In practice most people don't do this - we use heuristics and all sorts of approximations and social signals to choose our beliefs.
But I observe that many - particularly people in academia/tech - like to live under the delusion that they only rely on observable stats.
But I observe that many - particularly people in academia/tech - like to live under the delusion that they only rely on observable stats.
