- Knowledge and information are made transferrable by the invention of language.
- Knowledge and information are made collectable or archivable by the invention of writing.
- The symbolic elements of writing (whether single symbols like letters or pictograms, or clusters of symbols like words) enable both sharing of information and the integration of information.
- Knowledge and information are made much more sharable by the invention of printing.
- Sharing and integration of information is enabled on a much grander scale by the digitisation of information; this, combined with digital communication and digital computers makes bringing together different written elements easier.
- nformation and knowledge was always interpretable by (well-informed) people; but modern AI (transformers, and other learning techniques) enables a sophisticated form of statistical interpretation that (at least superficially) resembles interpretation by well-informed people.
- The digitisation of (virtually) all scientific and mathematical knowledge, as well as laws, etc., used in conjunction with huge learning systems (transformers etc.) equipped with persistence and internal evolution [1] (enabling long-term learning, after initial training) allows these systems to be able to interpret and integrate information very effectively.
Do these systems “understand” the information they are using?
But what do we mean by “understand”? Personally, I understand something when it connects on to the rest of what I know about – when it is integrated with the rest of my existing knowledge. However this is a very vague concept. Do machines understand anything? Does forming a hugely complicated network of weights and delays which encode deep statistical regularities in the training data equate to “understanding”?
[1] See the work of Ben Goertzel from Eurykosmotron.
Lestheprof's Blog 
Leave a comment