On this website (and still here to read), we posted a post that we had had artificial intelligence (AI) write all over it for fun. The post looked real, had even included some relevant-looking information from the internet and made up some names that provided quotes. Brilliant, on the one hand, but also a bit scary. The Literary Fund, for instance, was rightly concerned that the artificial intelligence had simply brought in that - beyond any doubt - institution. Consequently, the words attributed to the fund in the article were never used by that fund.
So such things are possible in AI. With GPT 3 (Generalized Parsing Toolkit, which develops the linguistic part of the algorithms, at level 3), if it were about self-driving cars, we could almost keep our hands off the wheel and still arrive safely at the destination.
Music
What impact this will have is difficult to estimate at this stage. For example, with Open AI you can already make music which is sometimes quite nice to use, although I wouldn't want to run it in the self-driving tesla just yet. But the time seems not far off when it will be possible.
For NBD Biblion, the library service that helps libraries buy books, the moment is already here. Even though humans are still watching: summarising and even assessing books to be purchased is now in the hands of an algorithm. Publishers provide PDFs of books, the AI scans them, summarises and gives a score on keywords. Objective and neutral, much faster than before and you can put 700 human reviewers on the street with an email.
Webinar
Last week there was a webinar on the development, partly to answer the question of why so suddenly all those reviewers were out on the street. The webinar itself cannot be watched back, but the organisation sent a Q&A report with answers to most questions.
It is quite comprehensive, but also a bit brief, especially when it comes to the answers to the questions asked by the participants. That helps with clarity, of course. Still, it very much speaks to the fact that NBD Biblion is mostly asking for trust from 'the field', and that all eventualities, such as an algorithm suddenly leading to bias, have been covered.
Gigabyte
When a participant asked if NBD Biblion could then provide examples of how the process worked, the answer was telling: 'After 10 years of work, we have about 130 gig of models and there are hundreds of different processes. A simple explanation does not do justice to reality. All the information used for the Al text comes from the book.'
That is a clear answer, which we will then rely on.
English (UK) 
Nederlands 
Français 
