Tamblyn (Rakuten Kobo): 'Ai creates a new relationship between readers and publishers'
According to the CEO of one of the ebook market leaders, however, reading remains a profoundly human experience
by Antonino Caffo
Will the bookseller of the future be an avatar, perhaps a digital twin of the trusted shopkeeper? Possibly, in fact, it is already partly so. There is a question that anyone who loves to read has asked themselves at least once, standing in front of a bookshelf - physical or digital: what am I reading now? Until a few years ago, the answer came from the friend with similar tastes to ours, the trusted shop assistant or the review found by chance in a magazine. Today, that answer is being taught by a machine. Rakuten Kobo, one of the world's leading players in the ebook market, has a precise vision of where digital publishing is heading: towards a model in which artificial intelligence stops being a cataloguing tool and becomes something more ambitious, a tailor-made literary advisor, capable of interpreting not only what we have read but 'the way' we have done it, the type of experience we were looking for.
To understand how radical the change is, we need to start from where we were. The recommendation systems of the first generation were basically statistical correlation mechanisms: they aggregated purchasing behaviour and extracted patterns from it. They worked, in a way. But they reflected the past, not the reader. They rewarded already bestsellers, amplified fashions, ignored the peripheries of taste.
Kobo's ambition is a different one: "The paradigm of digital discovery is shifting from passive browsing to active and meaningful connection," says Michael Tamblyn, the CEO of Rakuten Kobo, "we want to move beyond chart-driven consumption towards a personalised ecosystem that honours the reader's unique journey". The change in vocabulary is eloquent in itself - it is about 'journey', not 'purchase'; about 'honouring', not 'satisfying'. AI, in this vision, does not optimise a transaction: it accompanies an experience.
"Where algorithms only see data points and preferences, we look for the cultural resonance and connections that make reading a deeply human experience," Tamblyn continues.
The problem is that language models are creators of averages, not new ideas. Trained on what exists, they tend by nature to project the past onto the future, to reinforce what is already known, to ignore the emerging author who does not yet have enough reading data in his favour. An advisor who only knows the classics and best-selling books is not so different from the first-generation algorithm.
