Deterministic models are built on predefined rules and conditional logic. They function a lot like decision trees; they can only offer outcomes that have been explicity programmed.
When things can only respond to predefined outcomes, it’s impossible to account for every possible nuance or creative outcome.
Put another way, deterministic models are great at answering “if X, then Y” but struggle to go beyond those boundaries.
GenAI and multimodal systems operate fundamentally differently. Instead of relying on static rules, they leverage vast datasets, neural networks, and dynamic learning capabilities to generate outputs tailored to individuals in real time.
These systems don’t just categorise users based on past behaviour—they try to “predict” what you might want to engage with next, often before you even realise it yourself.
The Spotify Example
The personalised experience we have grown to love at Spotify would never be possible with deterministic models alone.
Deterministic models help to initially group listeners into broad categories like “loves country music” or “listens to classical”. While it’s a start, it doesn’t capture the complexity of your listening habits or get the product to a stage where it’s able to introduce you to music which hits just right.
GenAI doesn’t just see you as a genre lover; it sees you as someone who wants calm music for focus in the morning, high-energy tracks for workouts in the afternoon, and soothing melodies to unwind in the evening.
Notice that Spotify still makes use of deterministic models to help understand a listeners preferences and behaviour at a high level. Similar to carving out great chunks of a statue to start, then carving away the smaller details to get the representation exactly right, deterministic models help to carve great chunks of personalisation and then is best followed by GenAI to get the representation exactly right.
The Learning Example
Traditional education often rely on grouping learners into broad categories.
We often see this at school, with grouped classes based on “advanced” or “basic” streams. Singapore is especially well known for implementing this at a very young age.
This deterministic approach is a start, but makes the mistake of assuming all learners in a group benefit equally from the same resources and teaching methods. Just because I also like country music, that doesn’t mean me and Dolly Parton have the same exact taste in music or like listening to the same music at the gym.
GenAI and multimodal systems adapt to individual preferences, customising content based on type, tone, formatting, examples, and difficulty.
Even more importantly, it allows us to anticipate what a learner needs next.
A student struggling with calculus might have gaps in algebra they’re unaware of. Unlike deterministic models, GenAI can identify these gaps and dynamically provides tailored resources to help that particular student.
Understanding Context is Crucial
At its core, personalisation is about making technology feel human. It’s about making people seen for being them, rather than boxed into broad groups. There’s a fine line between wanting the security of feeling part of a group, and wanting to be treated as an individual. GenAI achieves this by creating experiences that resonate on a deeply personal level.
By understanding context, emotions, and preferences, these products can create an experience that really feels good to use - something that you feel is helping you personally rather than putting you in a box.
