Dynamic. Interconnected. Expansive.
Algorithms are assemblages of code and data furtively woven in our everyday surroundings, yet somewhat withdrawn to our eyes.
We encourage you to enter the playground.
Rethink the conglomerate of intricate systems.
Be mindful of their underlying bias, and their ethically-problematic roots, fueled by profit-driven, institutionalized power and exclusive values.
Embrace an attitude of compassion, vulnerability and care towards yourself. Towards your community. Towards the overlooked and forgotten.
Imagine alternative futures of inclusive technology in the context of recommender systems.
Soft, vulnerable, porous, empathetic, inclusive, fungal, grassroot, rebellious, situated and caring, how would YOU like your recommender system?
Curious to dig more about...
.why this experiment
.how this prototype works or whats behind these choices
.how to download your Google Historics
.designing YOUR own prototype.
We propose you to venture into prototyping your own recommender system.
As explained here, our prototype follows three steps of (1) Extraction >> (2) Profiling >> (3) Recommendation.
Both steps (2) and (3) relies on ChatGPT, which is a instruction-based language model.
It means ChatGPT can follow instructions you provide --commonly referred to as 'prompts'--, and generate text based on them.
A prompt is composed a sentence, paragraph, or input data, which should guide the AI model to generate relevant responses.
Each of the recommender in our prototype is based on a specific prompt (i.e. instructions on what to do) sent to ChatGPT.
More details about the prompts we use below.
To create your own recommender prototype, you need to design your own prompt for ChatGPT.
A few tips about prompt-design.
A detailed prompt may greatly affect the quality of the generated text, and avoid misunderstandings.
It can include instruction about the content, tone, or style you want ChatGPT to follow.
You do not need to use Filler Words in ChatGPT Prompts, such as "could you please", you can go straight to the point.
To guide ChatGPT further, you can also provide examples of expected answers.
Too generic, or abstract prompts ("what is life") may result in dull answers, while contextual information helps getting more tailored answers.
Keep in mind that recommendations Step happen after the Profiling step in which ChatGPT is asked to imagine your digital double, referred to as XXX here.
So when designing your prompt for the recommender, ChatGPT will have already some sketch of your digital double.
Lastly, if you want ChatGPT to tailor its answer to your data, you have to refer to your digital double as XXX in the prompt.
To give you an idea, here an example of the prompt we used for the 'vulnerable' recommender:
"Generate 7 unique suggestions for your character XXX to acknowledge, share and exhibit vulnerability towards events occurring on Earth or their personal life.
Vulnerability could relate to both animate and inanimate objects, sounds, gestures, processes, or events.
Maintain a creative, empathetic, quirky, and unconventional tone.
Each recommendation should be a single sentence, referencing a specific event, object or process in the novel and avoiding generic or romantic stereotypes.
Be original, quirky, radical and weird in your suggestions.
Examples: 'Make some tears at the sounds of a crumbled plastic bag.', 'Cry a little when feeling an irregularity in the concrete pavement.', 'Feel vulnerable when somebody mention the word ‘hard drive’'.