Page 428 - AI for Good Innovate for Impact

AI for Good Innovate for Impact

• The proposed use case promotes transparency by helping users understand the
legal terms they agree to, ensuring that companies are held accountable for unfair or
exploitative clauses. By flagging problematic clauses and providing clear summaries,
users are empowered to make informed decisions and seek legal recourse if necessary.
The proposed use case also aligns with the goal of building strong institutions by
advocating for fair and ethical business practices.

2�3 Future Work

Given more resources, more work can be done in terms of making a much more comprehensive
data set for fine tuning the AI model. More work will be put in to collecting more data for the
data set. This includes ToS and privacy policy documents of more digital services, and acquiring
professional help in summarizing such documents. The more comprehensive data set will then
be used to further fine tune the AI model in order to make it more accurate and provide more
trustworthy summaries to users.
Additionally, more work can be done in making the resulting mobile application even more
helpful and educational to users. Features such as comparing similar applications so as to
recommend the one with more favourable ToS, providing users with more educative information
on clauses, and helping users know what to look out for before agreeing to terms, will be added
in order to achieve this.

3 Use Case Requirements
• REQ-01: Create a comprehensive data set with ToS and privacy policy documents of
various digital services, their ratings, their summary, and their good and bad clauses. This
can be reviewed and standardized with the help of experts.
• REQ-02: Fine tune a pre trained legal LLM in NLP and ML using the created data set to
be able to read and summarise ToS documents and identify clauses
• REQ-03: Validate the output of the AI model
• REQ-04: Create a mobile application into which the model can be integrated and allow
users to interact with it

4 Sequence Diagram

392