Pratical guidelines for working with ai and data

The EBU is the world's foremost alliance of public service media organizations. Through an international project they have defined some practical guidelines for working with ai and data that is generic for all kind of operations.

Datasets and databases:

• Appropriate, unbiased and substantial datasets are crucial, especially when catering to minority communities and aiming to increase accessibility

• Automated approaches using large databases can help address diversity and equality related challenges by providing broad insights into psm programming

• ML and large samples can help save time and money, and importantly avoid bias generated by smaller or shorter samples

• Remember: categories are imperfect

• Datasets used in ml training can sometimes match in format without matching in content

Infrastructure and organization

• Having the infrastructure to easily access your own content, and making sure data is documented and accessible throughout the organization, is key to unlocking its potential

• New skills are needed, whether via training or hiring

• Constant attention to cooperation beyond team and system boundaries must be managed to prevent solutions never being integrate

• Collaborate with public institutions and private bodies to bolster excellence

• Maintain a direct link with the users

Work process & quality

• Detours and intermediate steps are sometimes necessary, such as translating into an intermediate format before the final one

• Ai combined with human effort increases quality

• “the algorithm is never going to be better than the data it learned from”

• Open-source tools can help jump-start in-house development

• Always ensure interoperability between systems

• The process from prototype to production can be an iterative one

• Quality needs to be continuously evaluated and validated

• Prototypes benefits from agile development and community sharing

Ethics and legalities

• Beware of the legal requirements surrounding the use of personal data, and opt for respecting personal integrity

• Clearly defined high-level principles for ai help guide applications and thereby ensure accountable use by psm

• When considering in-house development vs. Buying third-party products, it is necessary to balance cost and complexity with independence and the meeting of specific needs

• Having checklists eases compliance when working with data and ai

• When it comes to ethics, lean towards guidance over governance

• ML and AI pose specific challenges to ethical guidelines for psm: take responsibility for what is designed in-house and what is bought in. Understand how it works and whether user data is used on third-party platforms without prejudice to data protection regulation

• Ai and data can be used to fight (and fuel) disinformation