2/

4. building (or renting) adequate ML infrastructure
5. scaling to meet ML requirements
6. enabling smooth deployment of ML projects

Unfortunately, sometimes the proposed solutions highlight the terrible mindset of today’s approach to IT. An example (related to problem 1):

1/ Reading a very practical ebook from @OReillyMedia “Machine Learning at Enterprise Scale”. It describes 6 problems with modern #AI development:

1. reconciling disparate interfaces
2. resolving environment dependencies
3. ensuring close collaboration among all ML stakeholders

There are a lot of great recommendations in this document. I quoted just a few here.
My favourites, as usual, are related to the data economy necessary to fuel a fair #AI. Worth reading in full. https://t.co/nY3nyGUCWu

Quote 12: Create a data donor scheme, allowing individuals to donate data for specific purposes, surrounded by clear governance and transparency, and safeguards for privacy protection.

Quote 11 (continues): competition rules, particularly in terms of membership requirements, type of data shared, and access conditions for third parties. It should determine how intellectual property frameworks impact the allocation of property rights within trusted data spaces

Quote 11 (continues): of the value is being created.
Industrial policy can help specify sector specific openness and non-discrimination requirements for access to trusted data spaces, and clarify the conditions under which inter-firm data-sharing cooperation should respect

Quote 11: Foster the creation of trusted data spaces for specific sectors (e.g. healthcare, automotive and agri-food) where actors in a given value chain accept to share data, yet the rights related to the data rest with end users and/or the stages of the value chain where most

Quote 10 (continues): diverse datasets in the public domain. An EU system for annotating datasets and standardisation should be specifically supported in high impact domains with particular domain-specific content (such as employment, manufacturing or health).

Quote 10 (continues): The development of domain-specific data representation should be considered, as well as standards specific to particular domains. Standardised data and machine learning parameter schemes can help ensure a common data format that allows combining

Quote 10: Support an EU-wide data repository through common annotation and standardisation. This requires investment in the development of methods and tools that render datasets comparable and useable across EU borders.