I’m meeting a lot of CEO’s and their execs these days … helping them prepare their AI strategy for their BOD (Board of Director) presentation and/or validating what they have as a sound roadmap and feasible to deploy.
And I have to give credit to them … they have come up to speed with what AI is and what AI is not fairly quickly.
The part that still concerns me after having been a CDO many times over and having seen the data ecosystem for both small and large organization's is this … DATA is still the ruler of the AI kingdom. Whether developing Applied AI or Generative AI applications … the basics of a data ecosystems are still impeding the progress of true development. So outside of reading a Gartner or McKinsey report on the importance of data in any AI roadmap, it still hasn’t truly sunk in yet.
For example … in the task & approach of either fine tuning a model vs. taking a RAG approach
- Fine-tuning involves retraining an existing AI model on a new dataset ... so by default you need large amounts of training data that needs to be accessible, available, and of good hygiene.
- RAG involves connecting a retrieval system with a generative model … so taking information from a well known knowledge base ( that also needs to be accessible & of good hygiene) and using the generative model for search, understanding, and text creation.
Either approach has a dependency on accessible & good data … something we CDOs continuously need to persuade our stakeholders and peers for year after year during our financial cycles of “I need X to fix our Y” … yet we never truly get the investments we need, so the can gets kicked down the road.
My only request is, don’t develop an AI Strategy without parallel pathing your Data Strategy. The 2 need to sync!
And I have to give credit to them … they have come up to speed with what AI is and what AI is not fairly quickly.
The part that still concerns me after having been a CDO many times over and having seen the data ecosystem for both small and large organization's is this … DATA is still the ruler of the AI kingdom. Whether developing Applied AI or Generative AI applications … the basics of a data ecosystems are still impeding the progress of true development. So outside of reading a Gartner or McKinsey report on the importance of data in any AI roadmap, it still hasn’t truly sunk in yet.
For example … in the task & approach of either fine tuning a model vs. taking a RAG approach
- Fine-tuning involves retraining an existing AI model on a new dataset ... so by default you need large amounts of training data that needs to be accessible, available, and of good hygiene.
- RAG involves connecting a retrieval system with a generative model … so taking information from a well known knowledge base ( that also needs to be accessible & of good hygiene) and using the generative model for search, understanding, and text creation.
Either approach has a dependency on accessible & good data … something we CDOs continuously need to persuade our stakeholders and peers for year after year during our financial cycles of “I need X to fix our Y” … yet we never truly get the investments we need, so the can gets kicked down the road.
My only request is, don’t develop an AI Strategy without parallel pathing your Data Strategy. The 2 need to sync!