The advent of large, comprehensive AI models (such as OpenAI’s GPT-4) has had a profound impact on society, both beneficial and detrimental. The positive impact of AI includes stimulating innovation, generating audio, video, and text, and much more. However, the negative impact of AI includes environmental damage, data scraping that is mostly illegal, its immense costs, and it being the source of large amounts of misinformation, deep fakes, and so on.
There appears to be a tendency towards the adoption of smaller, more agile models that are specifically designed to efficiently perform a limited range of tasks. Such models are designed to perform these tasks at reduced cost in comparison to larger models. Such models offer advantages in terms of speed and can be deployed on local devices, thereby eliminating the need for constant cloud connectivity. This presents a substantial advantage for applications that require cost-effective, real-time processing and privacy.
Examples: Microsoft’s Phi models that, despite being 1/100th the size of GPT-4, can perform many tasks nearly as well. Or Apple’s initiative to run AI software entirely on phones using small models.
The reduction in data foorprint may prove to be a potential public relations asset.
The governance of information will be of significant importance in the utilization of these smaller models.
1. The functionality of small-scale AI models is contingent upon the quality of the data on which they rely. Consequently, these models require access to clean, accurate, and well-curated datasets to operate effectively. Information governance is a crucial aspect of this endeavour.
2. It is essential to guarantee that AI models adhere to regulatory standards and data privacy standards. Information governance frameworks assist organizations in navigating intricate regulatory terrains and implementing vital safeguards.
3. It is imperative that users and stakeholders have confidence in the impartiality, transparency, and fairness of AI models. The implementation of information governance practices, such as data lineage tracking and bias audits, facilitates the ethical design and use of those models.
What further arguments could be put forth to demonstrate the importance of information governance specialists within organizations?
Published in a post on LinkedIn in september 2024.
This post was a response on ‘Why Information Governance is Critical to the Success of Smaller, Nimbler AI Models’, published here.
