AI IN SUSTAINABILITY
What are the barriers and challenges in implementing predictive analysis sustainably?
CHRISTINA SHIM: I think there’ s a cultural component here. AI and sustainability can be used as real business drivers, and we are focused on maximising AI’ s value and minimising its costs. The challenge is that a lot of organisations tend to do things the way that they have been doing things. It’ s dependent on cultural management in terms of leadership and the way that it’ s being implemented based on use cases within the organisation.
If we can recognise that a lot of real value is coming from tackling these important problems in a more efficient and cost effective way with AI, then how do you go about doing that? It’ s about intentionally embedding it as part of your day-to-day business decision making and processes.
The other piece of this is reducing AI’ s costs and energy needs. When we’ re talking about Gen AI, trillion parameter models are not necessary for what the organisation may want to actually achieve. Think about the use case that you’ re trying to achieve.
Foundation models use the smallest possible model that’ s perhaps built on an existing foundation model so that you can amortise the training costs through further uses. As one example, our Gen AI model Granite has a choice of density, as well as the type of compute and energy cost, based on what’ s needed. In one case, a Granite model helped a global bank to do a project with 95 % less cost than an existing popular large model. They were able to do exactly what they needed to do regardless.
SUBHAGATA MUKHERJEE: Implementing predictive analysis sustainably presents several challenges.
With data quality and availability, high-quality and comprehensive data is essential for accurate predictive analysis. However, data silos and inconsistent data formats can impede AI model training and performance.
Training and deploying some AI models can be energy-intensive, potentially offsetting sustainability gains. It’ s crucial to assess and mitigate the environmental impact of AI operations.
Ensuring that AI systems are transparent, fair and respecting user privacy is vital. Addressing biases in data and algorithms is necessary to maintain trust and uphold ethical standards. sustainabilitymag. com 169