What are the stages of life for AI models?
It starts from the start: This involves defining the types of problems that the AI model solves and setting relevant goals.
After that, design and development will take place: At this stage, the author must build the algorithm, prepare the dataset, and build the model in the form of a cluster of computing functions. All this requires electricity (a huge amount) and water to ensure that energy-intensive data centers do not overheat.
verification: A wide range of test areas are run in Clusters to assess whether AI is working as intended. These tests require enormous amounts of electricity and water storage.
Expanding: Trained models are stored and replicated across servers spread across the world, balancing the load and making them work smoother and faster.
Operation and Monitoring: Once live, the AI model handles all requests through a powerful server.
Re-validation: As more data is collected and social context and user needs evolve, models are periodically retested and fine-tuned.
retirement: This happens when AI systems become obsolete. That model and server will be deprecated, discarded or replaced. This includes throwing out large quantities of chips, circuits and hardware that will degrade and poison soil or water for centuries, unless it is disposed of safely and responsibly.
Continuous learning: Throughout the lifecycle, AI models adapt by “learning” from new data and feedback, requiring additional calculations and storage, all at energy costs.
Can AI use anything other than water?
In the first study published in the Journal Communications of the ACM in June, researchers from the University of Houston at California Riverside, Texas and the University of Houston tracked how AI uses water and whether it is possible to use it smarter, more sustainable.
First, how to use it. The data center uses power plants that use large cooling towers.
Secondly, there are servers in your data center (thousands of servers and thousands of centres for each such center) and you need to keep it calm even if the semiconductor overheats during use. These are usually also connected to cooling towers.
If climatic conditions are appropriate, data centers can also use ambient air outside the facility to directly dispel heat, research notes. However, water will still need to be evaporated, especially if the outside air is hot. Water is also needed to achieve the ideal humidity level for the machine when the outside air is too dry.
While the use of ambient air is less resource intensive than water, it is difficult to practice this in busy areas with very hot areas and multiple data centers. Already, healthy pollution from large data center drones has led residents to protest in parts of the United States.
The best answer remains a smaller model, with more consideration being given to the use of this technology. Click here for more information.
