Responsible Generative AI
Sensors are also key to the sense-compute-actuate loop once the model training is complete . Several hazards exist between the physical process and the input to the AI inference engine . Sensors can be incorrectly connected to the physical system ( for example , a clogged port on a pressure sensor ). Networks can be congested or hacked . AI systems could come in handy to detect and correct many erroneous situations . Generative AI could create synthetic data to supplement or replace missing or untrustworthy sensor data .
6.3 INTELLIGENT DIGITAL TWINS
Digital twins use sophisticated simulations of real-world systems to predict their operations . The simulation testing of digital twins is costly and time-consuming and GenAI holds promise to make them affordable and timely for a variety of potential applications .
A blog by IBM 24 says that the use of GenAI increases the power of the digital twin by simulating any number of physically possible and simultaneously reasonable object states and feeding them into the networks of the digital twin . Some interesting use cases cited are :
• LLMs helping the identification of anomalies and damages on utility assets .
• LLMs based on time series data and its co-relationship with work orders , event prediction , health scores .
7 IMPACT ON ENVIRONMENT 7.1 ENERGY REQUIREMENTS OF GENAI
Deep Machine Learning has always been very time consuming and power hungry . While most of the power consumption happens during training ( building ) the model , the run-time ( inferencing ) is also expensive especially since there may be several users running the trained model concurrently on different applications .
GPT-3 was reported to have 175 billion parameters and required 355 years of single-processor computing time and consumed 284,000 kWh of energy to train 25 . Environmental impacts of the carbon footprint caused by LLMs are a major concern 26 .
GenAI will have a significant energy cost to run the very large computational and networking capabilities it will require for both its model training and inference phases . As GenAI snowballs in popularity , its global energy consumption will probably greatly exceed that of cryptocurrency
24
https :// www . ibm . com / blog / will-generative-ai-make-the-digital-twin-promise-real-in-the-energy-andutilities-industry /
25 https :// www . nnlabs . org / power-requirements-of-large-language-models /
26
https :// medium . com /@ sebastiaan . bollaart / the-environmental-cost-of-llms-a-call-for-efficiency- 206cbf352c79
Journal of Innovation 31