Responsible Generative AI
6 IMPACT ON IOT SCENARIOS
6.1 BACKGROUND
The Internet of Things ( IoT ) can greatly benefit from the careful application of AI techniques . In IoT systems , various types of sensors measure parameters in the physical world . These readings are digitized and transported by networks to computational capabilities that may exist in implementations like Programmable Logic Controllers ( PLCs ), edge computers , or cloud resources . Then , the results of these computations are sent over digital networks to actuators that can influence the physical world . AI techniques come in at the compute stage , where the capabilities of AI enable very sophisticated analysis of the sensor data , and equally sophisticated calculation of the best commands to send back to the actuators .
Although Generative AI is usually associated with unstructured content like text or pictures , largescale IoT systems can often benefit from careful application of generative AI techniques in areas like automated code generation , optimization of complex system parameters through the creation and analysis of multiple “ what-if ” scenarios , and the generative creation of digital twins and simulation models .
One important aspect of IoT is that the actuators allow software to control the physical world . So , if that software is inaccurate , unsecure , hallucinating or compromised in some way , those actuators can do physical damage in the real world in a way that IT systems without actuators can ’ t – in worst-case scenarios injuring or killing people . This is a great concern in traditional IoT systems that don ’ t use AI , but an even greater concern if we must worry about the potential for bias and hallucination of AI models that drive the actuators .
Great care must be taken at all stages of AI-based IoT deployments to ensure the model data is correct and consistent , that the model building process is accurate , the models are distributed in trustworthy ways , and that the inference engines that perform computations based upon those models are correctly connected and tested . Careful choices must be made to determine which physical systems we can entrust to Generative AI . Taken to extremes , there is concern of what is called the “ Skynet Scenario ” 23 where a large-scale AI system controls enough weapons to threaten humanity .
6.2 SMART SENSORS
The data from IoT sensors is given twofold use in AI-based systems . First , this data represents a behavioral record of the physical systems the sensors are monitoring in the physical world . This data ( perhaps going back years ) is ideal for training many classes of AI models .
23 https :// medium . com / data-driven-fiction / the-skynet-moment-looms-with-chatgpt-5ec9714f2750 30
March 2024