[GREEN TECH] - A Double-Edged AI Sword for Green Mobile Technology
The Synopsis:
Researchers at the Delft University of Technology have researched the implications of artificial intelligence on mobile devices and have noted the double-edged sword: AI uses a significant amount of energy, while also being used to create energy-efficient solutions for devices. The methodology of the research included a research goal and question, research process, publication selection, snowballing, data extraction, and data analysis. Nevertheless, the researchers gathered 34 papers, categorized them into 13 main topics, and summarized their findings.1
Below are the conclusions the researchers have made:
Research of artificial intelligence in mobile devices has become more common since 2019.
Of the 34 papers, many of them proposed solutions to mobile energy consumption, while very few papers were observational.
The use of AI to make mobile software more energy-efficient is being studied twice as much as the energy consumption of AI-based mobile software.
Of the 13 topics, Offloading, Context Adaptation, and Federated Learning were 70% of the papers. However, Approximate Computing, Model Design, Recommenders, Energy Measurement, and Resource Management were not studied as much.
Industry Researchers, compared to academic researchers, are scarce in the mobile energy consumption domain.
Studies to provide solutions to energy consumption with AI are plenty, yet only a small portion of the studies make the solution-based tools available online.
66% of the literature focuses on the green-mobile technology at the system (a network of devices) level, instead of the device level.
The Analysis:
The researchers’ diligence into exploring artificial intelligence in mobile devices, even though it has only recently gained popularity, is laudable. I believe their research is vital because green technology benefits us and our environment. However, I have an issue with the research for a few reasons, such as the sample size of the papers, the research-group makeup and size, software-only research, and search-engine preference. A sample size of 34 is relatively small to a corpus of work pertaining to AI and green-mobile technology, although it obeys the sample size rule of thumb. Additionally, due to the makeup of the research group, whom are English speaking, it was limited to only English-written research, which introduces innate bias in the selection process. Besides, there may be papers in other languages that address green mobile energy and AI. Google Scholar was used as the research engine; though, publications that may be on another research engine, like ResearchGate, could have been missed. In a research article titled, “ResearchGate and Google Scholar: how much do they differ in publications, citations, and different metrics and why?,” Google Scholar has had a higher amount counts of research compared to ResearchGate.2 Lastly, hardware is complementary to software, since without the correct GPU, RAM, or CPU, AI software cannot run efficiently. Research on hardware that supports AI software would allow for a comprehensive understanding of AI energy consumption, instead of researching the AI software alone. Despite my concerns with the research, I do believe that the researchers have done great work in highlighting the deficiencies in research of green technology and AI.
The Terminology / Research Process:
Research Question - What are the characteristics of the state-of-the-art research regarding Artificial Intelligence in Green Mobile Software?
Federated Learning - The application of AI to train a model across several decentralised devices with their own local subset of data, without the necessity of sharing data to all the involved devices.
Offloading - Approach to delegate the execution of a resource-intensive task from a lower- powered device, often a mobile device, to a different, usually more powerful, device or service.
Context Adaptation - The goal of the primary studies involving the topic is to improve the energy efficiency of mobile software by readjusting its execution according to the context.
Research Process:
Gather the initial set of publications
Choose publications (Include or Exclude Criteria)
Snowballing - Address the limitations in the query search
Data Extraction - Authors review the resources and categorize
Data Analysis - Address the research question
Data-Extraction Fields:
Publication Year
Study Type - The type of study the paper is presenting: either a position on AI in Green Mobile Software, a solution to tackle an issue on the topic, or an observational study.
Category of AI Role - The role AI has regarding Green Mobile Computing. It can either be the use of AI for improving the energy efficiency of mobile computing, or the study of energy consumption of AI-based mobile systems.
Topic - The topic the primary study is focusing on. For instance, context adaptation, in which the mobile software execution is readjusted according to the context, to improve the energy efficiency.
Level of Study - It corresponds to the scale at which the mobile software is studied (either at the level of the device, or of the system).
Industry Involvement - The involvement of industry in the authoring of the study, which can be either exclusively academic, exclusively industrial or a mix.
Tool Provision - The availability of the tool(s) to handle AI in Green Mobile Computing presented in the study (if applicable).
The Endnotes:
1 Siemers, Wander, et al., “The Two Faces of AI in Green Mobile Computing: A Literature Review”, Cornell University arXiv, accessed Mar 28, 2024,
https://arxiv.org/pdf/2308.04436.pdf
2 Singh, Vivek Kumar, et al. “ResearchGate and Google Scholar: How Much Do They Differ in Publications, Citations and Different Metrics and Why?” Scientometrics, vol. 127, no. 3, 2022, pp. 1515–1542, accessed Mar 28, 2024,