Generative AI is rapidly transforming the landscape of higher education, influencing how we teach, learn, and conduct research. As these tools become more integrated into academic environments, it is essential for faculty, administrators, and students alike to consider the ethical implications. Key concerns include equitable access, bias, copyright/intellectual property rights, privacy, and standards of research and scholarship.
Although many major AI platforms offer a free level of access, most provide significantly more powerful tools to paid subscribers. Such pay-to-play models create barriers that exclude users who are unable to pay for a subscription or who do not have access through a job or school.
All information is biased and it is unclear how (and if) generative AI reconciles these biases. We do know that AI platforms are trained on massive datasets many of which are openly available on the web. During training, human feedback is used to curate the data to remove harmful information, but human reviewers also carry inherent bias that can and will effect the training process.
The majority of platforms are built on the English language and programmed from a Western perspective. Additional social, culture, or political biases may also impact the output of AI platforms.
One of the major areas of concern for AI is its use of copyrighted data during the training process. Although most platforms are trained on the open web, many copyrighted materials have illegally found their way past paywalls and onto open websites. Due to the massive scale of data ingestion during training, copyrighted items are included in the training data set and ultimately incorporated in the AI without permission. Human intervention is intended to filter this material, but he scale is far too large to intercept all material.
In addition to lawsuits by copyright holders, there is a struggle to understand authorship and eligibility for copyright protection for AI-generated content. Regardless of their intent (or lack thereof) AI users may also be held in violation of copyright based on their use of AI outputs.
Since ChatGPT was released, many people have raised concerns about privacy. They worry about how these AI tools collect and use personal data to improve their responses. This raises issues like increased tracking by companies, potential harm to targeted groups, and the use of personal information without clear permission.
Congressional Research Service - May 23, 2023.
Stanford HAI March 18, 2024
Plagiarism and copyright infringement is notable concern in the use of AI, but concerns range beyond these issues. While many platforms are introducing research assistants, most do not provide a level of transparency that meets the rigors of scholarship. For example, overreliance on AI for literature reviews or systematic reviews can be problematic depending on the level of access the AI has to relevant content. Most current research is published behind paywalls where AI cannot (legally) reach. Without informing the user of its limitations, how will researchers identify gaps?
The promise and perils of using AI for research and writing
APA - October 1, 2024
Because we have to pay our power bill and fill our gas tanks, people will often consider efficiency and consumption when it comes to appliances and automobiles. What about free technology easily accessible on the internet? Few people give consideration to the requirements for building and operating an AI model.
Not only is there an environmental trade-off with the use of AI, there is also a trade-off in human labor. Training an AI takes thousands of hours of human oversight and intervention to focus and refine the tool. The more human hours invested, the better. Unfortunately, many AI models exported this work to data centers around the world, where workers were dramatically undercompensated for their work.
Explained: Generative AI’s environmental impact
MIT News - January 17, 2025
According to an article by MIT News:
Training OpenAI’s GPT-3:
▸ 1,287 megawatt hours (MWh) of electricity (≈ power for 120 U.S. homes for a year)
▸ Generated 552 tons of CO₂
Data center power usage (North America):
▸ Grew from 2,688 MW (2022) to 5,341 MW (2023)
Global data center consumption (2022):
▸ 460 terawatt hours (TWh) – more than Saudi Arabia (371 TWh), nearly France (463 TWh)
Projected global data center usage by 2026:
▸ 1,050 TWh – would place 5th globally, between Japan and Russia
ChatGPT electricity use:
▸ 5x more electricity per query than a traditional web search
Cooling requirement:
▸ About 2 liters of water per kWh consumed by data centers
▸ Poses risks to municipal supplies and ecosystems
GPU shipments to data centers:
▸ 3.85 million in 2023, up from 2.67 million in 2022
▸ Increase expected to be even higher in 2024
GPUs have higher carbon footprints than CPUs due to more complex manufacturing and raw material processing.
Generative AI workloads consume 7–8x more power than traditional computing tasks.
Frequent model updates lead to short shelf-lives, exacerbating energy waste and environmental strain.
Time Magazine exclusive report - January 18, 2023.