Artificial Intelligence and Education: More Questions Than Answers [Updated]

Update: At the end of the article, we add useful readings published recently that support and expand on this publication.

[Originally published on 07/04/2023]

At the beginning of the 20th century, and shortly after its discovery, the chemical element radium – very expensive to extract – was included even in the wool of baby clothes. Thanks to advances in science and research we currently use it only in beneficial applications and stay away from its radioactive effects. Ever since tools like ChatGPT have become available online, the same that happened with the discovery of radium is now happening with Artificial Intelligence (AI): it is becoming a necessity everywhere.

The current AI hype seeks to normalize its use in our societies so those who produce it can make huge profits as fast as possible. Although hype is common practice, the ways in which AI impacts Education lead us to commit to work constructively and responsibly, with a local perspective and respecting our values.

In this blog post, we share some critical thoughts and ideas about the potential uses of AI in Education. In this related post, we share some critical thoughts and part of AI’s potential in Education.

❓ We Have More Questions Than Answers

When we are asked questions about the intersection of AI and Education, before giving our expert opinion, we ask the following questions:

• To what educational task do you seek to apply AI? Will the AI tool assist instructors with a job that takes time away from strengthening their relationship with their students? Will AI be used for a task that focuses on the teacher-student relationship? As we expanded on in this post, although it still needs research, AI could help with repetitive tasks that do not involve students. The teacher-student relationship is the domain of humans, not algorithms.
• What is the goal of using AI? Is it to save money? Will teachers see their working conditions improve, stay the same, or worsen? Who will pay, and who will be charged for the work the AI system will do? It is crucial to assess that the impact on the working conditions of educational roles is positive or at least neutral, but it should never be harmful.
• Does using AI improve the teaching-learning process? Was it measured with good experimental design (e.g., randomized, masked) and with the consent of all participants? Did other actors, free of conflict of interest, reproduce the AI tool improvements? These educational tools are usually presented commercially as products that are difficult to resist. However, the evidence supporting them tends to be weak. It is common to discover sometime later that their use was harmful and the promises of improvement unfulfilled.
• What is the performance of the proposed AI system? How is that performance measured? Which students was it designed for? AI systems often make systematic errors against people not considered during the AI tool design time and tend to discriminate against them.
• What precautions have the people who produce these AI tools taken to avoid automating and amplifying harmful biases in our societies? There is no silver bullet to this problem. Thoroughly documenting the limitations and issues of these tools is a helpful minimum and a common practice in other disciplines with high social impact (e.g., Pharmaceutics and Transportation). This is still under discussion and far from agreed upon in AI.
• Where in the world and under what working conditions are these tools created? Big tech companies headquartered in high-income countries develop tools such as ChatGPT with significant profit margins. Often these companies outsource tasks such as data cleaning and labeling to companies in low-to-middle-income countries. The latter employ people in precarious, often unhealthy, conditions and with wages well below international standards for the AI field. The situation for these workers is highly problematic because, without their work, there would be no AI hype in the first place.
• What data was used to train the AI tool? There are lawsuits underway due to alleged violations of copyright laws during the training process of some currently available AI tools.
• What are the AI tool’s environmental costs (e.g., power usage)? The environmental impact of training tools like ChatGPT is very high. Experts warn that, amidst the climate change we currently face, we must restrict AI’s energy usage.
• Is it necessary to change everything in Education? What methods work well without modification? Should changes in Education follow the same pace as the current development of AI? What happens if we wait until these tools achieve minimum quality standards (e.g., minimize discrmination, avoid copyright infringement) and are regulated before implementing them in Education? Ideally we should not allow interests outside our communities and regions to hastily introduce tech, while portraying the process as inevitable, only to worsen the prevailing inequality in our geographies.
• What consequences could the immediate adoption of AI tools have simply because they exist and were made available by a few companies whose knowledge is limited to the AI field? Keep in mind that these are technologies with severe performance issues. For example, everything ChatGPT prints on the screen is an invention, and it is often difficult to detect that its outputs lack factual support.
• What are the consequences of the massive use of these technologies in Education in the medium and long term? How do people of different ages relate to AI tools that mimic human language? We do not have specific answers to these questions because there has not been enough time to study them in depth.
• If the new generations learn with AI, who are they learning from? Who designed the AI tool? With what values? Who will benefit from its implementation? For free tools, are they giving us something for free to extract something much more valuable, such as our data or intelligence?

🥁 A Closing Declaration

As we teach at MetaDocencia, Education is a social process that does not magically improve overnight with the introduction of any technology. Moreover, there is no universal solution for every context. Indeed AI, like radium, can bring benefits to our lives. It still remains to further explore these benefits and how to implement them well in Education. Such change must be democratic, honoring humans and our timing, and not because a handful of business people with accelerated profit motives want to impose their needs and solutions on us.

It is always better to prioritize people over tech tools. New AI tools generate enormous wealth for a few people and greater inequality for the rest of humanity. This is discussed further in the Montevideo Declaration on Artificial Intelligence and its Impact in Latin America, where more than 400 experts, mainly from Latin America, began to warn about this in March of 2023. We invite you to sign the Declaration and, above all, continue thinking critically. We invite you to sign the Declaration and, above all, continue thinking critically and, when you see the AI hype applied to Education, to ask yourself the questions that matter.

📖 Further Readings

• Miao, F. and Holmes, W. Guidance for generative AI in education and research. Published by UNESCO. September, 2023.
• UNESCO: Governments must quickly regulate Generative AI in schools. September, 2023.
• Higher education’s essential role in preparing humanity for the artificial intelligence revolution. A statement to be issued by a diverse group of educators and scholars at the 18th annual United Nations lnternet Governance Forum, October 2023, Kyoto, Japan.

Did you like this post? You can reuse it freely under CC by 4.0 license. Just cite it!

Here is the citation we recommend you use:

Laura Ación, Luciana Benotti, Melissa Black, Laura Ascenzi, & Paola Andrea Lefer. (2023). Artificial Intelligence and Education: More Questions Than Answers. Zenodo. https://doi.org/10.5281/zenodo.8120537.