Everything is 'Ask AI'

Original article (in Vietnamese): https://vnexpress.net/viec-gi-cung-hoi-ai-4852117.html

My colleagues and I suddenly have more work when grading students' assignments, because we have to determine whether they are using AI tools excessively.

The majority abuse AI in many ways, from simple things like using voice reading and drafting content to more difficult tasks such as presenting slides, building simulation videos, etc.

Applying AI to learning and work is inevitable, but abuse is a different story. This article will mention the unexpected impacts of AI on the environment if billions of people around the world continue to use AI irresponsibly, consuming energy for huge server networks.

Not counting the environmental costs of chip production and supply chains, the training process for a single AI model, such as a Large Language Model (LLM), can consume thousands of megawatt-hours of electricity and emit hundreds of tons of carbon—equivalent to the annual carbon emissions of hundreds of households in the US, according to a report by the University of California, Riverside, and the California Institute of Technology (Caltech).

Furthermore, the AI model training process can lead to the evaporation of a huge amount of fresh water into the atmosphere to cool servers and data centers, according to a recent report by UNEP - the United Nations Environment Programme.

More worryingly, global AI energy demand is expected to increase exponentially, to at least 10 times the current level, and exceed the annual electricity consumption of small-scale developed countries like Belgium by 2026. In the US, rapidly increasing AI demand will push data center energy consumption to about 6% of the country's total electricity usage by 2026, putting additional pressure on grid infrastructure.

In an era where environmental awareness is increasingly heightened, initiatives to promote AI sustainability and apply AI in mitigating climate change are gaining attention. "Sustainable AI - Responsible AI" is one of the recently noted initiatives. Sustainable AI is based on criteria related to the impact of AI on social life, aiming to adjust to fit community values through legal and ethical standards.

However, the sustainable AI indicators currently being built often prioritize easily measurable environmental data such as total carbon emissions and water consumption, without ensuring Environmental Equity. Harms and impacts need to be based on fairness in protecting individuals, groups, or communities from environmental dangers regardless of ethnic origin, race, or economic status.

Based on Google's 2023 environmental report, Google's data center in Finland operates with 97% carbon-free energy, while this figure at centers in Asia only fluctuates between 4-18%. The significant disparity in carbon fuel consumption and emission levels in each country and region entails differences in the level of impact and air pollution. Similarly, the rate of cooling water consumption also varies greatly between regions even within a country like the US; typically, Arizona, with its hot, dry climate and water scarcity, will require a much larger amount of cooling water than a center in Nebraska.

Vietnam, with developing infrastructure and a hot climate year-round, is very vulnerable. To prevent risks from artificial intelligence, regulations and institutions in AI research, development, and management play a very important role.

First, AI systems must ensure safety and security to limit immediate negative impacts and facilitate risk control. Vietnam has issued a National Strategy on Artificial Intelligence development, with a vision to 2030. The next step is to focus on prioritizing the construction of a set of principles on AI responsibility.

Second, the Ministry of Information and Communications has early on drafted the Law on Digital Technology Industry. This is a very important legal corridor to promote the digital technology industry with a focus on Vietnamese enterprises, aiming to shift from assembly and processing value chains... to creation, design, and mastery of more advanced technologies. Laws on Artificial Intelligence development need to be integrated directly into the Law on Digital Technology Industry, avoiding duplication and overlapping of regulations, creating a burden for businesses.

Finally, in many plans to minimize negative impacts on the environment, the easiest thing to do is always to reduce generation/use at the source (for waste, or food). Besides building sets of standards on AI responsibility, as well as managing and applying AI projects to environmental protection, it is equally important to educate and propagate about the negative impacts of AI abuse.

In the final lecture of the recent semester, a few students asked me: "Why shouldn't we depend on artificial intelligence?". My answer: AI, after all, is still a machine. If we are music lovers, we surely know the difference between sound from continuous signals (Analog) and digital signals (Digital). A machine cannot completely reproduce perfect sound like a continuous signal, because signals encoded in binary form as 0 and 1 cannot describe the variations like a continuous signal. The same thing happens with content produced by AI, as they cannot reproduce original and continuous intelligence formed in human thought.

Although artificial intelligence changes and will be optimized quickly, maintaining and encouraging the younger generation to use original intelligence and knowledge is the responsibility regarding AI of the previous generation.