Global Economic Development

Foundational Challenges in Assuring Alignment and Safety of Large Language Models

Anwar et al. (2024)

Source DOI

Sub-category

Risk Domain

6Socioeconomic & Environmental

6.2Increased inequality and decline in employment quality

Social and economic inequalities caused by widespread use of AI, such as by automating jobs, reducing the quality of employment, or producing exploitative dependencies between workers and their employers.

"Many of the themes and challenges that we discussed above come together when analyzing the socioeconomic effects on developing countries. The workforce of developing countries may suffer from a retrenchment of outsourcing as many simple cognitive tasks that used to be performed in developing countries — for example, in call centers –— can be automated with LLMs. This may adversely affect the economies of the poor countries (Georgieva, 2024)."(p. 95)

Entity— Who or what caused the harm

Human

Due to a decision or action made by humans

AI system

Due to a decision or action made by an AI system

Other

Due to some other reason or is ambiguous

Intent— Whether the harm was intentional or accidental

Intentional

Due to an expected outcome from pursuing a goal

Unintentional

Due to an unexpected outcome from pursuing a goal

Other

Without clearly specifying the intentionality

Timing— Whether the risk is pre- or post-deployment

Pre-deployment

Occurring before the AI is deployed

Post-deployment

Occurring after the AI model has been trained and deployed

Other

Without a clearly specified time of occurrence

Supporting Evidence (1)

"Additionally, in contrast to most other technologies, global adoption of LLMs requires that they are proficient in all world languages. However, even multilingual models perform worse in lower-resource languages (Etxaniz et al., 2023; Shen et al., 2024) and “think” in English even when fine-tuned for other languages (Wendler et al., 2024). This has severe safety and security implications as models that might be aligned in English might not be equally well-aligned in other languages (Deng et al., 2023a; Yong et al., 2023) or may perform worse in other languages (Aroyo et al., 2024; Holtermann et al., 2024). Additionally, LLMs may cost an order of magnitude more to use in some languages (up to 15 times in ChatGPT’s case; Ahia et al., 2023; Petrov et al., 2023b). Hence, ensuring that LLMs work well regardless of the language in which they are used is key if we want to ensure they are a tool for reducing global inequalities rather than further exacerbating them."(p. 95)

Part of Vulnerability to Poisoning and Backdoors