Artificial Intelligence is considered a technological breakthrough and is becoming an inseparable part of our lives, constantly evolving. However, this development still requires a significant human labour contribution to ensure that AI processes function efficiently and yield high-quality results. This creates a problem: in the labour market, people working in data labelling are often exploited, working under heavy workloads and receiving minimal pay, despite the critical importance of their work.
AI is often associated with the automation of tasks, which reduces job opportunities and the need for human labour. However, at least for now, AI cannot exist without human oversight, as it still requires manual data labelling.
This is the process in which humans review content such as images, texts, or videos and mark the relevant data. For instance, a person might be tasked with identifying and labelling pictures containing a dog.
Through this process, humans contribute to AI models’ learning, which later enables these models to recognize patterns automatically. In other words, AI learns to generate results from the datasets created by human labour.
Data marking is a crucial foundation for machine learning and deep learning systems, including computer vision and natural language processing.
As mentioned, this process is primarily driven by human resources. Challenges arise, as research in regions such as South America, East Asia, and Africa has shown that workers are often paid below the minimum wage in their respective countries, even though the work hours and nature of the tasks demand higher compensation.
Although data labelling has a direct impact on the quality of machine learning algorithms, with accurate marking ensuring more precise predictions, it is also a costly, time-consuming process that is prone to human error. Moreover, it sparks significant debates about ethics and the exploitation of labour markets.
Some companies openly admit to using cheap labour. For instance, in Finland, one company implemented this cost-cutting strategy by employing prison workers to train a large language model. These workers answer questions about text fragments and label data, earning as little as €1.54 per hour.
Such practices call for transparency in corporate labour policies, prompting us to reflect on whether the benefits of AI justify this tolerated exploitation.
One potential solution to the labour transparency issue is the automation of data labelling. Several tools and solutions exist that simplify and expedite the process.
One such tool is SentiSight, which offers smart solutions to reduce human workloads by automating certain aspects of the marking process. For example, its smart annotation tools enable quick creation of point masks, while AI-assisted labelling allows for marking images based on the predictions of pre-trained models.
Although artificial intelligence demonstrates technological progress and drives innovation, it is important to reflect on the ethical implications of its development. In this situation, essentially, human work must be based on transparency and fair work practices.
Meanwhile, automation tools are becoming the future to facilitate the data labelling process.
Sources: Access Partnership, The Conversation, IBM, SentiSight