The Finnish tech company Metroc has recently adopted an unconventional approach to train its language model for artificial intelligence (AI) technology. Instead of outsourcing data labelling work to low-wage labour markets in the Global South, Metroc has enlisted the help […]
The Finnish tech company Metroc has recently adopted an unconventional approach to train its language model for artificial intelligence (AI) technology. Instead of outsourcing data labelling work to low-wage labour markets in the Global South, Metroc has enlisted the help of prisoners through a prison labour program. In this program, prisoners are paid 1.54 euros per hour to answer simple questions about snippets of text, contributing to the data labelling process.
Data labelling is an essential part of AI development, where human workers label data sets to help AI models learn and generate accurate outputs. Typically, companies outsource this work to countries where workers are fluent in English and willing to work for low wages. However, due to the lack of Finnish speakers in these countries, Metroc has found a local source of cheap labour in prisons. Without this program, it would be challenging for Metroc to find Finns willing to take data-labelling jobs that pay significantly lower than the average salary in Finland.
While this cost-cutting strategy highlights the significant human labour needed to fine-tune AI, it also raises important ethical questions about the long-term sustainability of such practices. The use of prison labour in AI development highlights the larger story of the human cost behind AI’s rapid growth in recent years.
One prominent issue that arises from this model is the exploitation of low-wage labour. The reliance on outsourced and low-wage workers for tasks like data labelling is not exclusive to Metroc. Many leading AI firms use similar strategies. However, these practices often go unnoticed due to the excitement and attention surrounding AI technologies.
The implications of these labour practices are twofold. Firstly, the significant amount of human labour required to shape AI tools should make users more cautious when evaluating the outputs of these tools. Secondly, until AI firms take steps to address exploitative labour practices, users and institutions may need to reconsider the supposed benefits of AI tools.
Q: What is data labelling?
A: Data labelling is a process in AI development where human workers label data sets to help AI models learn and generate accurate outputs.
Q: Who are data labellers?
A: Data labellers are workers who perform tasks such as confirming image features or flagging offensive language as part of the data labelling process in AI development.
Q: What are the ethical implications of prison labour in AI development?
A: The use of prison labour in AI development raises concerns about the exploitation of cheap labour and highlights the larger issue of the human cost behind AI’s growth.