Hamilton resident Tina Lynn Wilson has been working as a freelancer for DataAnnotation since the beginning of this year.
Wilson, 45, expresses her enjoyment for the tasks she performs, which mainly involve reviewing output from an artificial intelligence model to ensure correctness in grammar, accuracy, and creativity. Her responsibilities demand analytical skills, attention to detail, and offer diverse projects, such as selecting the superior piece of poetry from two samples.
The nature of the work entails assessment of creative responses without the need for fact-checking. Wilson elaborates that the focus is on determining the better response and providing reasoning behind the choice.
This type of work constitutes a significant segment within the global gig economy related to emerging artificial intelligence technologies. Firms like Outlier AI and Handshake AI engage these workers as “artificial intelligence trainers” to train their AI models.
While some data annotation tasks are undervalued and potentially exploitative in certain regions, the scope of jobs in training, maintaining, and refining AI systems is extensive. Large tech corporations often overlook this labor force. As AI models evolve, the need for specialized training increases, potentially reducing the reliance on human annotators who have contributed to their current capabilities.
Companies are using AI hiring bots to screen, shortlist and talk to job candidates. Advocates say the technology frees up human workers from tedious tasks, but some applicants say it adds confusion to the process, and there are concerns about HR job losses.
Human Expertise
Generative AI systems are commonly trained on extensive datasets to understand the typical connections between human concepts. Although this initial pre-training is crucial, further refinement is necessary for these systems to generate accurate, relevant, and non-offensive responses, especially in specialized real-world applications.
This refinement process, known as fine-tuning, heavily relies on human expertise. It essentially involves gig work, where individuals work on a task-by-task basis without fixed hours. Canadian AI trainers, like Wilson, report earning around $20 per hour for general tasks, while specialized assignments can offer up to $40 per hour. However, income stability can be an issue.
Wilson emphasizes that relying solely on this work for income is challenging, considering the sporadic nature of assignments. Many annotators, including Wilson, view this work as a supplementary source of income.
Reinforcement learning from human feedback is a form of fine-tuning that heavily relies on human assessment of AI outputs.
Wilson’s role involves evaluating the human-like quality