Could AI benefit Low-Skilled Workers?

Displacement of labour as a result of automation has always been a controversial topic with some studies finding 48% of American jobs at high risk by 2035. But last year the World Economic Forum released a report identifying that AI is expected to displace 75 million but create 135 million new roles. This means that there is a net gain of 60 million jobs. The popular consensus has been that a very high majority of the new roles will be high skilled meaning they are not suitable for the low skill workers that AI generally displaces. Therefore meaning that AI will allow low-skilled jobs to disappear in favour of more knowledgable and skilled workers.

However, research suggests that a silent army will be needed to build AI. As most AI algorithms need to be trained on a voluminous dataset which will require supervised learning. This is because AI will eventually be able to make decisions itself but only after been exposed to enough situations with their respective solutions. For example, we can teach a neural network to identify pictures of a house buy feeding it with thousands of images of houses and specifying to the AI that this is a house each time.

This training data needs to be lead by a human who would essentially use labelling techniques to identify that the image contains e.g. a house. The extensive need for a human in training data is best highlighted by self-driving cars. For example, self-driving vehicles generate 60 terabytes of data for every 12 hours of driving which would initially require considerable human data training.

Basic Model representing the role of data annotators, source: Dataturks

Basic Model representing the role of data annotators, source: Dataturks

Generally, these data annotators are outsourced and located in low cost of labour regions such as India or many African regions with very big teams consisting of thousands of people. Data annotation centres can also be seen as the assembly lines in the technological age. As they are the cognitive equivalents of the traditional assembly line with workers engaged in repetitive cognitive efforts rather than physically manufacturing and demanding tasks. But in comparison to assembly line workers, data annotators are higher-skilled as the role requires training and meticulous attention to detail, for example, drawing precise shapes and images or pinpointing landmarks. All with extreme accuracy and precision as in some industries such as automated cars, poor inaccurate training data could lead to serious collisions. However overall data annotation provided a great opportunity for low-skilled workers particularly those in developing countries. For example, Deepen AI has started a venture called which focuses on giving refugees access to jobs and skills to improve their economic situation for example training Syrian refugees to become data annotators.

To conclude there may be many opportunities for low-skilled workers in the future as a result of AI. Especially as it is likely that the need for data (know as the new oil) and annotated datasets will grow exponentially in the next few years. However, to maximise profit margins, some AI companies may keep salaries for data annotators extremely low with some studies suggesting below $2 per hour. Therefore the nature and quality of these jobs are unknown and international regulation could be introduced to combat what is effectively slavery in the modern era.

FinanceArun Singh Dhillon