Enabling a new 'gig' frontier in Bangladesh and the globe
Offshore image annotation services for computer vision-based machine learning models
Let's be honest, COVID 19 is a hard slap on the face for the employed and the unemployed. Coming from Bangladesh, a country which recently attained a middle-income status, we observe a sizable advent of the unemployed as the job market faces imminent collapse in the wake of COVID 19. While the country’s socioeconomic status manifested in large part for the role of private sector enterprises and the impact it created in the economy, the pandemic and ensuing consequences of it are testing Bangladesh’s hard earned development over the decade; not only for the ‘talked-about’ ready-made garments (RMG) sector downpour, but also for the educated youths employed in private sector enterprises. Job cuts and pay cuts in the private sector have been myriad during the last six to seven months, some of which are reported and some are not. Just in June 2020, Uber Eats exited Bangladesh, an avenue for both gig economy players as well as employer extraordinaire for many of the country’s top-notch graduates. Yes, the number of the ‘educated unemployed’ are rising(reported to be 33.32% in 2016–17), possibly in a phenomenal rate — the question is, what can we do to combat this?
What is wrong with the contextual low-culture gig economy?
With consumers taking advantage of the e-Commerce platforms during the pandemic, an increase of 70–80% in order volume is reported. Noteworthy risers among these platforms are true e-Commerce like Chaldal.com and Daraz, food delivery services like foodpanda, commuting and delivery services like Shohoz and Pathao, and true delivery services like eCourier and RedX. All the players mentioned above employs a large frontline workforce of contractual workers or independent contractors, e.g. deliverymen, riders etc. — gig workers by definition. However, ‘educated youths’ refrain from engaging in these seemingly lucrative earning avenues because of it being viewed through the cultural lens as ‘low-culture’ in Bangladesh — or more specifically, work that they can’t put into their resume or work that they are ashamed to partake in for being of a perceivably less sophisticated nature. This begs the question:
What are high-culture gigs? What are socially acceptable parameters that would persuade these educated, yet unemployed, youths to work and earn?
Keeping it simple, high-culture is ‘computer-based’
Speaking with a few educated youths, we could surmise that the physical labour entailing the profession of a deliveryman or a rider is seen as a less ‘prestigious’ alternative to a profession where one works with a computer at any time of the day — a reference to high-culture gigs. The most popular reference to high-culture gigs are that of a ‘freelancer’ and working with individuals and brands to develop computer graphic works, data entry, and/or IT-based gigs. Platforms like Upwork and Freelancer.com are enablers of these gig works and these are preferred because of its hyperflexibility. A deliveryman can only operate from 11 AM to 10 PM wherein computer-based gig workers can engage in their assignments at any time provided it fits with the clients they are serving. These gigs can also take the form of digital brand promoters, ambassadors, content creators, enumerators and other gig-based professions. Now we are talking about people who are final year students or recent graduates from colleges. While the job of a delivery person is always consistent and one could feel a measure of financial security given the higher order volume of e-Commerce sites during COVID 19 season, ‘prestige’ is always a crucial social imperative for these youths.
Global decline in freelancing
Bangladeshi freelancers numbers around 0.7 million and earns well when they play their cards right. However, since COVID 19 hit the US, UK, and the EU in and around mid-February 2020, freelancing gigs had all but dried up — making times tough for high-culture gig workers in Bangladesh with massive layoffs and complete shutdown of many outsourcing businesses. Some evaded the brunt of the collapse, particularly legacy gig workers who have an established client base, but for most of them, ‘moving on’ would be the best possible option.
Cost-effective image labelling services for the booming AI market
Forbes reported the global AI movement and machine learning market forecast stands at a compound annual growth rate (CAGR) of 42%~ between 2017 to 2024. Cyber Security Intelligence, Tractica and many other think tanks conservatively predict the global AI market to be an industry worth north of USD $120 billion by 2025.
Banking on these numbers along with our own fascination with ‘thinking machines’, we began Acme AI in 2019, an image annotation outfit that feeds in training data to predominantly computer vision-based machine learning (ML) models and we are good at it. We make use of the enormous educated workforce available for the pandemic-induced job market collapse and give the unemployed educated youths an opportunity to join the AI-bandwagon to not only gain great pay and hyperflexible work hours, but great experience and social ‘prestige’ as well being at the frontier of the next technology revolution.
Since 2019, we have employed over 200 annotators (we like to call them ‘mutant gig workers’; X-Men reference) who works/worked on varying degrees of ML models — each unique to its own — and earns well over USD $200, greater than the national household income per month by a sizable margin. Some of them earn even more e.g. for medical annotators where we engage junior doctors fresh out of the academy and/or final year medical students who deal with a complicated workflow. But these are only the supply side of things. Being a company based in Bangladesh (and a young one at that), we can provide annotation services at a significantly lower cost than that of the global average — particularly in the space of geospatial intelligence, healthcare and retail automation, and autonomous vehicles. Similar service providers in India and the Philippines are also engrossed in this price competition because of having similar infrastructure and manpower support.
Fantastic infrastructural challenges and how we are solving them
Bangladesh has its own set of challenges, starting from the occasional load shedding and cost of internet during these trying times. Our annotators work through cloud-based annotation platforms, some of which pack a punch when it comes to data packets. While we put a few platforms to test, we hit a diamond in the rough with SuperAnnotate who boasts the fastest annotation platform with a series of automated functions that makes image labelling much faster; and we believe them! During times of uncertain electricity and data pack costs, speed is always a welcomed, and necessary, add-on, especially when we are cultivating computer vision-based training datasets. Not only does their cloud platform loads faster, but for some of our projects, we observed close to 10 times increase in service speed after adopting SuperAnnotate because of their automation algorithm. They have also released a desktop app which we haven’t tested yet, but the notion of offline work is quite intriguing and would solve our current predicament in its totality.
How to become an annotator? Is it hard?
Absolutely not. Anyone can become an annotator, and we designed our core training pedagogy accordingly. The project-specific training differs based on type and circumstances, but we have yet to see any issues in the uptake of image annotation skills — which can usually be transferred in the span of a day or less. Some educated youths who are inherently tech-savvy can pick up the tools in under 2 hours, +1 for SuperAnnotate in that department as well. We hold 2–3 training sessions per week, which are generally attended by 25–30 people per session. Many of our annotators are taking an interest in testing out key applications they develop after going through Daniel Shiffman’s Coding Train videos on YouTube; still nascent, but such a wonder to watch them as they grow.
Suffice to say that the concept of ‘gigs’ is evolving from the traditional models popular in Bangladesh. Educated youths are finding it a lucrative option in an era where jobs aren’t as plentiful (or even in a reality where it is) within their own cultural bubbles. Our hope is that this article, despite its limitations, reach these unemployed educated youths so as to begin a new train of thought on engaging with the space of AI, and subsequently its ML application. The industry is making noises, big ones like that of OpenCV, TensorFlow, PyTorch and an ever-expanding list of innovative, ML-driven solutions. It's up to us to support and benefit from this movement as much as we can.