Video originally published on The Medical Futurist YouTube channel on 10 June 2019.
Dr. Bertalan Meskó a.k.a. The Medical Futurist is correct and data annotators are indeed the frontliners who work to pave the way towards efficient AI solutions. Why? Because the accuracy of data labels ultimately affects the functional accuracy of an ML model. The need for data annotation services is well over the global supply of trained annotators in different fields such as geospatial and satellite imagery labeling, autonomous vehicle annotation, medical AI, human pose estimation, sports analytics, etc.
Case of Medical AI
Say for instance, Medical AI, a field that is so specialised that it requires doctors, trained nurses and/or annotators with special qualifications to detect and label foreign bodies in medical imagery. Going through the image below, I could not, for the life of me, differentiate between the two chest X-Rays. I know something is up on the right image given there is a bit of a haze on the lower lung, but then again, there is a bit of a black spot and minute hazes on the left image at the bottom right.
I showed the image to a doctor intern at Acme AI and he immediately recognised the right one to have a form of pneumonia. These expert tiers of annotators work well and are extremely precise. While annotation work for medical AI pays well, it might not be the go-to for doctors who earn the lion-share of their livelihood by providing medical services to patients.
From the perspective of the global south, recent trends showcase final year students from a pharmacy background and/or young doctors expressing an interest in providing data annotation services, predominantly because of it being an avenue for them to accumulate experience, earning remotely and on-the-go (a cultural pivot experienced during COVID-19) while the pull of working on a transformative medical AI model is also a major factor.
On a similar note, we saw a marked increase in quality and agility of annotation in the space of pose estimation by animators and/or people having a certificate in character modelling as opposed to their peers. This is because of their relative familiarity with rigging human and animal anatomical structure — giving them an advantage over specific tier of work. Replicate this by sect and there are specific people whose experience and education give them advantages in certain types of annotations.
The requirement for expert human labelers is tied to the rise of AI and the 4th Industrial Revolution that we are on the cusp of realising. It is instrumental for AI developers and researchers to have accessible, quality, and affordable training datasets — a formulae that we are experimenting with providing the best possible package for our clients and hopefully play a part to match demand of annotation services soon.
Data preparation and engineering tasks represent over 80% of the time consumed in most AI and Machine Learning projects. The market for third-party Data Labeling solutions is $150M in 2018 growing to over $1B by 2023.
COO, Acme AI
Have a data annotation need? Submit your requirements at https://www.acmeai.tech/requirements and we will get back to you.
Acme AI Ltd. is a data annotation and artificial intelligence solution developer. We support development and augmentation of artificial intelligence systems in the space of agricultural AI, retail automation, robotics, autonomous vehicles, medical AI, geospatial imagery among many more with a focus on computer vision-based labelling operations. Acme AI is a strategic partner to SuperAnnotate and Alegion and is one of the youngest companies to have won a Grand Challenge by the Bill & Melinda Gates Foundation.