In order to understand data annotation, it is essential to take a step back and first understand: what is the need for data annotation?
In 1889, Charles H. Duell who was the Commissioner of the US patent office, is reported to have said that “Everything that can be invented has been invented.” This was in the context of saying that the patents office may soon need to downsize, or even close as a result.
Did that happen?
Developments have been quite to the contrary and we have been witness to technological innovations rapidly gathering pace and affecting almost all aspects of our lives. Whether it is controlled flight or nuclear power or antibiotics or television or computers or the internet, all these have been developed or invented after Mr. Duell’s assertion.
Does the pace of development look like slowing down?
Quite to the contrary, the pace has never been hotter. It seems mankind is always on the cusp of breakthrough inventions destined to change our way of life.
In its identified niche of data services, oWorkers
provides data annotation services and other data-services support needs. We have been identified as one of the top 3
data entry services providers in the world.
Artificial Intelligence
One of the developments that has been gradually gathering steam in the background and is now entering mainstream usage in daily lives is that of Artificial Intelligence, or AI.
The road to understanding ‘what is data annotation’ passes through the center of Artificial Intelligence (AI).
AI is the term used for technologies, or software programs, that have the ability to mimic human behavior.
We know human beings are the most intelligent life form. At the same time, we also know human beings are irrational. We know human beings have good days and bad days. We know human beings have mood swings. We know human beings have prejudices and personal preferences.
What if we could harness human intelligence but deploy it in a manner that takes the human frailties out of the equation? Would that not be the perfect world?
That is the premise AI is based on. And that is the effort people have been making; develop AI algorithms or AI engines that can mimic human behavior in an impartial, objective, consistent, efficient manner.
But, of course, it is not simple. Human intelligence is the handiwork of millions of years of evolution. Expecting it to be replaced by a machine merely by snapping your fingers or flicking a switch is not possible. It is a slow, tedious, painstaking process also known as Machine Learning, or ML.
Human beings have known for many decades how to create a software program using formatted text, or software coding. Formatted text, or coding, and the subsequent interpretation of software code by a computer, drives computers, in a very layman-esque definition. If human beings had not developed coding and its interpretation, computers would not have done anything.
We are, in a way, going through a similar climb with AI. The additional challenge with AI is that the computer needs to interpret not just coded language which it has been doing so far, but raw data that it takes in without any formatting or coding. Hence, it needs to be taught how to recognize and understand the raw data that it is expected to encounter and respond to it in a manner befitting a human mind.
Let us take the most commonly quoted example of AI application these days, that of self-driving cars. The AI that controls the car, unfortunately, does not have innate intelligence like a human being. If a person is crossing the road, a human driver will slow down or stop or swerve to avoid hitting the person. For the AI that gets the image/ video of the road ahead, it is just raw data, without any attached meaning. It has to be taught that if in its ‘vision’ it encounters a shape that has certain dimensions, it means that it belongs to a human being and since we don’t drive cars over human beings on the road, it should stop or slow down or swerve to avoid that object. By doing this, say, a million times, the AI engine builds up a database that allows it to identify a certain shape or set of shapes with human beings who cannot be run over. This is what ML does. The million instances of raw data fed into the machine are known as training data sets that build the machine’s knowledge.
The end goal is that once training is over, and the AI is driving the car, it will recognize a human being if one comes into its range of vision and operate the car as required according to its programming.
The foregoing equips us to address the ‘what is data annotation’ question in the ensuing section.
As a leading provider of data annotation services, oWorkers has been supporting global clients in developing and shaping their AI models with the help of its experienced and trained workforce. Led by a management team with over 20 years of hands-on experience.
What is data annotation?
But what about data annotation, which is what we were trying to understand?
The process of enriching ‘raw’ data in order to create ‘intelligent data’ that can be understood by an AI engine, and that constitutes training data sets, is known as annotation.
It is actually pretty close to the English meaning of the word annotation which, according to
collinsdictionary.com, is ‘a note that is added to a text or diagram, often in order to explain it.’
Data annotation, as defined by
Techslang, is ‘the process of labeling information so that machines can use it. It is especially useful for supervised machine learning (ML), where the system relies on labeled datasets to process, understand, and learn from input patterns to arrive at desired outputs.’
In the example of the self-driving car, the process of identifying and marking the human on the road in a manner that makes it through to the AI engine, is the answer to ‘what is data annotation.’
Advantages of Data Annotation
Data annotation being a facilitator in the journey of building reliable AI, and not the final output, its advantages can be linked to making the AI engine effective and reliable. That is both its purpose as well as key advantage. Its advantages are inextricably linked to the advantages of AI
Data output often suffers from the GIGO (Garbage In Garbage Out) principle. The quality of output one can expect from a machine or a computer can only be as good or as bad as the input data it received and processed. While good input data might still be spoiled by a software program or human intervention, bad data can never lead to good outcomes. Hence the biggest benefit of data annotation is that, when done well, it leads to the creation of a reliable and smart AI engine.
A related benefit could be articulated as that of customer experience that will result from a reliable and smart AI engine as opposed to an AI engine that behaves like a bumbling idiot. In fact, in the right application, customer experience resulting from an AI engine could be far better than that from a human interaction. As a simple example, in case of a request for retrieval of information, an AI engine will probably do it much faster than a human.
It is a task of great responsibility as the future depends on it. oWorkers operates from secure facilities in three geographies and multiple centers across Egypt, Bulgaria and Madagascar and is not only GDPR compliant but also ISO (27001:2013 & 9001:2015) certified. Our partnerships with technology providers ensures that we have access to the latest technologies for data annotation and other work.
A good way to appreciate the benefits of data annotation would be to review a few ‘use cases’ or applications of AI that emerge after the model has been trained and implemented. Understanding the benefits delivered by a process often leads to a better understanding of the process, as should be the case as we try to unravel the layers of ‘what is data annotation.’
Enhancing Social Media Content Relevance
Have you ever noticed that if you look for flight tickets from, say, Miami to Detroit, for some time after that, you might start receiving pop-ups and advertisements with promotional fares for the sector. This is AI at work, based on the algorithm that the engine has been taught. A bad engine might just note the sector and start feeding you with promotions for the sector. A better engine might even note whether you managed to book or not, and send promotions your way only if you failed to book in that attempt.
Social media thrives on feeding customized content to users based on their profile and the footprints they leave behind on the platform while engaging with it. With the aid of data annotation, owners of the platform will strive to feed content that is relevant and personalized.
Security Monitoring
Streetside cameras, hundreds of them, have been installed in a particularly sensitive part of town where incidents of theft and mugging have been on the rise. The footage is beamed to a control room where the policemen on duty are expected to look at the feed coming in from the hundred plus cameras and identify potential flashpoints and alert the cops on patrol. It is a cumbersome exercise to constantly switch from one to the other and so on. It causes fatigue and several people have to be deployed simultaneously.
An AI engine was developed and taught to analyse the feed coming from the cameras and warn the cops on duty of potential danger. For example, if the AI engine has been taught to identify a weapon being carried by a passer-by and raise an alert, it can do so almost instantly. Not only that, it can monitor the hundred plus feeds all by itself, releasing the cops doing this to do the real work of patrolling the street and adding to the force available there.
Autonomous Vehicles
This being the current favorite example of AI application, does not need much explaining. It has also been covered in an earlier part of this post.
Search Engines
Operating in a manner somewhat similar to social media, the objective of AI engines used by search engines is to make the information contextual and relevant.
For example, if the query is about the weather, at the simplest level, knowing where it has been asked from will make a difference to the results. If the engine knows that the person asking the question is an avid skier, returning results relevant to the person’s passion might be an absolute delight for the user. All it needs is the training data to be annotated in a manner that enables it to recognize this fact about the user.
The oWorkers advantage
oWorkers has strategically adopted the ‘employee’ model as opposed to the ‘freelancer’ model for its operations. While it brings upon us greater responsibility with regard to our staff, it enables us to exercise greater flexibility in terms of client requirements. As contributing members of local communities we are established as employers where many would wish to work, which gives us a steady stream of incoming jobseeker applications, substantially reducing our cost and effort in recruitment and training, while reducing our attrition. It also gives us the room to cater to short-term spikes in client requirements.
Being located in cultural melting pots, our teams are multilingual and offer services in 22 of the most popular global languages. All our centers are equipped to operate 24×7 if client operations need it.
Our pricing is transparent. Usually we offer a choice between cost per unit of time and cost per unit of output to clients. We have been a steady profitable enterprise, with efficient operations allowing us to share benefits with clients, and operate as locally registered entities. Our staff regularly rate us above 4.6 on a scale of 5 on sites like Glassdoor on satisfaction.
Along with the ‘what is data annotation’ question, this should address the ‘who to partner with for data annotation’ question as well.
