10 Ways that Human-in-the-Loop Machine Learning is Used Today
From Human-in-the-Loop Machine Learning by Robert Munro
This article discusses concrete examples of human-in-the-loop machine learning in practical use and why you should learn more about it.
You can get the book for 37% off by entering fccmunro into the discount code box at checkout at manning.com.
One of the most important questions in technology today is how can humans and machines work together to solve problems? More than 90% of applications that use Artificial Intelligence improve with human feedback. For example, autonomous vehicles get smarter the more that they observe human drivers; smart devices get smarter as they hear more voice commands; and search engines get smarter by observing which sites people actually click on for each search term.
Human-in-the-Loop Machine Learning Machine Learning details the process for optimizing the interaction between Machine Learning algorithms and humans who create the data that powers those algorithms. The book goes into technical detail about every part of this process: from interpreting where algorithms need the most help, to quality control over the human feedback component. It covers the foundational techniques in areas like Active Learning and more recent advances like incorporating Transfer Learning into Human-in-the-Loop architectures.
This article complements the technical components of my book by highlighting 10 different ways that people are using Human-in-the-Loop Machine Learning today, each focusing on a different advantage that it brings. Ten advantages of using Human-in-the-Loop systems over purely automated systems are: avoiding bias, creating employment, augmenting rare data, maintaining human-level precision, incorporating subject matter experts, ensuring consistency & accuracy, making work easier, improving efficiency, providing transparency & accountability, and increasing safety.
This article covers each example in turn. Every example is taken from a news article that was published in the first month of 2020, highlighting just how widely Human-in-the-Loop Machine Learning is being used today.
1. Avoid Bias
Machine Learning models can easily become biased because they are trained on data that is itself biased. Having a human in the loop can detect bias early, as Dianna Paredes, CEO of Suade, recently said at the World Economic forum:
2. Create Employment
While the focus on artificial intelligence is often about how automation can take jobs away, it is also creating new jobs in data labeling. Because humans are needed to train most algorithms, there are now tens of thousands of people who specialize in labeling data for machine learning. People labeling data outnumber the people building machine learning algorithms and tend to come from much more diverse backgrounds. So, like this example in India shows, Human-in-the-Loop machine learning can allow people globally to benefit from the AI boom from new jobs:
3. Augment rare data
Most popular machine learning algorithms require large amounts of labeled data to produce accurate results. However, there are many cases where there is not even a large amount of unlabeled data to draw from. For example, if you are looking for examples of fake news in a language with only a few thousand speakers, there might not yet be any examples of fake news in that language. Therefore, the algorithm will have nothing to learn from. In this case, keeping humans in the loop can ensure the same level of accuracy even for rarer types of data, like in this example about monitoring at Facebook:
4. Maintain Human-Level Precision
There are many applications where you never want the AI to fall below human-levels for a task. For example, if you are manufacturing critical equipment for an aircraft, then you can improve safety by using machine learning for the inspections, but you don’t want to sacrifice safety for the sake of automation. So, you still need a system that can be monitored by humans to ensure you always get human-level precision, like in this article:
5. Incorporate subject-matter experts
If you have subject matter experts (SMEs) creating the training data, then you can create some very sophisticated applications. There are many industries where SMEs work closely with machine learning-driven technologies, like this is use case in finance from Nasdaq which is a great example:
6. Ensure consistency and accuracy
Machine Learning models are often more accurate on some types of data than others. This can result in applications that are much less consistent than humans, excelling in some areas and operating poorly in others. For critical tasks like audits, this could be unfair to the targets of that audit. So, the “Big 4” accounting firms are investing in Human-in-the-Loop Machine Learning architectures to ensure consistency:
7. Make work easier
The most obvious benefit from Machine Learning is that you can make work easier by automating many of the tasks. This is the case in this cyber-security example, where it is not possible to fully automate the detection of security threats for computer systems. However, you can still semi-automate as many of the tasks as possible to make the jobs of security professionals easier:
8. Improve efficiency
Improving efficiency typically comes hand in hand with making work easier and many people have debated whether or not healthcare can (or should) be automated. Like this article points out, a recent Stanford study suggested that Human-in-the-Loop systems out-perform humans or AI alone:
9. Provide accountability and transparency
Interpreting the decision of a machine learning model can be very difficult. If the model has thousands or even millions of parameters, which is common, then any “interpretation” of that model will have to be an approximation because there is no way that a human can truly understand the complexities of a model that large. For life-changing tasks, like this example about whether or not to give a visa to someone who wants to enter a country, you can have potentially negative decision always back-off to a human to provide accountability and transparency:
10. Increase Safety
There are many ways that machine learning is improving our safety and the most obvious example is when autonomous vehicles will have fewer accidents. Even non-autonomous vehicles can leverage Human-in-the-Loop Machine Learning to improve safety, as with this example from Formula 1, where professional gamers are helping with simulations of actual driving conditions:
Each of these examples of Human-in-the-Loop Machine Learning come from articles published in the first month of 2020, covering a very diverse set of use cases: international standards & governance; data labeling; content moderation on social media; manufacturing for semi-autonomous vehicles; monitoring for financial market manipulation; auditing; cybersecurity; healthcare; and motor-sports.
Despite the large number of different industries using Human-in-the-Loop Machine Learning, it is still a relatively new field. The primary evidence for this is that my book is the first one dedicated to this topic! If you’d like to learn more about how these kinds of technologies are implemented, please pick up a copy of the book and look out for other articles like the ones we shared here!
If you want to learn more about the book, check it out on our browser-based liveBook reader here.
This article was originally posted on Manning’s Free Content Center: https://freecontent.manning.com/10-ways-that-human-in-the-loop-machine-learning-is-used-today/