Why you should do it, how it works, and how we can help make it work for it
Human-in-the-loop (HITL) is a branch of artificial intelligence that leverages both human and machine intelligence to create machine learning models. In a traditional human-in-the-loop approach, people are involved in a virtuous circle where they train, tune, and test a particular algorithm. Generally, it works like this:
First, humans label data. This gives a model high quality (and high quantities of) training data. A machine learning algorithm learns to make decisions from this data.
Next, humans tune the model. This can happen in several different ways, but commonly, humans will score data to account for overfitting, to teach a classifier about edge cases, or new categories in the model’s purview.
Lastly, people can test and validate a model by scoring its outputs, especially in places where an algorithm is unconfident about a judgment or overly confident about an incorrect decision.
Now, it’s important to note that each of these actions are comprise a continuous feedback loop. Human-in-the-loop machine learning means taking each of these training, tuning, and testing tasks and feeding them back into the algorithm so it gets smarter, more confident, and more accurate. This can be especially effective when the model selects what it needs to learn next–known as active learning–and you send that data to human annotators for training.
Human-in-the-loop is an approach Figure Eight has championed for years. We’ve seen it help improve models of every stripe, whether they’re text classifiers, computer vision algorithms, or search and information retrieval models. We can create vast quantities of highly accurate training data for your unique use case, then tune your model with human insight, and test it to make sure its decisions are accurate and actionable. If you’d like to learn more, please don’t hesitate to reach out.
The human-in-the-loop approach combines the best of human intelligence with the best of machine intelligence. Machines are great at making smart decisions from vast datasets, whereas people are much better at making decisions with less information.
For example, people are great at looking at a complex image and picking out discrete entities: “this is a lamppost” or “that’s a cat but you can only see its tail.” This is the exact sort of information a machine needs to understand what a lamppost or a cat looks like. In fact, a machine needs to see a lot of different lampposts and cats–from different angle, partially occluded, in different colors, etc.–to understand what one looks like. A robust dataset of these labeled images (i.e. human intelligence) teaches a machine to see those items (i.e. machine intelligence). And at some point, with enough data and enough tuning, those machine algorithms can see and understand images incredibly quickly and incredibly accurate without the need for people to constantly tell it what exactly a cat (or a lamppost) looks like.
– For training: As we discussed above, humans can be used to provide labeled data for model training. This is probably the most common place you’ll see data scientists use a HitL approach.
– For tuning or testing: Humans can also help tune a model for higher accuracy. Say your model is unconfident about a certain set of decisions, like if a certain image is in fact a cat. Human annotators can score those decisions, effectively telling the model, “yes, this is a cat” or “nope, it’s a lamppost,” thus tuning it so its more accurate in the future.
Active learning generally refers to the humans handling low confidence units and feeding those back into the model. Human-in-the-loop is broader, encompassing active learning approaches as well as the creation of data sets through human labeling. Additionally, HitL can sometimes (though rarely) refer to people simply validating (or invalidating) an output without feeding those judgments back to the model.
HitL can and is used for manifold AI projects. This includes NLP, computer vision, sentiment analysis, transcription, and a vast amount of other use cases. Any Deep Learning AI can benefit from some human intelligence inserted into the loop at some point.