Transfer Learning in Customer Service Automation

Understanding transfer learning

Transfer learning is key when it comes to solving certain problems with artificial intelligence (AI). But to understand exactly what transfer learning is, let’s take a step back from AI for a minute.

You probably know how to ride a bike. Let’s say, you were to attempt to ride a motorcycle for the first time. Some of the skills you know from riding a bike — like how to steer using handlebars and how to balance on two wheels — would come in handy on the motorcycle. That’s transfer learning. In the same way a human would use their knowledge of cycling to make riding a motorcycle easier, AI can use transfer learning to tackle new problems more efficiently and effectively. 

Why do we need transfer learning? 

If you’re familiar with machine learning (ML), you might be wondering why you’d use transfer learning instead of deep learning. If you’re unfamiliar with these concepts, here’s a quick explanation:

• Machine learning is a type of AI that allows software applications to get better at predicting outcomes without being explicitly programmed to do so. 
• Deep learning is a type of ML that attempts to mimic the human brain so it can “learn” from large amounts of data.
• Transfer learning is an ML method where knowledge gained from completing one task helps solve a different but related problem.

Deep learning uses algorithms inspired by the structure and function of the brain called artificial neural networks (ANNs). ANNs enables machines to “learn” (read: make very accurate predictions) from large amounts of data. 

For example, deep learning enables AI to learn how to convert between Celsius and Fahrenheit by analyzing examples of conversion pairs (0ºC = 32ºF, 10ºC = 50ºF, etc). This can be useful in complex cases where we do not know the function (formula) between input and output. 

Deep learning allows AI to handle increasingly sophisticated functions, but current systems require a lot of data. The more complex a system is (for example, learning a language is more complex than temperature conversions), the more data the ANN will require to learn how the rules operate.

So what to do then, when data is limited, but data quantity is so highly tied to solution quality?

That’s where transfer learning comes in. Transfer learning allows machines to use the knowledge gained from other tasks in order to tackle new but similar problems effectively. 

How we leverage transfer learning at Ultimate

Transfer learning is one of the fundamental ways we develop our AI here at Ultimate. Two keys ways that we leverage transfer learning are to create a language-agnostic solution and to solve the problem of having small datasets in customer service automation. 

Leading the market in languages: the first language-agnostic solution

‍Ultimate is a Finnish startup and so had the pleasure of starting with one of the world’s toughest languages. For deep learning AI, Finnish faces two problems: 

1. It’s really, really complicated (lots of rules)
2. With only approximately 5 million Finnish speakers worldwide, there isn’t much data on it

Ever heard that learning more languages gets easier as you go? Maybe from a friend that’s boasted that speaking Italian and French have made Spanish a breeze? This phenomenon isn’t just true for humans. In Google Translate’s Zero-Shot Translation algorithm, probably the most famous beneficiary of transfer learning for scaling languages, Google uses a single model to translate between multiple languages, scaling the learnings from one translation pair (English →Japanese) to another (English →Korean).

At Ultimate, we use a similar technique for customer service automation. Like Google, one of our adoptions for this is understanding languages. Solving the Finnish problem with transfer learning prompted us to develop our architecture to use a single model across all clients and regions. Today, this development story has made us the only truly language agnostic provider of customer service AI. This allows us to scale easily and to help our customers serve their customers across all markets. And the more we scale, the better we get. ‍

Learning from limited data: cracking the data problem in customer service AI

More interestingly, by being able to apply ways of thinking from one task to another, transfer learning unlocks deep learning potential from smaller datasets. 

Let’s go back to our original example: humans develop basic knowledge of riding a motorcycle from riding a bike. It’s a similar case with AI for customer service. Each time our AI is trained on a new customer service use case, e.g. adding an ecommerce company to our portfolio of telco and travel customers, its previous experience in identifying customer problems in other industries elevates its accuracy with the new case. As such, as we scale at Ultimate, we get better at using less data to achieve higher and higher accuracy levels.