How to Train Your Robot
Step one: think like a human
Here at ATB, our Bot Squad has been busy.
Yes, you read that correctly: we’ve got a Bot Squad whose work is centred on training software robots to automate banking processes like payments, cash advances, and fee and interest reversals on ATB’s Mastercard operations team. The first of those robots is Botson (named by an ATB team member in a contest—way to go, Myvan!) and we’ve just closed in on the end of Phase 1 of training and exploration with it.
Unlike Pepper, Botson is a software robot, which means it doesn’t embody a physical robotic form (no labcoat-wearing scientists chasing after a robot with clipboards and stopwatches here!). This type of software is referred to as robotic because it’s malleable and can be trained, which is where the Bot Squad comes in. And interestingly, a lot of the training process has us thinking more like humans than robots…
See, we initially thought we could train our robot by sitting down, recording the steps necessary to complete a task and then loading them into the robot’s software and walking away. What we quickly realized is that completing multiple tasks throughout a day takes a lot more than the basic steps. Because every time a new task arises, it can bring with it a unique set of circumstances. So what Botson needs to be trained on are the exceptions to rules, the "what ifs" and the analysis of multiple outcomes that can be required for a single transaction. In other words, the critical thinking skills our human team members possess in droves.
What this means is that we need the experts on hand to train Botson—team members who know the work better than anyone else—who can quickly identify exceptions that may come up and who can call a spade a spade when it just doesn’t make sense for a robot to complete a certain task. When the robot is suited for a task, it needs to be broken down into small, logical pieces; robots work very well in situations where high volume, short task processes need to be completed. They also do well with highly structured data, and with moving data between systems. Where they need some help is in the more human realm of critical analysis and situational decision-making.
Lucky for us, we have an amazing team of humans to do the human work, so our main goal with robots like Botson is to offload repetitive tasks to the bots to free up time for our highly skilled employees to learn more, innovate, and foster relationships with our clients across Alberta.
Our experience thus far with Botson has taught us a lot about how to train a robot. Here are our top three takeaways, should you be planning on training a robot of your own in the near future:
Think about today and tomorrow, and the “why?” - It’s not just about solving today’s problems in the realm of robot training. It’s also about getting critical about today’s problem-solving and how it can be improved, and also about looking forward to what may need solving in the future.
Put on your pessimism hat - Just kidding (mostly). When training a robot, you do need to consider all the ways a task or process can go wrong, however, so that you can them how to deal with all possible outcomes. Spread the love - Once you’ve got a robot trained to complete certain tasks, it’s helpful to think about other contexts those problem-solving skills could work within. Think about this like getting more robotic bang for your buck.
The beauty of all of this is that it’s just the beginning. We’ve got Botson completing a couple types of tasks, but as we train and iterate its processes, we’re already envisioning the significant possibilities of this work, and the significant possibilities it will open up for our human teams to explore, grow and excel.