Little Learning Machines Preview

From Code Romantic to Autonauts, I’ve always loved programming games with a charming aesthetic, so I was really excited to give Transitional Forms’ Little Learning Machines a go during its Early Access phase. In it, you’ll befriend and train adorable AI robots — Animos — teaching them how to complete increasingly complex skills and navigate all sorts of obstacles as you visit whimsical islands.

These robot friends are brought to life by an inquisitive blue crystal, Imogen, who also serves as our tutorial instructor. Through Imogen, we learn that each Animo comes with its own neural network, which is essentially a brain with a blank canvas, and you’re the painter. Your first robot is a delightful yet goofy blue being named BillyBob (note: you can customise the names, but I stuck with the game’s suggestion). The first task is to teach it how to collect crystals because it can’t really do much without our input, and its movement by default is very random.

All training takes place in the Training Cloud, a magical purple world where you can create custom environments for your Animo to try out its skills. Here, you’ll use positive and negative reinforcement to teach the robot what to and what not to do by using two special variables: Learning Value (Love) and Failure Reason (Fear). So, to complete your first quest of collecting crystals, you’ll tell the little machine, well, that it loves collecting crystals! Then you can assign it a reward of either +1, +10, or +100 Love for each one it grabs, using positive reinforcement to encourage it to keep on gathering. It’s not as easy as setting everything to the highest amount, though. Too much Love for one action might mean that your robot only does one thing and nothing else. Too much Fear and the AI might just stand still, literally paralysed with fear. This is because those neural networks are always seeking to optimise, which can lead to some funny scenarios as the machines essentially try to cheese the system.

For instance, I ran into a tricky scenario trying to teach an Animo how to pick up a watering can and water flowers. However, the AI discovered that it was getting more Love from the action of picking up the can versus the actual watering, so it got trapped in an infinite loop of picking up and putting down the item. It was getting rewarded for this, so why would it stop? I solved this by setting -1 Fear for setting down the watering can but then ran into an issue where the robot was pouring water on nothing and still gaining Love for simply using the item. And back to the drawing board, I went.

Simply put, you’ll run into quite a bit of trial and error when training your robot friends. Thankfully, the game makes setting these parameters easy to grasp, as you’re given a set menu of actions and values to select from versus having to write them in yourself. Unique icons also illustrate the actions, such as picking an item up or stepping on a surface; you’ll see these visuals often as your robots carry out tasks, so you can see exactly what they’re up to and when. The game has done a great job so far of making its mechanics accessible, distilling rather complicated concepts into something players of all levels can have fun with. If you ever forget what something means, as I do quite a bit, there’s a detailed info panel that breaks down all the ins and outs of the AI.

While Little Learning Machines isn’t an idle game, it does have some idle elements. Training the Animos isn’t an instant process, as you’ll have to let them go through a trial simulation to fully learn whatever it is you’ve assigned them to do. At the very start of the simulation, the robots will just take random actions, but as time passes, they’ll start to do what you’ve programmed. It definitely takes longer than you expect, and there’s not a clear indication of when your Animo is ready to head back into the world, so you might be staring at the screen waiting for some time. There is an option to speed up the simulation, but the game notes that it doesn’t speed up the actual training, so I’m not honestly sure why that option is there. It certainly seemed like it sped up the training process when I used it, but hopefully, the dev will give more clarification about that feature later on during Early Access.

The world of Little Learning Machines is the definition of cosy, with relaxing music to top off the wholesome experience. You’ll visit Dog Bone Island, Cottage Cove, and more. Teaching Animos how to pet adorable pups and how to water gardens of wilting daisies are just a few of the skills you’ll give these tiny robots. Additionally, as you complete objectives, you’ll receive customisation options such as headphones, glasses, and even eggshells. Dress the robots to your heart’s delight!

Little Learning Machines plans to be in Early Access for less than a year. The version I played definitely feels close to a finished state. Performance-wise, it ran great, with no crashes or bugs throughout my time playing it. Developer Transitional Forms stated that it does plan to add better balancing for quests, new cosmetics, and new interactions for items based on community feedback. So, if you’re interested in a cute game about little machines, AI, and reinforcement-based learning, keep an eye on this one as it develops!