Many years ago I read about artificial intelligence, and one specific subset caught my attention: Machine Learning (ML). I googled it and learned that algorithms and statistical models are used to make predictions or decisions without being explicitly programmed by someone to perform a particular task. This blew my mind. When I started going deeper and working on a PoC, I realized that it wasn’t as straightforward as I had anticipated and that I ought to specialize in a new branch of computer science and become a data scientist. Unfortunately, life took me along other paths, and I couldn’t pursue that specialization.
Today, there are a bunch of new services to run AutoML, that allows non-experts to make use of machine learning models and techniques without any advanced knowledge in this field. Basically, with an AutoML system, you can provide the labeled training data as input and receive an optimized model as output.
Some research suggests that the percentage of development teams using this kind of service to add AI capabilities to their applications is around 2% these days. By 2023, this number is expected to grow up to 40%. By 2025, 50% of all data scientist activities will be automated by AI. This will improve the ROI of data science investments, and bring down the time necessary to implement ML.
Due to the complexity of a common Machine Learning process, the demand for these kinds of methods, that can be easily implemented without expert knowledge, has been on the rise in the past years. This makes sense if you think of all the tasks involved in an ML project, that can be sped up and simplified in an AutoML project:
A traditional machine learning workflow involves a lot of time of experts in different fields working together to solve a sophisticated problem. But many of these tasks (grey items) can be automated almost up to 100%, which opens the door to non-experts to generate, deploy, and manage models to produce inferences.
There are many tools to implement AutoML projects, like H20 and Datarobot. All the mainstream cloud providers also offer services to use them. For example, if you want to generate a model to make predictions with images (object detection or image classification), there are some tools that you could use:
- Amazon Rekognition Custom Labels
- Google AutoML Vision
- Microsoft Custom Vision
In conclusion, this technology is still evolving and new opportunities are being created constantly in this field. The market is hungry for this kind of service, that is both easy to use and to implement. Of course, this isn’t something to be applied in every situation: if you need something more complex, you would still have to talk to a data scientist to generate a custom model, according to your needs.