Every day, business is concern about improving the performance and increase the efficiency and ROI. And one of the most progressive solutions is AI technologies.
And each day AI consultants get hundreds of requests to implement business optimization via machine learning techniques.
To better understand what we are talking about, let's go through the terminology. ML is a class of artificial intelligence methods, a subclass of strict functional programming languages. ML allows you to develop an algorithm, carry out machine learning, analyze information flows, and, based on a complex of data, make predictions that form the basis for decision-making. To better understand the idea, check how different models work here. But the use of ML is not magic, it is a complex system that is not always relevant and can not do everything. Let's figure out in what cases it is rational to use ML, what it can do, and what not.
When ML is not desirable
Please note that the ML models is not always the best solution! I propose to consider a few real examples from the practice of programming.
- You have a set of data with obvious patterns and a simple task. In this case, it is quite possible to write code that will be automated without ML. Unconditional Automation is better than ML. You will spend less time and resources, and get 100%
- The client set the task to classify the documents received daily. The data visualization software identified the keywords that distinguished these documents. This made it possible to write code with which documents could be distributed according to certain keys, namely, to automate information. If we decided to go along the path of building an ML model, we would have to annotate some documents, before training the ML model, then set the time and money consumption. As a result, the ML model could give a false prediction that needs to be Good explanation of the false positive and false negative prediction is provided in this video.
As practice has shown, the written lines of code turned out to be more effective in solving such a problem, but as for using ML, it is better to think carefully and work out alternative solutions.
- One more example. The task is to perform segmentation on satellite images for green areas and swimming pools. The quick solution was to use a static program, base. Colours were used for visual segmentation, the so-called segmentation with a parser. The order was closed quickly and without any problems. Note that the ML model was not required to solve the
As you can see, the application of the ML model depends on the established scenario. Sometimes it is more rational to use standard programming rules that differ from how ML works.
What Machine Learning can't do?
Above, we considered in which cases it is better not to use the ML model since there are more rational options for solving certain problems. However, it should be understood that ML may not cope with a certain scenario at all. For example, a customer asks to build a model that automatically performs customer support tasks. Namely, it automatically gives answers to questions, deals with complaints advise on the company's services. The maximum that the ML model can do in this case is to organize a complaints route leading to the right team, organize an analysis of customer reviews. However, it is difficult to imagine such an ML model that will completely replace the call centre and will adequately respond to customer complaints.
When artificial technologies are the best solution?
There are many problems that no one can solve better than the ML model. These are automatic text translations, face detection, and much more. Just think about the possibilities of machine learning (ML): “An ordinary person can solve any mental task in a split second, while the ML model can automate such work, and this is possible now.” One feature of ML is that machine learning is impossible without the availability of training data, and the larger their volume, the more effective the constructed ML model will be.
Guest post courtesy of Paul Taylor