If a human can do it, AI can do it more efficiently

There is no clear definition of Artificial Intelligence, different people define it in different ways.

It is an ambitious and strong statement. Of course, not all human data can be performed by AI today, but many, more than one can believe, defined tasks are fully possible to automate using AI.

For us, AI is a Machine Learning based system that you can train beyond the limits of a traditional non-ML system. The AI-system is not by itself smart, it becomes smart through training. In that way, it can be thought of more like a resource than a system.

Other people would say that only deep neural networks are AI. This will probably change in the future because in, let’s say, 20 years from now, the definition of AI will be different from the one we have now. This is because AI is constantly evolving.

Among else, some of the projects we've been working on in our Seavus AI department is data validation and product classification.

Data validation is about training an AI system to detect errors in data by being shown good product data and problematic product data. It can be any types of error and it covers all those cases were the traditional rules would be unfeasible because of the complexity and number of rules needed.

Our product data classification system works by eliciting information from a database with products in some product hierarchy and classifying it in another hierarchy, possibly enriching it with data features extracted from secondary sources. This is a very time-consuming and error-prone work if done manually but can successfully be performed with an AI.

The most challenging part is predicting the results of an AI effort. It's not like in traditional programming, where you define the testing criteria and make a system that does the allocated processes. The AI can, and will, learn what it’s being shown. By saying that, it means that the basis for what the AI can achieve lies in the patterns there are to discern in its training data.

AI is powerful because, even though it needs training (in some cases a lot of training), it usually decreases the amount of time and resources needed to solve a problem. In the end, the AI training part might be much cheaper and faster than the programming part in a traditional system.

Plus, when you introduce an AI in a business environment you can extend its knowledge and use it for other purposes. Thus it becomes more and more mature in time. Who can benefit from this? Well, basically everyone.

AI can generally replace or enhance a lot of workflows and processes where manual work, validation, categorization or data extraction is needed – basically everything that is repetitive. A huge number of business cases exists within text-based data processing and that’s what we're focused on at the moment. Most of the digitalization and automatization of administrative processes include text-based data processing, traditionally performed either manually or by static IT system, and it’s here that AI can drastically enhance the workflows.

In the end, what could be said about AI is that if a human can perform a certain task, then a machine can do it as well. It's just a question of how complex the system needs to be and the amount of resources needed to train it. 

It's ambitious but it's manageable. And it's exactly what we do.