There seems to be much confusion among business users about the extent to which Intelligent Automation solutions actually involve artificial intelligence. This article gives a simplified practical overview, starting with the basics.
Why is there so much confusion around Artificial Intelligence (AI)?
There are many definitions out there, and they change all the time
Collective term AI mixes together concepts, methods, philosophies and applications
There is a lot of hype, and the specific terms are often misused
Human mind typically cannot think in multidimensional spaces
Is there a simple way to understand AI?
An AI model is an algorithm which fits a function to some kind of data. In essence, all AI technologies are evolutions of line-fitting. Its simplest form is simple linear regression, fitting a straight line to two-dimensional data. That is easy to imagine. Then the models become more complex, increasing the polynomial degree and the number of dimensions, until we cannot visualize them anymore. However, in essence they remain the same.
Why AI model training is so important?
"Out of the box", an Artificial Intelligence model is just an algorithm. There are a bunch of different types of AI models (neural network being perhaps the best-known). It needs to be "trained", i.e. fitted to a particular dataset, before it becomes useful. The larger and more comprehensive the dataset, the more powerful the model becomes.
AI models are widely accessible, and many great ones are open-source. However, the magic happens in choosing (or developing) the right model for the application, and then properly training it on a sufficiently large body of data. In this article we elaborate on the potential of pooling companies' resources for training AI to transcend the boundaries of human learning and by extension the current limitations of knowledge works.
What are the business applications of AI?
Any context where reliable data is available and accurate predictions can be useful (e.g. predictive maintenance, supply chain management, consumer behaviour prediction) is a natural application field for Artificial Intelligence. However, AI can and should be applied in a much wider variety of use cases to augment algorithmic automation.
Artificial Intelligence used in conjunction with Robotic Process Automation (RPA) is called Intelligent Automation.
Conclusion
The main benefit of Intelligent Automation lies in non-algorithmic decision-making and adaptability. While most objectives can be accomplished with just RPA (avoiding the need for extensive AI model training), AI helps increase the flexibility, adaptability and, ultimately, self-sufficiency of the solution.
Ultimately, algorithmic decision-making is limited by the ability of the system's architects and users to apprehend the kinds of decisions it needs to make. In contrast, AI enables flexible decision-making by analogy: if a new data point is sufficiently similar to a meaningful number of earlier occurrences, the system can infer that it should be treated the same way without the need to program a rule for this decision, simply by observing the decisions in the training data set. It can also do so based on massive amounts of data which would not be possible to process manually.
AI is a very strong tool, and we are constantly finding new applications for it. At the end, as with every tool, solid understanding and experience are required to achieve good results.