Turn on more accessible mode
Turn off more accessible mode
Skip Ribbon Commands
Skip to main content
Turn off Animations
Turn on Animations

Skip Navigation LinksAddBlog

  • Home
  • Blogs
  • Computers are no longer an “Idiotic Machines” with AI and ML

Augmented intelligence (AI) and Machine Learning (ML) has created a lot of traction in many industries. Customers across different industries have started to think about adopting ML in their business and IT process automation. People imagine that ML can completely take over the human tasks which are completely unreasonable expectations about ML’s capability. Right from our child hood, we have been taught that “Computer is an Idiotic Machine” and they don’t think on their own. Even a most advanced super computer is used only when it is programmed. The definition of ML which is “the ability of computers to learn without being explicitly programmed” completely contradicts with our thoughts. 

Data scientists are now trying to mimic the human learning process on machines. Basically, the success of human’s learning is dependent on some basic artifacts like training material, a method of training, and selecting the right learning tools. Scientists have completely understood this and tried to replicate the same while training the machines. ML’s architecture will have components like learner (a machine program), training material (Database or the repository of the training data), the method of training (Model of ML selected for training), and advanced learning tools (associated software components to process the training data).

ML algorithms such as supervised, unsupervised and reinforcement algorithms solve classification, regression, clustering and behavioral analytics problems. The classification through ML is highly successful if the system is trained with larger training input data set. These training data can be of digitalized images, audios, structured data from the database with rows and columns or unstructured data in the form of natural language. Training is usually done offline and the model parameters are static on new predictions. The training model has to be periodically overlooked and updated with a larger training data including newly observed data. These newly observed data can be immediately fed back to the training model to continuously improvise in real time. This would look like a closed loop feedback control system.

The accuracy of ML is related to the volume, quality of the training data and accuracy of the training model selected for solving the problem. Natural language processing (NLP) of unstructured data is the need of the hour in many industries. However, current advanced NLP technology doesn’t completely solve the knowledge extraction from the unstructured data sources. ML algorithm working on top of these inaccurate processed data doesn't produce high precision of results. This often requires manual interventions for “Continuous assisted learning” to upgrade their training data.

To conclude, ML is a set of cognitive services running on computers which to certain extent can mimic human like learning behavior. ML works on different kinds of digitized data like audios, images, structured texts and automates tasks performed by the human which can’t be addressed by the traditional office automation software. NLP has boosted the learning capability of the ML to process and learn about the unstructured data which is in the form of scanned PDF’s, images of primitive human written documents. Advances in ML and NLP technology will improve the accuracy but a complete replacement of humans by machines is still a long shot.  The cognitive automation through ML can add significant value to businesses, in terms of cutting down costs in human capital investment and improving overall productivity.