In the recent years, we have seen tremendous achievements in the field of AI from outsmarting humans in many logical tasks to identifying faces of people sometimes even better than a normal human being. You cannot name one field where AI is not being used, albeit in smaller capacity, but it does play a major role in almost every field. Now AI is being tested for its metal in an entirely new field, language. Mastering the unique and complexity of the languages used by humans to communicate will be AI’s most challenging task yet. But seeing the rapid pace of advancements in AI, this dream is not far off.
How will AI Transform Communication?
The ability for computers to understand human languages will revolutionize how we communicate with brands, organizations across the world. In this modern day and age majority of the companies do not have the time to answer queries of each consumer individually. But if these companies leverage the power of AI in understanding human language this cumbersome task will be eased. Companies can listen, understand, and answer almost every query anytime through any channel boosting their ability to take note of precious customer feedback. A happy customer is always beneficial for the company.
Difficulties ahead
Designing an AI model that is capable to understand human language is not a piece of cake and it will take time for such AI models to be designed, trained and then to be unleashed in the wild. For reference it took until 2015 for computer scientists to finally understand on how to build a human face recognition algorithm that can accurately identify faces comparable to that of humans. Facebook’s Deep Face AI model is accurate up to 97.4% of that of a human performance. All this was possible because Facebook solved two key problems that were being faced in order to solve this task. Firstly, how to design an architecture that could mimic the way humans learn and second was where to provide data that would be used to train this AI model. Similar approach could be used for achieving the similar result in the field of language, but It is not always that simple.
Language AI models being developed in full swing.
In 2013 Tomas Mikolov along with his team at Google made a breakthrough, he discovered on how to create an Algorithm that able to learn what individual words meant by mapping synonyms on top of each other. Still much was to be done as context was still missing from this model and it was a necessary requirement for AI to language. This was achieved when Google introduced the BERT model recycled from an old architecture whose main purpose was machine translation. Google trained this model by teaching this model to fill out missing words in Wikipedia articles, the team finally achieved in teaching the context of a sentence to the AI model. In 2020 Google researchers were finally able to beat human performance in language by training the BERT model on much larger network with even more data and name it the T5.
The True Cost for AI-based Language Understanding
All this tech is not cheap, the T5 algorithm a matured version of Google’s BERT costed $1.3 million in cloud compute. Normally these AI models are not shared on the internet or any person until they buy a license from its developer. However, as it is a new technology its still very difficult to come up with more efficient solutions to solve the same problem. But the future is promising as companies start to realize AI based language’s true potential and start investing in its R&D.
I’m Ava Paul, an experienced news website author with a special focus on the entertainment section. Over the past five years, I have worked in various positions of media and communication at World Stock Market. My experience has given me extensive knowledge in writing, editing, researching and reporting on stories related to the entertainment industry.
