Conversational A.I. — The front door for human-machine symbiosis
Everything we say or do with an A.I. platform is a valuable input in its learning cycle as Natural Language Processing (NLP) has undergone a major paradigm shift in the last decade.
Natural Language Processing (NLP) is a branch of artificial intelligence that helps computers understand, interpret and manipulate human language. Its abilities have led to a compelling SaaS (Software as a Service) atmosphere, as apps started to adapt to user behavior and habits rapidly without the need for back end re-working. We have seen the adoption and utilities of interactive AI in every social fabric, the great thing about them is their computation and memory, and it is infinite and pure. Automated assistants tend to excel when conversations are narrowly defined and linear, but this is beginning to change, thanks to advances in Natural Language Understanding (NLU) and Computer vision. As the technology progresses, we would see AI assistants which can recognize and respond to human cues such as facial and body language, vocal modulation and other emotional signals.
AI giants like Google assistant and Amazon Alexa, uses Bidirectional Encoder Representations from Transformers (BERT) and Transfer and Adapt (TANDA) approach respectively. They have set the bar high by enriching their vast data corpus and language models working towards the ultimate human-machine experience.
Amazon’s Echo speakers dominated the market with a 69.7% share in 2019. Coming in at a distant second is Google’s Assistant-enabled speakers, with 31.1% of the market.
The reason for Alexa’s dominance persists because of their grass root methodology as Amazon researchers first tapped into two popular NLP frameworks namely Google’s BERT and Facebook’s RoBERTa. They measured accuracy with mean average precision and mean reciprocal recall and reported that both the BERT and RoBERTa models with fine-tuning on TANDA provided a “large improvement” over state of the art, and that they are “an order of magnitude” less affected by the insertion of noisy data. In a second experiment, the team built questions sampled from Alexa customers’ interactions. They say that using TANDA with the aforementioned RoBERTa produces an “even higher” improvement than with BERT, and that TANDA remains robust against noise. In short, Amazon made a smart move by combining the best of both worlds.
Back in 2017, you could achieve state of the art by using Bi-lstm and attention mechanism for encoding the input data. It eventually lagged with lot of limitations like establishing deep contexts, tons of background knowledge and failing to capture polysemy. By 2019, this space has transformed by game changers like BERT which created new methods of training models where pre-trained word representations can be context-free, meaning that a single word representation is generated for each word in the vocabulary, or can also be contextual, on which the word representation depends on the context where that word occurs, meaning that the same word in different contexts can have different representations. In a span of 10 years, word embedding has revolutionized the way almost all NLP tasks can be solved, essentially by replacing the feature extraction/engineering by embedding which is then fed as input to different neural networks architectures. This newfound state-of-the-art is bringing a paradigm shift in the realm of Conversational AI.
In this day and age, as we migrate towards sophisticated AI-first enterprises it is of paramount importance that the diagnose-prescribe-predict-personalize loop coupled with smart integration to services is the only key in moving forward to a true level of loyalty and retention.
Gartner says that by 2021, 15% of all customer service interactions will be wholly taken care of by AI. By 2022, 70% of white-collar workers will interact with conversational platforms on a daily basis. That is an increase of 400% from 2017 and customer support is just a single part of the story. By integrating your products and services, exploiting the data and by improving operational excellence you could create lucrative business models using Conversational AI.
Deep learning’s relationship with enriched data builds a virtuous circle for strengthening the best products and companies: more data leads to better products, which in turn attract more users who generate data that further improves the product. It naturally trends toward winner-take-all economies within an industry.
