Dec 17, 2021
A small chapter in Natural Language Processing, A huge leap for General AI
Advancements in NLP like Google LaMDA are promising and in the right direction to realize our vision of general artificial intelligence.
Language is the cornerstone of civilization. It reflects the dynamism and interaction in cultures and evolves with communities.
This very nature of language makes it extremely difficult to create true human-like Conversational AI.
But the critical question is, what makes an interaction a conversation? Is it the context, multi-directional nature, or the free-flowing open-endedness of jumping from one topic to another while retaining context?
In his 1950s paper “Computing Machinery and Intelligence”, Alan Turing described a ‘Thinking’ test to determine whether a machine can act indistinguishably from how a thinker (human) works. Since then, we have seen NLP models evolving from the very basic non-semantic based models like Bag-of-Words to prediction-based models using advanced neural network techniques to most recently ‘attention based’ networks allowing focus on specific sub-set of inputs.
The problem remains, all these models work well for highly specialized tasks but fall apart as soon as the application areas become more generalized, or simply put, they lack common sense. This challenge leads the way to a new era of highly complex & large language systems. First to launch was OpenAI’s GPT-3 (Generative Pre-trained Transformer 3) with 175 billion parameters. It took the world by storm, promising creation from fiction to poetry to music, and most importantly, it brought reasoning and intuitiveness to the mix for the first time.
Recently launched LaMDA (Language Model for Dialogue Applications) from Google took it a notch up by bringing several nuances of open-ended conversation in the model. One in specific is sensibleness, which means does the response generated by the model makes sense in the context of the conversation? Google showcased a cool demo in a candid talk with Pluto (yes, the earlier Ice planet) during their 2021 I/O event.
Though far from human-like dialogues and conversations, these advancements are very promising and in the right direction to realize our vision of general artificial intelligence, and who knows, maybe one day we might have our own personal JARVIS!
