LaMDA(Language Model for Dialogue Applications) – An AI Chatbot from Google

 

Google announced their next Gen AI Chatbot i.e. LaMDA on 18 May, 2021.

To understand LaMDA, let’s first get a brief about what is NLP

Natural Language processing(NLP) and AI-technology for businesses is an increasingly popular topic and all but inevitable for most companies. It has the power to automate support, enhance customer experiences, and analyze feedback.

NLP describes the interaction between human language and computers. It’s a technology that many people use daily and has been around for years, but it often taken for granted.

A few examples of NLP that people use are:-

Spell Check

Autocomplete

Voice text messaging

Spam filters

Related Keywords on search engines

Siri, Alexa or Google Assistant

Where NLP outperforms human is in the amount of language and data it’s able to process. Therefore, its potential uses go beyond the examples and make possible tasks that would’ve otherwise taken employees months or years to accomplish.

 

History Behind LaMDA

LaMDA is not the first of its kind technology introduced by Google.

The new language model is built on Transformer(It is an architecture that adopts the mechanism of attention and influence the different parts of input data), neural network architecture that google invented and open-sourced in 2017. The architecture is then used to produce a language model that can be trained to read many words. The model can read a sentence or a paragraph, understand how the words relate to one another and then predict what is to come next in the context.

Google provided a hint of such a model in 2018, with Bidirectional Encoder Representations from Transformers, or BERT. Based on the same neural architecture(The architecture that produces a model that can be trained to read many words (a sentence or paragraph, for example), pay attention to how those words relate to one another and then predict what words it thinks will come next. ) as LaMDA, the technology enables anyone to train their own question answering system.

BERT models could understand the full context of a word by analysing the words that came before and after it. The ability was found to be particularly good in the case of preposition like ‘for’, ‘to’, which are often misplaced in search queries.

See How it gave response to the user in below image. It feels a bit funny.

 

How does LaMDA works?

Unlike most other language models, LaMDA was trained on dialogue. Since these conversational dialogues can be loosely associated with each other, LaMDA’s primary focus is to stay on track with the conversational flow.

Normal language models are not able to do so. This happens because every conversation does not contain a question mark. It might be exclamation or a simple acceptance of what AI stated. This is the point where most of the AIs go “Sorry, I don’t understand.”

LaMDA shines in the above situation because of its focus on expression of open-ended conversations.

Another element is the wittiness(clever humour) of the response. It is one thing to hold the conversation and make it interesting. Google says that it is working on LaMDA to bring in that element.

Conclusion

We have all seen one or the other AI assistant in a sci-fi movie. Take Iron Man’s Jarvis, for example. The superhero’s constant companion did not just answer the simple questions but it converses with him at every level. It would not just power his thrusters; it would tell him how long he has before he drops like a dead fly from the sky.

Google’s LaMDA is a step in that direction. We can very well envision many of our smart devices being controlled through such voice commands in the near future.

So, imagine coming home from office and asking your smart door/security system if anyone visited in your absence. Or imagine asking your smart oven for the right temperature to bake your food item.

Maintaining a conversation will be a crucial step for the next-gen language model.

If and when Google succeeds with its LaMDA project, it will allow the company to do exactly that. Google would also be able to assign different personalities to different devices, or even better, create a singular control system that understands your needs. More importantly, it will be able to talk to you in a human-like manner throughout the conversation.