4 mins read

Google LaMDA: A Conversational Intelligence Revolution

4 mins read

Pivoting on conversational technology, many tech giants unite day and night to pull back the drawbacks and limitations that still exist in voice assistants and robotic chatbots.

The pre-coded set of options and responses delivered by them over one-on-one user conversations are yet not 100% efficient and that ultimately, allows us to find them useless at most times.

Well, among many struggling to leverage the language models, transforming and vetting AI and ML algorithms, Google, the overlord, has come up with an invention named LaMDA (Language Model for Dialogue Applications) that aims to replace AI robotic conversations with more natural dialogues.

LaMDA: A Conversational AI that Goes Beyond the Existing

We know that every company with a new conversational feature launched in its product claims to make enhanced human-like conversations but Google’s LaMDA is an approach to create bots that are capable to handle their conversations in a more natural, human-like manner.

It presses more on ‘dialogue-based context’.

Unlike other conversational paradigms, LaMDA is trained on the concept of dialogue and picked up several nuances that distinguish open-ended conversation, one of the major nuances being sensibleness and specificity.

Satisfying responses like “that’s nice” or “‘I don’t know” are sensible but may not be specific to the query asked by the user at times. And this is what exactly is witnessed today when users communicate with voice assistants or chatbots.

If some input is given to a chatbot that is out of its scope, it is usually answered along the lines of:
“Sorry, I cannot understand what you mean. Would you like to take the conversation forward with one of our representatives?”

Here, Google’s LaMDA is made more advanced to handle such conversations where a chatbot will resolve a user query itself to a great extent without directing the query to a human executive at an early stage, thereby, being highly sensible and context-specific.

LaMDA: The Internal of the Technology

LaMDA is built on Transformer which is a neural network architecture and open-sourced in 2017. The recent algorithms of BERT and GPT-3 also work on Transformer.

The architecture uses an attention mechanism and allows the language model to predict text(s) following the previous words/conversations related to each other.

Though it uses some principles from the other existing models; nonetheless, its distinction to manage the ‘open-ended nature’ of conversations and engage in a free-flowing way with an endless number of conversational topics, helps the newly emerged concept to unlock more natural ways of interaction and communication between humans and machines.

As human conversations are distinctive and chaotic, LaMDA is prepared to handle such unexpected situations where a conversation initiates with one topic and in a few moments ends with a different one.

With LaMDA tackling communications with a range of connecting topics, it will bring revolutionary changes within Chatbot Technology, making chatbots more humanly- intelligent, interactive, and engaging with people than ever before.

LaMDA’S Core Conversational Principles

[Sensible, Specific, Interesting, and Factual]

As briefly told earlier, LaMDA is trained in rendering dialogue-sensitive responses. As the quality of a chatbot is highly proportionate to the SSA (Sensibleness and Specificity Average) which is one of the human evaluation metrics, LaMDA’s competence went a step beyond and included two more factors of being interesting and factual.

Excelling at detecting sensibleness in the context of a conversation and responding in the most specific manner, LaMDA is also expected to provide interesting answers alongside considering the responses to be appropriately factual (which the powerful GPT-3 systems still lack).

Interesting, insightful, and compelling responses won’t matter to individuals unless they are truly correct.

Furthermore, Google is also concerned about whether their technology will adhere to the AI principles, thus, making all the righteous sense to all types of conversations encountered (like biases, hateful speech, deceitful information, unfavorable comments. etc.)

Being mindful of such risks, the tech giant wants to put ‘responsibility first’ as an ethical principle to be incorporated by LaMDA to minimize the misuse of such models.

The Latest Status of LaMDA

Well, it’s in the making and soon to be launched for official use.

Till then a lot of developments and testing cycles are being conducted so that LaMDA does full justice to its invention and leaves no stone unturned to augment the conventional language and conversational models.

Demonstrations on LaMDA’s competence are being conducted and in the recent one, LaMDA was seen conversing with its human counterparts being the planet Pluto and a paper plane.

And surprisingly, it was seen rendering answers based on its fundamental propositions of sensibleness, specificity, fact-based, and compelling. Below is an excerpt representing LaMDA’s conversation impersonating Pluto, keeping the order of the dialogue:

Image source : https://towardsdatascience.com/googles-lamda-the-next-generation-of-chatbots-62294be58426

Google LaMDA is the Next Big Thing in Conversational AI!

We expect LaMDA to eliminate the inconvenience users face while conversating with voice assistants or robotic bots. The test conducted must focus on the degree of humanness it can achieve and bring breakthroughs in the conversational intelligence of machines.

Well, I believe that getting deep into human interaction is out of reach for AI as human conversations are acutely dynamic and have an ever-changing context of speech while communicating.

What do you think?

Will AI be able to acquire pragmatic knowledge just like us? Let us know your thoughts on the same and if you feel to discuss and know more on this trending subject,
get in touch with our AI experts and know how your business can take advantage of conversational AI and LaMDA-enabled chatbots.