Google’s Meena: Why it’s a breakthrough
Just like me if you too were amazed by Spike Jonze’s directorial ‘Her’ where the protagonist Theodore (played by Joaquin Phoenix) falls in love with an AI Samantha(played by Scarlett Johansson). Actually, this movie was one of the reasons which hypnotized me to chose Computer Science(especially Machine Learning & Artificial Intelligence) as my professional career and so whenever some breakthrough like this is made in the field, it feels like a personal gain to me, that’s why i’ve decided to write this blog.
Overview
What is Meena?
What is its history?
What will be its impact?
Why it’s a breakthrough in right direction for multi-turn chat bots?
what to expect now?
What is Meena?
Meena is a chatbot released by Google recently claiming it to be the best voice assistant at the moment. We already have some very popular chatbots like Apple’s Siri, Google’s Google Assistant, Amazon’s Alexa, Microsoft’s Cortana, etc. But they aren’t good enough as the quest of all the virtual assistant AI is to match human-like experience which seems to be very near after looking at the results of Meena.
Now, what is this Sensibleness and Specificity Average (SSA)?
It’s a human evaluation metric that decides how humanly(Sensibly and Specifically) do you answer. This metric is designed by Google which argues that SSA captures basic yet important attributes for natural conversations.
What is its history?
Chatbots have been here for more than fifteen years now, and tech giants like Google, Facebook, Amazon, Microsoft, IBM, etc are betting big on this, the availability of chatbot frameworks like DialogueFlow from Google, Watson from IBM’s, Bot Framework from Microsoft, RASA(an open-source framework) seduces developers and enthusiasts to have their eye on it constantly in the name of achieving the technological singularity.
ELIZA is considered to be the first chatbot in the history of Computer Science which was developed by Joseph Weizenbaum at MIT. ELIZA operates by recognizing sentences and keywords from its provided keywords to reproduce a response using those keywords from pre-programmed responses. Like, Someone says ‘I love my Company’ then ELIZA would pick up the word ‘Company’, and reply by asking an open-ended question ‘Wow! Tell me something about your company’. Which felt like an Auto-Chatter but it was just a programmed response and nothing more but that was the inception of chatbots and our quest for humans like chatter which was later followed by ALICE(a natural language processing powered bot) and Steve Worswick’s MITSUKU
These days even you may go on to create your own chatbot using these above-mentioned frameworks with a very minimal effort. I highly encourage you to go on and try it (especially if you’re not from a Computer science background.)
What will be its impact?
Since Meena model has 2.6 billion parameters and is trained on 341 GB of text, filtered from public domain social media conversations for thirty(30) days on a TPU v3 pod (2,048 TPU cores). Compared to an existing state-of-the-art generative model, OpenAI GPT-2, Meena has 1.7x greater model capacity and was trained on 8.5x more data.
More data means better results.
Why it’s a breakthrough in the right direction for multi-turn chatbots?
Although Google, Amazon and Microsoft’s assistant are hugely successful they are miles away from Samantha from the movie “HER” :-)
Another reason why our expectations are so high because Google has recently achieved the Quantum Boost, So it’ll be easier, faster and far more efficient for them to achieve a human-like chatbot than any of its adversaries.
What to expect now?
Although Google claims it to be the best but at its heart lies the seq2seq architecture which isn’t something completely revolutionary and is common in the community but since Google has quantum power which allows them to feed Meena a huge data to achieve better results which means we’ve come a long way but clearly not there yet.
