Google’s Meena is a Conversational AI That Is Coming
Google Brain has a product that I’m watching closely. Like they say in their blog, modern conversational agents (chatbots) tend to be highly specialized — they perform well as long as users don’t stray too far from their expected usage. However the dream of a conversational AI is real.
Companies like Google, Huawei, Baidu and Amazon are hard at work on this. But Google Brain might be in the driver’s seat to attain it.
As platforms like Apple and Facebook evolve, Google’s aim to unify some of their products is really interesting. Google with their uptake in the Cloud is getting to the point where an acquisition of something like Slack becomes more likely, but there’s also huge potential fortheir AI to become useful.
Conversational AI is a catch-all term for natural language models for artificial intelligence that can interpret human words, speak to people or carry out tasks or computations with natural language. We’re in a golden age of NLP and converstaional AI will do really well in the 2020s.
While the chatbot hype didn’t exactly live up to expectations, Google Assistant is improving and Meena represents another level.
Today, conversational agents are a bit limited, and Google is working towards a human-like chatbot “that can chat about anything.”
Companion Level Conversational AI Is Coming Soon
The Google Duplex demo raised a bit of hype that Google is on the right path for conversationally useful AI. The idea of chatting about anyting with an AI is appealing for some of us who also want the smart home to be fully voice enabled with something a bit more shiny (and relatable) than dull and stupid Google Assistant.
To better handle a wide variety of conversational topics, open-domain dialog research explores a complementary approach attempting to develop a chatbot that is not specialized but can still chat about virtually anything a user wants. This sort of companion-AI is really the holy grail for companies like Baidu, Amazon, Alibaba and Google who all have very competitive voice assistants with speakers and screens.
It’s not a question of if but a question when.
Google’s Meena has a 2.6 billion parameter end-to-end trained neural conversational model. Google thinks Meena can handle multiturn dialogue and claims it’s better than other AI agents built for conversation and available online today.
While Google’s CEO trumpets the potential of AI, Google pointed to the published paper, which details a “human-like” chatbot that can “engage in conversation on any topic”. That’s what Apple said about Siri, and now Siri lags behind the times.
Google today also released Sensibleness and Specificity Average (SSA), a metric created by Google researchers to measure the ability of a conversational agent to maintain responses in conversation that make sense and are specific.
Humans rank around 86% in SSA and in initial tests Meena scores a high of 79%. Mitsuku, an AI agent created by Pandora Bots that’s won the Loebner Prize for the past four years, got a 56%, while Microsoft’s XiaoIce, which speaks Mandarin Chinese, got a score of 31%.
Curiously unrelated to Meena, Google is working on another communications application, this one for workplaces, that will combine several different platforms it already operates, according to a new report from The Information. Just as Facebook seeks to centralize its encrypted chat, Google appears to be doing the same in its own way.
Apple’s wearables growth appears to be an on-ramp into its ecosystem as well. EarPods and wireless ear buds are obviously where the majority of conversational AI would have to take place. Nobody wants to talk to their phones or use “keywords” to wake their smart speakers constantly throughout the day or at work.
It’s not clear with Google Brain how close Meena is to action with the consumer. Google Duplex itself does not seem very fast to be implemented. Google keeps on tangling AI hype without a good relationship with or understanding of consumers.
Google Duplex and Meena however do show that Google is doing important AI work. Meena is trained on 40 billion words and utilizes a seq2seq model and a variation of the popular Transformer architecture. Google first released Transformer in 2017, but since then the language has grown to rank among the highest performing language models around.
You will know you grew up in the 4th industrial revolution if you primarily deal with voice to navigate how AI augments your life. That’s the defining mode of search and control in the IoT world of the “Alpha” cohort. (If Gen Z was born in 1995 to 2009, the Alpha gen are those from 2010 to 2024).
Meena is detailed in “Towards a Human-like Open Domain Chatbot,” a paper published Monday on preprint repository arXiv. Google Brain is a deep learning artificial intelligence research team at Google. Formed in the early 2010s, Google Brain combines open-ended machine learning research with systems engineering and Google-scale computing resources. You can read their blog here.