Whilst the payout wasn't huge within the early days of Amazon, those who got in early are now seeing huge rewards, with 38% of shoppers starting their buying journey within Amazon (source), making it the number one retail search engine. Some studies are suggesting that Amazon is responsible for 80% of e-commerce growth for publicly traded web retailers (source).
As I tinker with dialog systems at the Allen Institute for Artificial Intelligence, primarily by prototyping Alexa skills, I often wonder what AI is still lacking to build good conversational systems, punting the social challenge to another day. This post is my take on where AI has a good chance to improve and consequently, what we can expect from the next wave of conversational systems.
Nowadays a high majority of high-tech banking organizations are looking for integration of automated AI-based solutions such as chatbots in their customer service in order to provide faster and cheaper assistance to their clients becoming increasingly technodexterous. In particularly, chatbots can efficiently conduct a dialogue, usually substituting other communication tools such as email, phone, or SMS. In banking area their major application is related to quick customer service answering common requests, and transactional support.
In our work at ZipfWorks building and scaling intelligent shopping platforms and applications, we pay close attention to emerging trends impacting digital commerce such as chatbots and mobile commerce. As this nascent trend towards a more conversational commerce ecosystem unfolds at a dizzying pace, we felt it would be useful to take a step back and look at the major initiatives and forces shaping this trend and compiled them here in this report. We’ve applied some of these concepts in our current project Dealspotr, to help more shoppers save more money through intelligent use of technology and social product design.
Google, the company with perhaps the greatest artificial intelligence chops and the biggest collection of data about you — both of which power effective bots — has been behind here. But it is almost certainly plotting ways to catch up. Google Now, its personal assistant system built within Android, serves many functions of the new wave of bots, but has had hiccups. The company is reportedly working on a chatbot that will live in a mobile messaging product and is experimenting with ways to integrate Now deeper with search.
On the other hand, early adoption can be somewhat of a curse. In 2011, many companies and individuals, myself included, invested a lot of time and money into Google+, dubbed to be bigger than Facebook at the time. They acquired over 10 million new users within the first two weeks of launch and things were looking positive. Many companies doubled-down on growing a community within the platform, hopeful of using it as a new and growing acquisition channel, but things didn't exactly pan out that way.
It's fair to say that I'm pretty obsessed with chatbots right now. There are some great applications popping up from brands that genuinely add value to the end consumer, and early signs are showing that consumers are actually responding really well to them. For those of you who aren't quite sure what I'm talking about, here's a quick overview of what a chatbot is:
As artificial intelligence continues to evolve (it’s predicted that AI could double economic growth rates by 2035), conversational bots are becoming a powerful tool for businesses worldwide. By 2020, it’s predicted that 85% of customers’ relationship with businesses will be handled without engaging a human at all. Businesses are even abandoning their mobile apps to adopt conversational bots.
Smart chatbots rely on artificial intelligence when they communicate with users. Instead of pre-prepared answers, the robot responds with adequate suggestions on the topic. In addition, all the words said by the customers are recorded for later processing. However, the Forrester report “The State of Chatbots” points out that artificial intelligence is not a magic and is not yet ready to produce marvelous experiences for users on its own. On the contrary, it requires a huge work:
In business-to-business environments, chatbots are commonly scripted and used to respond to frequently asked questions or perform simple, repetitive calls to action. In sales, for example, a chatbot may be a quick way for sales reps to get phone numbers. Chatbots can also be used in service departments, assisting service agents in answering repetitive requests. For example, a service rep might provide the chatbot with an order number and ask when the order was shipped. Generally, once a conversation gets too complex for a chatbot, the call or text window will be transferred to a human service agent.
I will not go into the details of extracting each feature value here. It can be referred from the documentation of rasa-core link that I provided above. So, assuming we extracted all the required feature values from the sample conversations in the required format, we can then train an AI model like LSTM followed by softmax to predict the next_action. Referring to the above figure, this is what the ‘dialogue management’ component does. Why LSTM is more appropriate? — As mentioned above, we want our model to be context aware and look back into the conversational history to predict the next_action. This is akin to a time-series model (pls see my other LSTM-Time series article) and hence can be best captured in the memory state of the LSTM model. The amount of conversational history we want to look back can be a configurable hyper-parameter to the model.
Perhaps the most important aspect of implementing a chatbot is selecting the right natural language processing (NLP) engine. If the user interacts with the bot through voice, for example, then the chatbot requires a speech recognition engine. Business owners also have to decide whether they want structured or unstructured conversations. Chatbots built for structured conversations are highly scripted, which simplifies programming but restricts the kinds of things that the users can ask.
We then ran a second test with a very specific topic aimed at answering very specific questions that a small segment of their audience was interested in. There, the engagement was much higher (97% open rate, 52% click-through rate on average over the duration of the test). Interestingly, drop-off went wayyy down there. At the end of this test, only 0.29% of the users had unsubscribed.
If it happens to be an API call / data retrieval, then the control flow handle will remain within the ‘dialogue management’ component that will further use/persist this information to predict the next_action, once again. The dialogue manager will update its current state based on this action and the retrieved results to make the next prediction. Once the next_action corresponds to responding to the user, then the ‘message generator’ component takes over.
Being an early adopter of a new channel can provide enormous benefits, but that comes with equally high risks. This is amplified within marketplaces like Amazon. Early adopters within Amazon's marketplace were able to focus on building a solid base of reviews for their products - a primary ranking signal - which meant that they'd create huge barriers to entry for competitors (namely because they were always showing up in the search results before them).
Back in April, National Geographic launched a Facebook Messenger bot to promote their new show about the theoretical physicist's work and personal life. Developed by 360i, the charismatic Einstein bot reintroduced audiences to the scientific figure in a more intimate setting, inviting them to learn about the lesser-known aspects of his life through a friendly, natural conversation with the man himself.
In 1950, Alan Turing's famous article "Computing Machinery and Intelligence" was published, which proposed what is now called the Turing test as a criterion of intelligence. This criterion depends on the ability of a computer program to impersonate a human in a real-time written conversation with a human judge, sufficiently well that the judge is unable to distinguish reliably—on the basis of the conversational content alone—between the program and a real human. The notoriety of Turing's proposed test stimulated great interest in Joseph Weizenbaum's program ELIZA, published in 1966, which seemed to be able to fool users into believing that they were conversing with a real human. However Weizenbaum himself did not claim that ELIZA was genuinely intelligent, and the introduction to his paper presented it more as a debunking exercise: