Increasingly, we’re spending our time online communicating not with real, flesh-and-blood people, but with machines. But exactly how do these machines work, and why are they so important? Let’s find out.
Artificial machines and humans communicate in very different ways. Human language is a hodgepodge of semi-improvised and often illogical transcriptions of utterances whose meaning depends on context, body language and other implicit information. Machines, on the other hand, communicate via a resolutely logical series of zeroes and ones.
The challenge of building a chatbot that responds in the same way that a human does lies in overcoming this language barrier. To say that this is something of an engineering challenge is an understatement. Even simple requests, like asking whether there’s any cheese left in stock at a given shop, are tricky for a machine to handle. Subtle conversational undertones and idiosyncrasies are considerably more so.
Modern chatbots meet these challenges through a series of techniques, collectively called Natural Language Processing, or NLP. This term covers a machine’s ability to break apart incoming verbal communication and convert it into a structure that’s comprehensible to a machine. A separate piece of code, called the Natural Language Generation part, then takes the machine’s reply and converts it into text.
So, suppose that you run a business selling pipe fittings. A human customer might get in touch, providing a vague description of the sort of thing that they need. A human salesperson might know from experience and context how many of a given item might be required. Moreover, they’ll be able to sense subtext while listening to the customer – sensing dissatisfaction and adapting their tone accordingly. The more advanced NLP engines become, the better able chatbots will be to do this.
When a new user lands on a website, they’ll have limited knowledge of what the site does, and whether it fits their requirements. Reading the site might provide a general overview, but more specific needs can best be met through conversation.
At their simplest, a chatbot might act like the Frequently-Asked-Questions section of your website. They’ll interpret a user’s communications by searching through a database of predefined questions and answers. More sophisticated chatbots will present the information alongside a sensitivity to the customer’s expectations, providing the tact and flexibility we associate with a great sales rep. But they also offer advantages which even the most patient and understanding salesperson can’t match.
While human operators will expect to go home at around five, and won’t handle calls well in the middle of the night, a chatbot can be made available constantly. At it won’t become tired or irritable as a result! This allows urgent tech-support queries to be handled even at unsociable hours.
A single chatbot can handle thousands of conversations at the same time. This allows it to effectively do the work of thousands of real-world call handlers. This eliminates errors, and slashes the wage-bill considerably. Humans are poorly-adapted to the repetitive, menial questions that constitute the overwhelming majority of calls. Chatbots, on the other hand, can be built for the purpose.
If your business sells across many different territories, language barriers can pose a problem. Humans who can speak several languages are in high-demand, and their labour is thus costly. Chatbots can detect the language being spoken and reply in kind, allowing even small businesses to more easily venture into newer markets. There’s no such thing as a ‘native-speaking’ chatbot; learning a new language is a simple matter of plugging in the right software.
A good human salesperson will be well-versed in the products being sold. But there are limits to human knowledge. What if there are millions of products being sold? What if the database is being updated with thousands of new ones every day? A machine will have encyclopaedic knowledge of even the largest and most fluid catalogue, and be better positioned to provide prompt and informed answers to customer queries. Never again will customers need to wait to see whether an item is in stock; the chatbot will know instantly.
Writing intelligible sentences is one thing. Convincingly aping the intricacies of human speech is quite another. Developing a chatbot that can do this requires an enormous amount of machine-learning, informed by an enormous amount of data. Fortunately, the internet is awash with exactly the sort of data that’s required, in the form of the communications we all make via email and social networks. But leveraging this information to produce a convincing chatbot will still take time and patience.
As it stands, chatbots can play an important supplementary role, dealing with common complaints and enquiries and forwarding more arcane ones to human operators. If the chatbots are convincing enough, the human customer might not realise that a transition has taken place. Ultimately, chatbots will grow so sophisticated that no such transition will be necessary!