Remember, you’re a bot: Why product managers must be the boss of NLP
When it comes to the more creative tasks, natural language processing (NLP) lacks one thing: the human touch. However, this could soon change thanks to one simple technique we as humans use to learn language: listening. Google is now training its NLP algorithms with human dialogue.
We’ve all heard about the capabilities of natural language processing (NLP): from its ability to provide almost instantaneous translations to its potential to write articles and even books. When it comes to the more creative tasks, though, NLP lacks one thing: the human touch.
However, this could soon change thanks to one simple technique we as humans use to learn language: listening. Just as infants learn to form sentences and understand meaning by listening to the constant babble of voices around them, Google is now training its NLP algorithms with human dialogue, enabling the technology to replicate a more conversational tone.
Businesses and technologists alike have also been eagerly anticipating the arrival of AI-powered machines that are indistinguishable from human operatives. But before they hand the customer service reins over to a chatbot, product managers must ensure that they continue to capture the invaluable insight that springs from real customer interactions.
With great power…
NLP has become embedded in our everyday lives, and although the term isn’t commonly used, what it does is widely understood. Almost everyone knows that the ads they see online aren’t generated at random but are tailored to our search history.
This is NLP in action, but it only scratches the surface of what the technology can do. The latest NLP iterations are being used to replicate human communication patterns so closely that it’s becoming even harder to tell if you’re speaking to a bot.
We’re all familiar with chatbots, and most of us have had good and bad experiences with them. As customers, we don’t really care about how closely they mimic human speaking patterns: what matters is how well they understand us. And this depends on their ability to learn from the data they gather, including from each interaction with a customer.
The advantage of NLP is that it has the potential to deliver much richer and more relevant data than simple search terms. If I type “Bluetooth speaker” into a search bar, all the retailer knows is that I’m interested in that product. If I were having a conversation with an operative – whether a bot or a human – I’d likely yield a huge amount of additional information: for example, how much I’m willing to spend and what features I’m looking for.
The announcement of Google’s two new datasets, ‘TimeDial’ and ‘Disi-QA’ promises a leap forward in AI’s ability to learn from human speech, further extending the promising capabilities NLP can bring to enterprise organisations. It’s crucial that product managers understand how much power is at their disposal…while also being aware of how to avoid the potential pitfalls of relinquishing control to the machines.
…comes great responsibility
As with any superpower, the old adage always rings true: “with great power, comes great responsibility.” And, let’s be clear, NLP is a business superpower if it’s used correctly. But, with any technology, there are also limitations that can’t be ignored.
In order to get the most value from NLP, product managers must remember that NLP is not a fix-all. For example, while “human-style” interactions may generate huge amounts of relevant and timely information, it can hugely complicate the task of categorising, analysing, and turning this data into insight.
Going back to our example of the Bluetooth speaker enquiry, it’s easy to see the benefits of simple search queries: they are clear and unambiguous, enabling businesses to identify patterns and trends. While the same is true of conversation-based queries, it’s much harder to mine unstructured data, extract the value within, and aggregate it.
It’s also important to guard against the eternal problem of incorrect, inaccurate or biased data sets. AI devours data, but unlike humans, it can’t always discount obvious errors. And the consequences can be more severe than lost sales or inaccurate analytics. For example, if a healthcare chatbot is trained using an outdated medical journal, then it might provide incorrect information to a patient. This can either cause unnecessary panic or result in the patient taking the wrong action.
These scenarios can be avoided with proper training, which is a task for data scientists. But product managers also play a critical role: only they know what insight they need to capture and what errors must be avoided. Above all, product managers are responsible for being the voice of (human) reason. Like the auriga whose job it was to remind Roman emperors that they were only human, someone needs to be the voice that reminds the organisation that bots are still several orders of magnitude less sophisticated than the human brain. When NLP’s limitations are fully understood, that’s when organisations can truly unleash the power of its potential.