What’s the first thing that pops into mind when you hear the term “talkbot”? Is it Apple’s Siri telling you a joke? Or Amazon’s Alexa playing Despacito? Or Ok Google reminding you to buy eggs and cheese for your planned carbonara fiesta later tonight?
At the Google I/O developer conference held in May 2018, Google CEO Sundar Pichai unveiled their smart AI assistant Google Duplex to the world. Whilst watching the virtual conference at home, I remember feeling a mix of absolute awe coupled with something else.
Mind-blowing, right? Not only does the AI talkbot mimic the tonal inflexions of an actual human voice, but it also adds in natural verbal responses like “um” or “mm-hmm” much like what a human would do. In straightforward interactions like this, I wouldn’t be surprised if no one on the other end of the line had doubted that they were speaking to a fellow human.
This level of sophistication is precisely what many companies are trying to achieve with their conversational talkbots: a natural human voice using natural human tones. But, as much as we are in awe of the human-like responses of the AI assistant, there is also an undeniable feeling of fear deep down – a worry that AI like this could be abused for profit at the expense of others.
Does your bot have morals?
Ethical issues concerning AI have been like a sea of dark clouds threatening a thunderstorm ever since the advent of AI technology. Despite the initial excitement over AI talk bots, many people – from tech experts to restaurant owners – have expressed ethical concerns over the proliferation of AI talk bots. Ironically, making talk bots extremely human-like poses the biggest ethical concern: How would we know if the person we are talking to is really who they say they are?
This concern is not a trivial one by any measure. In an era of fake news and identity theft, it is extremely worrying that a single person with nefarious intentions could so easily disseminate false information and plunge communities into a state of panic. What’s disturbing is that this is already happening without the help of AI.
With the entrance of AI, electoral processes could be sabotaged, fake kidnappings could occur, and phone scams could become a daily occurrence.
Thankfully, all’s not lost yet. With the help of Wiz.ai, an industry leader in localised conversational AI research and application, we can begin to understand AI talkbots and how WIZ.AI intends to harness the power of AI while preventing AI-rmageddon from happening.
An honest conversation
To learn more about conversational AI1 and its associated ethics, I had a chat with Nicholas Ko of WIZ.AI, who is an avid fan of Japanese culture and how they have managed to seamlessly blend technology and tradition. Below is a transcript of the most exciting tidbits of conversation exchanged regarding this evolutionary technology.
Okay, first things first! What’s so special about you guys?
At Wiz.ai, we are revolutionising the customer service industry by using human-like Voice Artificial Intelligence to digitalise the process of in- and outbound calls.
The quest to increase revenue and reduce operating costs have pushed corporations towards mass automation at almost every customer touchpoint, resulting in impersonal, text-based, self-serve solutions which skimp on customer satisfaction.
Our goal is to use Voice AI to help companies re-engage with their customers at scale through the familiar, interactive medium of a simple phone call while ensuring cost efficiencies.
Our bespoke handcrafted customised Voice AI solution incorporates ASEAN languages and accents into the repertoire of our bots. We deliver an enhanced customer experience to ASEAN customers no matter what language or accent they speak in.
Our talkbots are designed to imitate human speech patterns to give a realistic human-to-human conversational experience – over 90 per cent of our users cannot tell the difference between our humanistic voice AI and a real person.
Is your technology essentially rendering call service workers obsolete and taking away more jobs than it creates?
We are in the midst of the Fourth Industrial Revolution, the new age of AI, networks of highly interconnected smart devices, and smart systems that collect and analyse user data.
However, while AI may be more efficient at completing predictable rule-based tasked within a given framework, quickly and with a consistent level of quality, they are not better at everything. They generally lack empathy and human connection.
Wiz.ai works together with call centres to improve efficiency. By implementing Wiz.ai talkbots to handle repetitive rule-based tasks, our unique deployment of voice AI technology allows businesses to position their employees to more high-value priority tasks while the bots take the time-consuming service calls.
In a typical case, when you call in to a customer service centre, agents first capture relevant information during a call, triggering a series of rule-based questions like, “Hi, what’s your name?” or, “What’re the last four digits of your bank card number?”. They then perform robot-like actions on the backend to transfer data, retrieve data from legacy systems, offer templated replies, or transfer the call to other relevant departments.
The whole process of verifying the customer’s identity and retrieving their data from the system, till the point where customers can actually make their query takes a very long time. Imagine that being repeated a hundred times a day for that single call centre employee.
And so what WIZ.AI does is we streamline all these frontend parts – we get all the identifiers, we automate the rule-based tasks, templated replies and if needed, transfer the caller to the relevant human counterpart. This reduces the time the customer spends on hold and frees up the call centre to focus on more high-value tasks.
Another thing that we help companies with is giving them access to a whole new set of customer data. Usually it is only possible to track the customer journey online, however we provide companies with rich customer data through our client portal, which automatically records and categorises intention data from the call. This comprehensive depth of data allows companies to identify high-value customers, create a more comprehensive customer persona, and uncover any gaps in their customer experience easily.
How sophisticated can AI talkbots get? Can it ever achieve human-like intelligence?
It’s true that AI is a game changer, however, the dream to create an AI with human-like intelligence is not as simple as it seems.
The way humans think and translate words into concepts is extremely complex. Having an AI understand basic words is a relatively simple process, but, once you add slang or colloquial phrases into the mix, or even just words with more than one meaning, for example, “right”, things can get exponentially more complex.
We have to look at what kind of tasks make sense for an AI and where a human would be better suited to perform that task. Generally, we want an AI efficient at performing basic rule-based tasks quickly and consistently.
Our talkbots are designed to appear more human-like, they are crafted to imitate human speech patterns to give a realistic human-to-human conversational experience. It has features such as natural pauses, the ability to handle interruptions, continuous listening, use of chasing and clarification statements, and the ability to understand the customer’s intent from their speech.
To achieve this, the Wiz Talkbot uses multiple proprietary deep technologies – automatic speech recognition (ASR), natural language understanding (NLU), real-time text-to-speech (TTS), multi-round dialogue technology. Furthermore, our proprietary natural language understanding model has an extensive library, which is robust, customisable, and adaptable, able to quickly learn new domain knowledge from scratch.
For example, the bot we built for one of our clients in the banking industry, is used to call customers with an overdue credit card bill reminder. It calls the customer, gently informs the customer of the overdue bill and asks if they have made payment. If they have not made payment, it follows up with when the customer expects to make payment and recommends some payment options, it is even able to negotiate for the customer to pay a smaller amount if they are unable to pay the full sum.
This scenario may seem simple but in fact, has a complex multi-layer dialogue management framework which is capable of learning and evolving from every interaction to grow in sophistication.
So far we’ve talked about the benefits of AI talkbots. Are there any dangers we should be aware of?
As with all technology, there is always the risk that it might be used for nefarious purposes. Take Google Maps for example. Used correctly, it can help save countless hours giving you the quickest route to your target location. However, in the wrong hands, this technology could potentially be used to coordinate attacks.
One area I see which might be prone to abuse are scam calls, which I’m sure you have received in the past in one form or another. In America, tax scammers impersonate IRS officials to contact people demanding payment or to steal personal financial information. They do this through the phone, e-mail, text message and even through social media.
In 2019 alone, the FBI received almost 300,000 complaints to the total of US$1.3 billion, this number they estimate is only 15 per cent of the total.
Now, if we assume that one human can contact a hundred individuals a day, a bot caller can easily make 3-4 times that number. They follow their programming until they are told to stop – they are efficient, relentless and tireless. And what’s so scary about this is that they are able to be spun up exponentially with the only limiting factor being the number of phone lines the scammers have.
And this whole thing becomes just a numbers game. It basically boils down to the number of people you can call until you reach someone who is actually going to believe the caller.
Then again, conversational AI is not a new technology – the first-ever chatbot called Eliza was developed in the 1960s. The important part lies in how well you can refine the technology, which is essentially our biggest differentiator from other mass-produced bots. Compared to our handcrafted AI bots, their technology is a far cry and should not pose any real trouble.
Is there a way to definitively prevent people from abusing the tech?
WIZ.AI is extremely selective about who we work with and what they would use our technology for. We always ensure that we do our due diligence when taking on new clients. Each talkbot is also protected with multiple layers of encryption and security.
As the creators of our talkbots, we are morally obligated to be the gatekeepers of it. However, as much as it frustrates us, as long as there are people with ill intent, they might use voice AI technology for the wrong reasons. The only saving grace is that mass-produced bots are easy to spot and producing a multi-dimensional talkbot requires a whole lot more information than just a couple of voice recordings. There is a lot of technology, AI training, and artistry behind the scenes in order to achieve the level of sophistication seen in our talkbots.
The only way to effectively prevent people from abusing the technology is to implement regulations that protects the customers. Similar to how the personal data protection act gives individuals control as to how their personal data is being used.
The “guns don’t kill people, people do” rhetoric is always an all too easy fallback for people who shun ethical concerns regarding advanced tools. Technology and AI communities will definitely be better off with responsible inventors and creators like WIZ.AI who use their technology to do good.
While the bulk of us may not be the decision-makers, raising such ethical concerns are just as important. After all, we are each one of us a crucial stakeholder in this high-stakes game of humans versus AI.