The computers that talk like us: How conversational AI could change lives, for better and worse

Artificial intelligence India
(Image credit: Google)

Off the back of advances in compute performance, data management and software design, artificial intelligence (AI) has come a long way over the last few years and is now being deployed across all manner of industries

One particular subdiscipline, conversational AI, is paving the way for systems capable of holding a discussion at a human-like level, opening UP various doors in fields from customer services to sales and marketing.

However, when it's no longer possible to distinguish between man and machine, a whole host of issues are bound to crop up, especially if AI models have not been developed and audited in a responsible manner.

To hear more about the opportunities associated with conversational AI, as well as the potential pitfalls, we spoke to Dinesh Nirmal, head of data, AI and automation at IBM.

How would you define conversational AI for the layman?

Conversational AI refers to the different types of AI software or solutions that are designed for people to talk to and communicate with. We train conversational AI using vast amounts of input data, essentially language and phrases that teach it how to recognize words and imitate human interactions. In a business context, this is useful in many different situations, especially customer care. With conversational AI, we can train virtual assistants to help customers solve common problems, more quickly find insights in their documents, or automate repetitive tasks using natural language commands.

Natural language processing (NLP) is one of the techniques we use, along with machine learning and sometimes other forms of AI as well, to train conversational AI. It is focused on teaching computers to analyze language. A robust conversational AI that’s in the field and helping customers resolve their issues, buy products or execute tasks is the end result of combining NLP with other forms of AI that put that comprehension to work. 

At IBM, we’ve developed several different types of conversational AI solutions to help businesses deliver better customer care experiences and help their employees be more productive. For example, IBM Watson Assistant, which is our core conversational AI solution, automates engagement with customers to accelerate the resolution of their issues or answer questions that might be buried in hard-to-read FAQs and webpages. It uses machine learning and natural language processing to understand customers, in the appropriate context, and to provide fast, consistent, and accurate answers across any application, device, or channel. We’re also developing conversational AI solutions that are being put to work for automated order taking in quick service restaurants as well as solutions for digital employees that allow your team members to use natural language to automate repetitive tasks.

What are the main opportunities associated with conversational AI, as you see it? Which industries in particular are ripe for disruption?

AI and automation are already having a growing impact on how brands manage customer and employee interactions. The business market for virtual assistants will grow 15 percent to more than $7 billion this year, according to a Gartner prediction. That growth is coming from companies looking to automate customer interactions, sure, but also increasingly in areas like advanced search and document insights, so that a user could say “pull up last year’s August sales” and an assistant could surface the relevant document.

In the long run, we believe that all enterprises across industries are poised to benefit from conversational AI solutions. This will only continue to grow as advancements in AI enable these systems to understand the unique language of their industry and business, automate tasks and extract even more precise insights from complex documents and data – without requiring sophisticated data science skills on the part of users. Additionally, advancements that enable companies to build deeper integrations and to break down internal data silos will result in better, more personalized experiences. 

When it comes to customer care, the Covid-19 pandemic accelerated adoption of conversational AI solutions across industries, with many early adopters in retail, healthcare and financial services. For example, to help their customers navigate the Covid-19 vaccine rollout process, CVS Health engaged IBM Consulting and began using IBM Watson Assistant to respond a deluge of common customer questions about eligibility, side effects, needed documentation, and more. This freed up their human agents to handle more complicated issues. This is an extreme example, but as we shift back to a new “normal”, we see conversational AI playing a massive role in ensuring information, direction, and overall engagement is streamlined between the brand and the consumer. 

Another area that we think is particularly ripe for disruption with conversational AI is in employee experience and productivity. Many organizations today are facing skills and labor shortages, and they are looking for ways to free up their employees to focus on higher value tasks. Conversational AI can play a key role in this space. Many employees in fields like HR, sales and marketing are constantly bogged down with repetitive tasks like scheduling meetings, updating calendars, requesting agendas, sending reminders, etc. With an intelligent automation solution that can be controlled using natural language, employees can automate these repetitive tasks and reclaim their time to focus on what matters most.

A digital face in profile against a digital background.

(Image credit: Shutterstock / Ryzhi)

What are the risks that biases make their way into NLP models that end up powering a wide range of services? What's the worst-case scenario here?

As AI adoption continues to grow, the stakes are high, and guardrails are needed to ensure we can trust AI systems and their outcomes. Technology must be transparent and explainable, and that means that businesses must be clear about who trains their AI systems, what data was used in training and, most importantly, what went into their algorithms’ recommendations. Whether you’re using AI to help solve customer challenges, screen prospective job candidates or to help streamline your IT environments, we all have a role to play to ensure that AI is explainable, fair, robust, transparent and respects consumer privacy.

Ensuring your AI systems aren’t perpetuating biases or causing unintended harm isn’t something that just happens when you’re first building models. Businesses need to make sure that they are establishing trust throughout every step of the AI lifecycle – from data collection and cataloguing to model development, observability, optimization and monitoring.

Bringing automation into your AI governance processes helps with this. Many problems can arise when organizations are using manual documentation and validation processes to govern models – not only is it slow, it often requires hundreds of pages of documents, can easily result in human errors or delays in getting key information to auditors, leading to penalties and fines.

To help with this, one innovation I’m excited about is our AI FactSheets, which are part of IBM Watson Studio on IBM Cloud Pak for Data. Think of AI FactSheets like a nutrition label for food. They are an idea that was born in IBM Research that we’ve made available to businesses so they can automatically capture metadata on their models and closely monitor for things like quality, fairness and drift. In addition to understanding your model history and enabling ongoing monitoring and management, organizations also need to ensure they have a diverse set of practitioners and skillsets supporting the development of conversational AI and NLP models.

It feels like NLP has taken huge strides as a discipline over the last couple of years. How quickly do you expect to reach a point at which humans can have a seamless conversation with AI?

Like any emerging technology, there is always an adoption curve. If a human can quickly solve a problem by using AI, then that’s a win for everyone. But the purpose of AI is to augment – not replace – human intelligence, so while the interactions will continue to get more seamless, the need for a human-in-the-loop will remain.

There is a lot of exciting work happening in the field of NLP that makes the experience more and more seamless. This includes advances that enable AI to achieve higher accuracy, from smaller data sets, as well as advancements so that systems can handle the complexity and nuance of human communication like misspellings or mispronunciations, idioms and industry specific phrases, topic changes and disambiguation. There have also been advances so that systems can support many different languages and dialects, without needing to be retrained every time they encounter something new. Advances in NLP models for document understanding help us uncover more precise insights from complex document types like PDFs, charts and tables, which results in more accurate answers ending up in the hands of end users. Ongoing research and development in this space will continue to result in even better experiences at a rapid pace.

Another factor that will lead to more seamless interactions are deeper integrations that enable companies to build systems that reflect the full context of the user’s journey. For example, if you discover a product on your bank’s website, fill out a form and go into your local bank branch bank to apply for that product, you’ll have a better customer service experience down the line if the virtual agent you end up interacting with has the context of your past interactions readily available. This will lead to more meaningful interactions and faster resolutions – regardless of what channel you are interacting on.

A developer writing code

(Image credit: Shutterstock / Elle Aon)

What do you think advances in fields like NLP mean for creatives in industries like script- and fiction-writing, or journalism?

I don't think AI is putting Steven Spielberg out of work any time soon, but I do think the next generation of creatives are extremely likely to be using AI more and more, if they aren’t already. In the long run, AI is going to transform the way we all work for the better, free people from tedious, non-value adding work and ultimately help create new types of jobs around the world.

I believe AI is ultimately going to transform virtually all jobs in some way, whether through virtual agents and assistants who help us with tasks, to employee assistance and training tools, to the better forecasting and prediction that AI allows. AI is designed to augment human potential and free up time from mind-numbing, creative tasks. That’s wonderful news for creatives, because it means more time that they can spend being creative.

How should we think about the dangers of advanced conversational AI, whereby it might be difficult to discern whether we’re having a conversation with another human or not?

The benefits that companies are gaining with conversational AI should not come at the expense of transparency. Even as the technology gets more and more sophisticated, we believe organizations should be up front about when, where, and how they are using AI.

Many organizations approach AI from the lens of a company problem that needs to be solved, when really we all need to be considering the human problem. At the end of the day, it’s the human who is interacting with the AI that we care about. At IBM, we call this approach “human-centered AI.” Building an effective, trustworthy conversational AI system requires asking yourself at the beginning: Who is going to be using this? How are they using it? Why are they using it? The goal shouldn’t be to design a system that can be mistaken for a human, it should be about how we can best use AI to augment human expertise, judgement, problem solving and decision-making.

It’s no secret that the benefits of conversational AI can be huge – from cost savings to productivity improvements. But we’ll only reap the benefits of conversational AI if society continues to trust it. That’s why trust and transparency are fundamental to AI innovation.