Conversational AI Poised to Be Major Disrupter
Chatbots and conversational AI systems got an extended tryout during COVID as companies scrambled for ways to keep their operations running amid lockdowns. The technology fared better than expected, and now is on the cusp of a major breakout in 2023 as companies look to build on those accomplishments and reach new heights in office automation.
Hayley Sutherland, who tracks the market for conversational AI tools and technologies as a research manager for IDC, says most conversational AI deployments were in pilot or proof-of-concept phases before 2020. But thanks to COVID, which provided an “extreme test case,” companies found success in their conversational AI deployments.
“I think that conversational AI has really earned this place in the last couple of years as a mainstream business application,” she tells Datanami. “Conversational AI has really become business-ready.”
The market has grown quickly, with hundreds of vendors developing a variety of tools, technologies and platforms for everything from first-generation chatbots all the way up to the most sophisticated conversational AI systems. Thousands of successful deployments over the past few years have shown that conversational AI can deliver 24/7 service, as well as a positive financial ROI.
We’ve come very far from the early days of rules-based chatbots, which proved frustrating to many people, Sutherland says. The advent of large language models, such as BERT (open sourced by Google) and OpenAI’s GPT-3, as the core of conversational AI deployments has been a critical factor in their success, she says.
“As technology matures…you have some people who are just realizing, wow, this is smarter than I thought it was, this is better than I thought it would be,” Sutherland says. “A big part of that has been the advances in deep learning and machine learning in these foundational large language models and the open sourcing of those large language models from some of the big vendors and the big research teams in the world.”
Large language models underly many conversational AI deployments today, but there’s a wide variety of other tools and capabilities that enable companies to deliver a finished product. Before the pandemic, developing a conversational AI system would have required a large team of developers to make it work.
But since then, a slew of low-code and no-code conversational AI platforms has emerged that can be effective in helping companies get started with conversational AI without making big investment in highly skilled data scientists, Sutherland says. However, that doesn’t mean companies can successfully deploy converstaional AI without any skilled individuals.
“Increasingly, conversational AI vendors are coming out with these tools that recognize that the teams that build conversational AI applications, or that are building successful ones, maybe they include data scientists,” Sutherland says. “But even if they do, they also need to include a line of business people who understand what does the good conversation look like? What information will the bot need to know in order to answer questions?
“I think even without data scientists,” she continues, “there are tools out there that organizations can use to get conversational AI quickly up and running.”
This is a great time to invest in conversational AI, as companies have many options available to them. However, it’s important to realize there is no one-size-fits-all solution, Sutherland says, and what works for one organization may not work for another.
The first step in figuring out what path to take is gauging the level of available data science talent at the company, Sutherland says. If companies want to build the whole conversational AI system themselves, they may need a different level of talent versus companies that choose to partner with a vendor to develop the application. The company’s industry also impacts the availability of pre-built templates that can jumpstart a project, she says.
“I think those are things to consider, because there is a variety out there,” Sutherland continues. “Some platforms will provide all kinds of testing and monitoring capabilities, which might be better for an organization that is really developer-heavy. Others might focus more on those low- and no-code tools and making them well-integrated into the overall business workflow.”
The availability of training data is also big differentiator in conversational AI. Some vendors may bring training data and have pre-trained models available for specific industries, while in other cases, the customer will need to bring their own training data to tune the large language model to work in their specific industry, Sutherland says.
“Low- and no-code tools, in combination with pre trained models–which some vendors are offering, that are essentially pre-trained for certain industries–can provide those quick starting points for organizations with greater accuracy out of the gate without necessarily having to hire a whole team of data scientists or even one data scientist,” Sutherland says.
There’s a huge swath of potential use cases for conversational AI; it’s not limited to just replacing or augmenting human customer service representatives. That variety in use cases, and the specificity of the industry, will impact how much additional training and tuning will be needed.
For example, a biotech firm that’s developing a conversational AI system to assist with the development of novel compounds will likely much more specific data than, say, a mattress store would need, Sutherland says.
“I think to get to a certain level of accuracy, there is always going to be some level of tuning,” she says. “I think the question is how automated is that, and to what extent does an organization to do that in house versus having the vendor assist with it.”
Another thing that companies should be aware of is that some conversational AI platforms work well for voice and can be hooked into the integrated voice response (IVR) systems that human agents use, while others platforms are designed for digital channels, such as Web and mobile; some can also do both.
As conversational AI spreads, new questions and challenge have emerged. One of those is determining what metrics to use to gauge the success of the conversational AI and the impact that it’s having on the business, Sutherland says. Conversational AI is also hungry for computational power, especially some of the latest large language models, so enabling companies of average means to partake in the fruits of this AI form is another concern.
As adoption of conversational AI spreads and companies become more aware of its benefits and limitations, finding the right balance between AI and the humans will become more critical, Sutherland says.
“I can’t necessarily replace every single human I have with AI. That’s not going to solve all my problems. In fact, it might create new ones,” she says. “So in fact the best way to leverage AI may be an understanding how can we use it to augment human workers and how can we leverage that so that AI is doing what AI does best and humans are doing what humans do best. And I think that is a balance that we’re really seeing organizations start to come through with, especially in the last year as they wrestle with what’s being called the Great Resignation.”
There are some conversational AI platform vendors that provide only technology, while others intentionally bring human operators into the mix. The conversational AI firm Simplr, for example, can help a company get started with conversational AI while also bringing a team of human agents to assist with customer service. The company’s goal with its “human cloud” is to provide real-time training for the conversational AI algorithms while simultaneously leveraging the automatically generated AI insights to elevate the human operator’s ability to deliver good customer service.
“I think that it’s going to be really important to look at where does AI work best? Where does a human work best? Sutherland says. “And I think that’s something that that Simplr has the potential to bring to its customers, through the combination of providing AI and human services as well as augmenting their own human network with AI.”
Related Items:
Google Debuts LaMDA 2 Conversational AI System and AI Test Kitchen
AI Is Not Sentient Yet. But That Doesn’t Mean It’s Not Useful in the Enterprise
A Data Platform for Chatbot Development