In conversation with ChatGPT: What do we really want out of chatbots?

While the chatbots from six or seven years ago might be like “embryos,” today’s AI chat tools are more like a “really smart college student,” O3 World cofounder Mike Gadsby said.

Years ago, you had to teach chatbots how to do everything, but today’s are quite smart. With all the information available on the internet, natural language processing tools can gather research for a presentation, write emails, code and even pen your wedding vows.

You might think of them like a virtual assistant, Gadsby said Monday during the user experience and strategy firm’s Innovation Day. Throughout the event, which took place during Philly Tech Week 2023 presented by Comcast, Gadsby and senior engineer Josh Friedman explored the evolution of natural language processing tools over the last few years.

Gadsby started with a real-life example. In 2017, he and his wife worked with Lowe’s employees on a kitchen renovation. Over a few months, they had multiple meetings with the Lowe’s team answering questions about the size, budget and design goals they had. It was a “very offline” experience, he said, but they worked with the same three or so employees over time, who had the context and capabilities to guide them through the project.

Now, if they were to get started on that project today, they might head to Lowes.com, where they’d find a pretty standard retail chatbot. Gadsby modeled to the audience how that conversation went: He explained he wanted to work on a kitchen renovation, and the chatbot served him a security warning, telling him to not give it personal information. After a few standardized questions, it connected him with a “live agent” who took a few minutes to review the conversation so far and decide he ultimately wasn’t specialized enough to serve the information Gadsby was looking for. The agent instead provided a phone number to the nearest Lowes. Gadsby still wasn’t really past step one of his project after this interaction.

Humans aren’t really looking for the technology to predict the future, but they are looking to it as a way of summarizing ideas and taking a next step in the experience.

The introduction of OpenAI’s ChatGPT tool late last year has opened up a lot of digital doors, the exec said. And it hones in on some of the key things humans want when they interact with chatbots, starting with personalization. Today’s natural language processing tools can be used to offer up very specific suggestions, based on prior knowledge about you.

“So saying, ‘Hey Mike, because you bought XYZ, and you like these things, these are other things that you will also like,'” Gadsby said, “that level of hyperpersonalization is available to us, and it’s one that we’re really exploring in a major way.”

Humans are also looking for more conversation-like interactions with their natural language processors as they make their way though a website or app, as well as predictions. Humans aren’t really looking for the technology to actually predict the future, Gadsby said, but they are looking to it as a way of summarizing ideas and taking a next step in the experience.

And lastly, users want automation or streamlining of processes if they’re interacting with a chatbot: It should be sending you specific links of where you’ll want to go on the site, or help you fill out applications or forms.

To test how the Lowe’s chatbot feature would work if it used ChatGPT integration, Gadsby and Friedman built a version they called Lowe’s Kitchen Assistant in less than a day to demo. With the audience, the pair started a similar conversation about a kitchen renovation. This tool, though, displayed the characteristics — personalization, conversation, prediction and automation — of an ideal chatbot. Check out a snippet of the conversation:

Had a fun little conversation with #ChatGPT at @o3world’s innovation day today! #ptw2023 pic.twitter.com/v0nLEvJAcX

— Paige Gross ✨ (@By_paigegross) May 8, 2023

The bot affirms the style choices we made and even offers some suggestions about complementary “cozy” flooring options. It ends the conversation by noting the client’s availability and location, and says they’ll be reaching out to set up an appointment on Saturday.

“You can see the difference in terms of detail and service direction, but it’s not so stiff,” Friedman said.

There’s lots of ways to take these tech tools, the O3 World technologists said. They’re both very aware of the security and privacy concerns raised about AI, and keep them in the back of their minds as they build integrations for clients. But concerns that it will lead to a dystopian society where humans don’t talk to each other or abuse machines doesn’t seem as likely to Gadsby.

In a conversation with his 14-year-old son, he heard that the teenager doesn’t care if he’s talking to a human or a robot in these types of interactions, he just cares that it feels like a conversation.

“Now we sort of engage with technology and we sort of expect, you know, that human interaction treated with human respect,” Gadsby said. “And it’s such a beautiful thing to me that, I don’t know that really matters, at the end of the day. But as far as what we’ve seen, what people like is that they really want to say please, and thank you. They’re having real conversations, because I think the responses [by AI] are such that it makes them feel that way.”