Even though physical evidence might point in the opposite direction, AI chatbots are not as effective as they may seem. Here’s why
If a food delivery order fails, you can ask a chatbot to issue a refund. If you want to reserve a table for a date night, a chatbot can help you with that. If a potential partner turns you down, a chatbot can provide emotional support. From your banking needs to healthcare, customer service and more, chatbots are everywhere. Using natural language processing and sentiment analysis, chatbots—artificial intelligence (AI)-enabled software—interact with humans or other chatbots.
They reduce the strain on humans and save companies money they would otherwise need to spend on multiple people. However, as the reputation of bots goes, chatbots aren’t all that effective and can annoy users (more so if they are already annoyed).
Here are some reasons chatbots have failed over time.
Failure to perform simple tasks
If a chatbot’s purpose is simply to refer the user to the human support team, is it really even necessary? That’s a question Swedish furniture giant IKEA had to grapple with after launching its AI chatbot Anna in 2008. When a customer wanted to return a bed, he reached out to IKEA customer support, emailed IKEA and spoke to Anna—and what an experience that was.
Anna, for one, didn’t understand the user’s question. Then, when asked to do something (anything at all) to assist the user, Anna told him to contact the company on call for assistance. Not only was Anna pointless, given that her immediate response was to refer the user to on-call support, but she was also useless and unable to perform a simple task.
Ultimately, the user took to writing an open letter to IKEA’s CEO on LinkedIn.
Lack of human support can prove detrimental to the business and frustrating for humans. A 2019 Forrester report found that over 50% of U.S. shoppers had a negative experience with chatbots. Due to their limited ability to understand humans, chatbots cannot function without actual human support if they want to be effective.
Lack of humanity
In 2016, Microsoft launched Tay, a Twitter chatbot built based on “relevant public data”. It was a machine-learning project designed to imitate humans online. It was the ultimate bid to make chatbots appear human. But gosh did it take a turn for the worse.
Based on what users tweeted at it, Tay made extremely controversial statements, ranging from “I f*cking hate feminists” to “Hitler was right”. Leaving chatbots to their own devices can be super problematic and inefficient. It needs humans at the helm to filter responses and ensure they are not controversial.
Chatbots are ultimately programmed solutions. They run based on coded responses, and they can only do so much with it. They lack the emotional quotient humans have to be considered alternatives to customer support.
Inability to understand human input
When a user asked Meta’s Poncho AI about the weather in Brooklyn during an upcoming weekend, the AI chatbot asked if the user was on a boat. Then, when prompted again, it gave the real-time weather of Brooklyn. Finally, the user yelled (i.e. used all caps) “WEEKEND” at the chatbot, which responded nonchalantly with “Sorry, dozed off for a second. What were you saying?”
It is challenging, though not impossible, for AI chatbots to understand anything off-script. A user could be trying to report a poor product, and the AI would just be concerned with whether they received the product at all. It’s like conversing with a gaslighting partner. You want to throw something against the wall after a point.
Trust and privacy concerns
Research has found that people simply do not trust bots. Even though companies attempt to make chatbots conversational and give them a humanlike quality, it is difficult to gain users’ trust. Experts feel that chatbots lack a contextual awareness of what they say. Ultimately, they are just programmed systems designed to persuade users.
This programming often works, convincing users to converse with chatbots as if they were humans. However, because of that, users might end up sharing personal information that could put them at risk of privacy breaches.
Business-focused instead of user-focused
Finally, managers’ misplaced priorities can also result in chatbot failures. Typically, chatbots are created not to make the customer journey easier but to simplify business processes. Businesses want to reduce costs, save time and reduce redundancies within. So, they create chatbots.
However, chatbots’ inability to understand human input or perform tasks without human assistance defeats that purpose. Businesses do not consider what users want or the complexities involved in assisting actual people, which is often where chatbots fall short.
Will ChatGPT fail or succeed?
All hope is not lost with chatbots. OpenAI is promising to consider these abovementioned flaws and create a new chatbot—ChatGPT—that can, as per its website, “answer follow-up questions, admit its mistakes, challenge incorrect premises, and reject inappropriate requests”. It takes conversational AI up a notch and claims to be actually effective. It can write essays, tell jokes (even if they aren’t to everyone’s taste) and play games with you. Currently, it has mixed reviews, with some calling it “overhyped” and others saying that it will “kill Google”.
While many are impressed by how effectively it can imitate humans, the above examples show that that can have its downfalls. The success of ChatGPT rests in its code and its ability to gain users’ trust.
Whether it will meet its inevitable downfall—as many chatbots before it did—or actually thrive is yet to be discovered. If it’s the latter, it could be a milestone for AI as we know it.
Also read:
- Why Are Bots Taking over the Internet?
- Chatbots: Friends, Companions, And Assistants Of The Future
- Next-Generation AI May Look Like Us
Header Image by Freepik
