Four Reasons Why You Can’t Trust ChatGPT With Your Customers
Pity our friend ChatGPT — he is misunderstood and destined to disappoint every C-suite executive who sees him as a magic force that eliminates the need for human agents in their contact center. He is a powerful fellow that will have an outsized impact on the contact center, but not yet and not on his own.
Meet ChatGPT
ChatGPT is becoming like Zoom or Kleenex: a product name that describes its whole category. But ChatGPT is a product. It is made by a company called OpenAI. The company uses large language models to provide two very exciting capabilities: DALL·E is an application that creates images based on prompts that describe whatever picture is desired, and ChatGPT is an application that writes text based on prompts. These are generative AI applications that perform astonishing feats by combining large language models with other techniques to create new things. There are two things that OpenAI did to make its offerings unique:
- Get “smart” by building the largest large language model. A chatbot uses a language model to understand what is being said. A large language model is as good as its name: lots of data. In the realm of language, the more data you feed AI, the better able it is to discover semantic connections between different concepts, which makes it “smarter.” For example, a large language model can understand that a phone stored in a glove box is not visible from outside of a car. This is significant if the insurance company does not cover a stolen phone that was left in plain sight. It’s a simple example, but it implies an impressive understanding of the relationship between things. ChatGPT’s language model is notable for its size and breadth. Open AI scraped massive amounts of the public-facing internet late in 2021 to create what may be the largest language model to date. The massive scale of this language model allows ChatGPT to make more connections on more topics. It can answer questions about everything from pre-Columbian ball courts in Teotihuacan and their archeological significance to the messy implosion of the Sex Pistols in 1978.
- Use that model to create a powerful generative AI algorithm. Generative AI is used to create new content expressed in different ways, such as text, image, or video. OpenAI uses a variety of techniques to fine-tune ChatGPT’s language model to generate useful text. The result is remarkable: ChatGPT can credibly write college essays, computer code, or pithy tweets. If you are building a chatbot, it sure is tempting to let ChatGPT loose and answer all questions for your brand. Before you get too excited about this possibility, though, read on, please.
On a lark, I asked OpenAI’s image-generating offering, DALL·E, to paint me a picture of its sibling, ChatGPT — here is what I asked: “1700s-style portrait painting of ChatGPT.” And here are the results:
All the above look like men to me. Ergo, I’m labeling ChatGPT he/him. This is also indicative of one of the four reasons why you can’t trust him alone with your customers.
Four Reasons Why You Can’t Trust ChatGPT With Your Customers
- ChatGPT is biased. As I already mentioned, the four pictures above all look a lot like men. Are DALL·E and ChatGPT sexist? Not really, but my speculation is that, when DALL·E looked at portraits from the 1700s, there were a lot more men than women, so the odds of getting a man generated is much higher than getting a woman.
- ChatGPT lies. Not intentionally, but he is easily confused, and he doesn’t let on. Several analysts at Forrester asked ChatGPT to construct a biography for them. A bio for Max Ball came out fine, a clean and concise summary of what Forrester says about me on our website. Will White, on the other hand, was a mix of his bio and who knows how many other Will Whites.
- ChatGPT doesn’t do CRM. ChatGPT doesn’t know anything about the data from your back-end systems. While OpenAI is building tools to connect and extend ChatGPT, your brand will best be served by a traditional chatbot or contact-center-as-a-service provider acting as an intermediary between ChatGPT and your back-end systems, at least in the short term.
- ChatGPT is out of date. ChatGPT does not know that my contact center analyst partner, Christina McAllister, works at Forrester — remember, it last scanned the internet in 2021, before she started working here.
In The End, ChatGPT Is Just A Tool
Companies are flocking to build an infrastructure around large language models and ChatGPT to cover up his faults. Microsoft, which made a significant investment in OpenAI, already has a Bing front end that works with dynamic information to address the “out of date” issue. There will be a lot of activity in the contact center space to patch the holes in ChatGPT to deliver appropriate customer service solutions. ChatGPT clearly promises a new level of intelligence that will allow conversational AI to become ever more valuable to brands and their customers, but there is a lot of infrastructure that needs to be built for this to make sense. We’ll see how long that takes. Even then, our C-suite executive’s dream of eliminating the need for agents will have to wait — they are not going away anytime soon.