ChatGPT: A Cautious Field Guide

Within just a few months, ChatGPT, the artificial-intelligence-driven chatbot launched by OpenAI in November, has gone mainstream. Though it’s essentially a prototype, the bot has the attention not only of Big Tech, but its competitive capital, too. On Monday, Microsoft announced its “multibillion dollar investment” in OpenAI, and now Google is bringing back founders Larry Page and Sergey Brin to help jump-start the company’s AI development. The ironic part is that Google actually pioneered some of the core technologies that helped OpenAI to create ChatGPT, which has the potential to disrupt its search engine business. 

This seemingly simple interface, run by “generative pre-trained transformers” (GPT), represents a huge leap forward in natural language processing (NLP). Without getting too deep into the technical details, this means AI-driven computers essentially have the ability to “understand” language in a remarkably humanlike way.

This brave new future is already here. This tech is going to show up as a core feature of new startups and as enhancements to many tools and products we use every day. Microsoft is already looking for ways to integrate it directly into its Office software and its Azure cloud computing platform, for example.

But only fools rush in. Like any new medium or technology, there are limitations and potential risks involved in using it.

Wait… Is This Safe?

One of the big limitations of ChatGPT is potential biases or inaccuracies in the training data. It’s a very large set, at around 800 gigabytes of data (in version 3), but even at that scale there’s a fair amount of room to get things wrong. Also, it’s important to note that OpenAI makes no guarantee that its chatbot’s answers will be factually accurate, and in many cases the responses are misleading. It’s better at creative, open-ended questions than highly specific questions with exact answers. For instance, CNET is reported to have used AI to write a series of explainer articles on its website, and the information about how interest rates for CDs work was incorrect.

Training data is another factor. These large sets of data are coming from various sources, and one of those sources is scraping content found on the internet, which is legally sound for search engines, but an open question for AI training models. Indexing information and then linking to the source material is very different from consuming source material and generating something new. Ownership and intellectual property rights are a bit fuzzy, as the idea of fair use may or may not apply to the outputs of these technologies. 

Currently, there’s a legal challenge brewing for Midjourney, Stability AI and DeviantArt brought forth by a group of artists who believe that data scraping for training violates copyright law. Even though these technologies are different from ChatGPT, given that they are generating images, the process by which they consume large sets of public work by humans has similar legal and ethical implications. The outcome of the legal battles of one will surely impact the other, so it will be important to see how these products approach fair use — something to be cautious about when creating generative AI-enhanced work.

Walk Before You Run

At this point, you shouldn’t necessarily avoid these tools, but step one for adopting them into your business is setting policy guidelines, no matter how you choose to pilot a program.

First, be as transparent as possible with all parties involved that you are using the technology, and make sure everyone is aware and on board with its use. Next, be considerate of intellectual property (IP). Don’t use brand names or personal names or likenesses (e.g., Barbie, Taylor Swift, Stephen King), as the results could lead to infringement. Also, make sure whatever comes back as a response doesn’t directly reflect or feel overly derivative of IP or someone’s likeness. (If it feels wrong or too similar, don’t use it.) 

Of course, you might want to consider just employing a human. There is still no replacement for authenticity. For now, the tech is not anywhere as good as a talented writer; so far, it’s just earning a passing grade. Yet it’s easy to see how a utility like this will find its way into mainstream writing tools in the near future. Longer term, the big advantage will be in helping everyone reduce or eliminate the need to write anything formulaic or rote. New tools and systems will allow for a lot of this work to be automatically generated.

Where This Is (Most Likely) Headed

No one knows what ChatGPT will become, but some applications of its technology seem inevitable. 

Customer service call centers, for example, have been trying for years to fully automate, but often lean on outdated chatbots or “smart” phone systems in an effort to save labor costs. A few platforms have already announced intentions to integrate OpenAI or a similar technology into their product offering. The e-commerce company Shopify also has a bot using the same underlying technology as ChatGPT as part of its platform.

The search engine business is another target. Generative AI will likely transition search from providing links based on a topic to just answering the questions itself. This will change search engine monetization and the entire SEO industry as these new tools shift from driving consumers to external websites to one where the AI gives them everything they need without having to go elsewhere. The engine becomes the place where more of the experience happens.

This may result in a big shift in how we interact with technology. It may actually deliver on the experience people originally expected from assistant technologies like Alexa and Siri.

In the not too distant future, open text prompts might be the first way into many apps and websites that originally forced users to navigate through various menus and a maze of user-interface elements. Completing complex multistep interactions could turn into a simpler chat or voice-based experience.

What if you could get all your taxes filed over a simple chat with a bot that does the hard work? One could even imagine a future where legal contracts are negotiated by competing chat AIs and mediated in nanoseconds. The possibilities really seem endless.

Though I can’t predict exactly how this will play out, I’m betting this technology is going to find its way into everyday life over the next several years. It’s going to fundamentally redefine table stakes in our digital transactions — and ultimately shape our digital lives.

Jon Hackett is the group technology director for transformative technology at Huge.