In the last few months, the major news outlets have been whipping themselves into a frenzy over the current and potential future implications of artificial intelligence (“AI”).
Time Magazine warns that comedians are worried that AI is crafting jokes that are a just little too clever. AI has been shown to outperform doctors in making diagnoses. AI can write wedding vows– and later encourage a spouse to abandon their marriage. And most disturbing of all, a writer for the New York Times was so “deeply unsettled” by his conversation with a chatbot that he had trouble sleeping.
While greater minds can ponder those broader inquiries, I have but one question: can AI replace the role of the travel blog or travel writer? Are travel writers doomed, soon to be swapped out by more efficient, more intelligent machines?
First, for those less familiar with AI, let’s start with simple a definition:
Artificial intelligence (AI) involves using computers to do things that traditionally require human intelligence. AI can process large amounts of data in ways that humans cannot. The goal for AI is to be able to do things like recognize patterns, make decisions, and judge like humans.Aruna Pattam, Artificial Intelligence in Simple Terms
To test whether AI could in fact create a helpful travel blog post, I sampled the new-ish tool everyone is buzzing about, ChatGPT. (Read the resulting blog post I wrote using ChatGPT, the Top Souvenirs from all 50 States, here.)
My process for Using ChatGPT AI to Create a Travel Blog Post
Naturally I had my biases going into the experiment– I questioned whether AI could provide specific enough content to actually be useful, or whether it would just spit out generic, unhelpful results.
Step One: Select a Suitable Topic
My first step was to decide on a topic– I decided to ask ChatGPT the following:
What specific souvenir-related topic is better suited to have chatGPT write an article about rather than a human?
ChatGPT indeed felt it was up to the task, promising that its “ability to gather, analyze, and present large amounts of data and information from multiple sources makes it well-suited” to write the blog post.
Of the four options ChatGPT provided, I decided to go with ChatGPT ‘s offer to “gather information about the most popular souvenirs in different destinations, drawing upon data from tourist guides, travel blogs, and social media.”
I went a bit beyond ChatGPT’s generic recommendation for “popular souvenirs in different destinations” and decided to get more specific and use a topic that would be difficult and time consuming for me to compile and create on my own– the best souvenirs from all 50 U.S. states.
Step Two: Create Relevant ChatGPT Prompts to Deliver Helpful Content
I quickly learned that getting ChatGPT to create this 50 state souvenir list was not a “one and done” deal. I decided limiting the result to just one souvenir per state was just too restrictive. I settled on a minimum of five souvenirs per state.
My first ChatGPT prompt attempt left me with vague results that no one would be interested in (like “souvenir pottery” from Indiana, for example). With each subsequent ChatGPT search I learned to refine my technique.
- To force more specific responses, I asked ChatGPT to name a shop where each souvenir could be purchased.
- When ChatGPT starting recommending things like Butter Burgers, a diner staple in Wisconsin, I added “and not perishable and can be easily transported” which helped somewhat.
- When ChatGPT recommended University items from every state’s schools, and sports items from every state’s sports teams, I tried to create prompts to exclude those categories of items.
- I attempted to force ChatGPT to make judgments about the souvenirs (nothing tacky for example) but it claimed it could not make such human distinctions. I tried everything, even asking that it “pretend it was human” to no avail.
As a former practicing attorney, I was experienced in creating searches in online legal databases to retrieve relevant legal materials. Creating prompts on ChatGPT was not that much different and definitely more fun.
Step Three: Weed, Vet and Curate the ChatGPT content
After running multiple searches on ChatGPT, I ended up with a lot of results– sometimes 250 suggested souvenirs per prompt. There was no one “perfect” search, I ended up using results from multiple searches. These search results sometimes overlapped and other times were wildly different, even if the prompt was the same or very similar.
While ChatGPT did a great job in amassing a huge amount of information, it could not sort through and vet that information into a usable blog post– that’s where the human element as the curator comes in.
As I am pretty familiar with souvenirs, I was able to eyeball the list for obvious errors, which were surprisingly few.
I next eliminated all the generic souvenirs that could be found in any state– according to ChatGPT nearly every state has a “Ye Olde Fudge Shop,” and souvenir worthy handmade soaps, glass and pottery. I didn’t exclude every shop though– this step required more vetting in case a particular state was very well known for its pottery or glass, for example.
I then chose five souvenirs from the ChatGPT data sets for each of the fifty states. This step involved a lot more vetting. I spent hours googling suggested souvenirs to see if they actually exist and if they were a “thing.”
I then again used my human skills to curate the remaining results. I used criteria like uniqueness, interest level, relevance to the state, cultural traditions and diversity of items to cull to the final lists of five per state. If there were more than five souvenir items for a particular state and I was torn on which to choose (since I often did not have the personal experience to make subtle judgments), I threw the extra options under “honorable mentions.”
Googling the less familiar souvenir items could have possibly been a shorter process, but my passion for the topic often led me down a rabbit hole going far beyond the parameters of the post. I can’t help that I have an insatiable curiosity and Iove learning about new topics. For example, when I researched an unfamiliar Nevada souvenir, basque chorizo, I learned that Nevada has a sizable Basque population that came over during the gold rush. That led to me researching the history of Basque settlement in Nevada.
My curiosity got the best of me and I repeated this time and again. Do I now know the difference between shoefly and gooey butter pies? Yes, and I have recipes I plan to try too.
Step Four: the Final Result, How did we do?
Now onto the results, how did the blog post come out? In my opinion, pretty damn good for what it is.
The post provides a birds eye perspective of souvenirs across America all in one compact post. While there is no context (adding it would have made the post too long and further challenged ChatGPT’s limits) I think the spare list is fun to read and draw comparisons across states.
Are these the best souvenirs for every state? Definitely not. Some states are better than others (New York was awful, but using all my restraint I did not add my own suggestions). And while I tried to get ChatGPT to generate images, they all had error messages, so I added my own photos for visual interest to break up the long text.
But taken as a whole, the ChatGPT souvenir suggestions aren’t bad, and more importantly they are a starting point for additional research– which brings me to what I think in its current state AI and ChatGPT are good for.
From my use of ChatGPT at least, I don’t think AI can yet tell the story of a place. When I write about souvenirs, it’s never about the thing alone– it’s a way bring home the feeling you have when you travel. AI also cannot yet inspire an interactive experience. Some of my blog posts have dozens of comments from readers who whether they agree or disagree contribute their own insight, creating a communal experince.
Maybe someday ChatGPT can fake that, but until then travel bloggers are safe from AI forever replacing the travel blog.
Tell us what you think of AI and ChatGPT in the comments below!