No ratings yet.

Although AI is able to generate good-looking images from a textual prompt, it is too sure of what it is creating – it is being asked to draw things correctly and it is intended to do so. Human drawings, on the other hand, do not have to be correct to be creative. My writing will begin with this conclusion.

I often wonder how these “beautiful” AI-generated images are created – I use quotation marks here to indicate that I tend to disagree with the adjective, because I often find these images to be mechanical, impersonal, unconscious patchworks. Surely there are methods of writing THE “spell” to produce an interesting work that can visually engage the viewer, such a process of finding the right prompt, however, later I learned through my own experimentation that it takes many trials, and hopefully sometimes one can find something that fits well with one’s imagination among ten or more works that AI has generated on request.

I tried to use the same prompt to ask for a film festival poster with a rabbit in the centre, using different image-generating AIs. This choice of rabbit animal had nothing to do with my personal preference, but with a poster design requirement I had recently heard in a friend’s task. The results I got are shown below, and I noticed that these AI-generated contents either got the textual information for the film festival wrong (also because I did not specify which film festival) or tried to depict a realistic animal regardless of style (I did not specify a style unless it was mandatory to produce the image).

To provide some examples of human-made posters as a control group, I simply typed “rabbit poster” into the search box on Pinterest, a social media service where people can store their inspiration in image form and display it in the form of pinboards. The contrast is already apparent here: being creative is more about “showing without showing exactly”, and the rabbit is sometimes seen only as lines or colors, sometimes the animal is indicated by the limited presentation of certain traits such as long ears and/or whiskers.

Unlike many popular ideas that put a question mark over AI’s potential to replace human creativity, I am on the side of asking why AI will not replace human creativity, although it could be argued that I have not fully experimented with everything that image-generating AI is capable of.

If we first consider the mechanism of generating visually appealing content, it has a lot to do with machine learning algorithms and training models with complex datasets. In order to provide the generated image, the algorithm first identifies and extracts patterns and structures from its datasets and then uses this information to create new content that matches with the previously defined characteristics (Reinecken, 2023). This process is data-based, which means that it is essentially a transfer and transformation of information, rather than the creation of something that has never been created and stored before. For this reason, a certain amount of “soul” is missing from the process of creation.

Human creativity, however, is based on the neuroscience of the human brain, which has a distinct ability to recognize and generate new ideas or possibilities while we are going through emotions and subjective experiences (Reinecken, 2023). Moreover, human creativity does not follow a single process of identifying logical patterns and structures and then extracting these fragments precisely from a data set that has been modeled and encapsulated as a static, closed, even dead circuit. No two human brains are supposed to think alike, because the process of thinking itself is bound up with cultural, social, and individual differences that provide ever-changing variables in the production of ideas and images. If one argues that AI can also operate on a dynamically changing model that integrates randomness every few minutes. Does that bring the AI close to a human brain in terms of creating new things? The answer is still no, because it is so rare that we human realize the role that our body plays in generating emotions and new ideas – our creativity. Simply think of how often ideas surface while we are walking or running, not to mention that emotions are a human response to the world around us.

That was perhaps too far from my argument that AI cannot produce images in the same way as humans, because the AI does not get the idea of “showing without showing exactly”. But my reflection on this question of the existence of creativity stands, and it actually points to the idea of “embodied creativity” (Davis, 2012), that we live and experience through the flesh of the body, and therefore human creation can only be seen as a product bound up with the participation of a human body.  In conclusion, the reason why AI will not replace humans in the creation of images goes back to the most fundamental difference between these two existences: while one exists only virtually, the human exists as a physical creature, therefore with the utmost capability for experience, reflection, love, renewal and becoming.

References:

Reinecken, E. (2023). Is generative AI replacing human creativity?. Retrieved from https://simpleshow.com/blog/generative-ai-replacing-human-creativity/

Davis, J. (2012). Science of Creativity Moves Into the Body. Retrieved from https://www.psychologytoday.com/intl/blog/tracking-wonder/201211/science-creativity-moves-the-body

No ratings yet.

The answer would be no, which is what I expected, because not only was I browsing information available on different websites, but my brain was constantly comparing options and making choices based on my travel experiences. Following some previous examples of using ChatGPT for travel planning (Beebom.com 2023, Halpern 2023), I experiment with two different generative AI models – ChatGPT (version 3.5) and Bard powered by Google. I see the characteristics and limitations of both AI models.

This year, I decide to make early plans for the coming holiday season before the price of flights and hotels skyrockets. Ever since I took on the role of “planner” for assignments and later at work, I know that I am good at making travel plans for myself and people who travel with me. My plans always make sure to include the major tourist attractions, shopping spots, nice restaurants, and cozy stays within the agreed budget. However, my perfectionism often slows me down when making such plans. For example, as well as looking at distances to attractions, I would also look at safety information before deciding where to stay.  Reading through low-star reviews to have a realistic expectation of the hotel is another habit of mine. I wonder if I could be assisted by generative AI, having learnt about many applications of this technology in the class.

Since the default version of ChatGPT-3.5 has very limited access to data after 2021, I also went to Bard for real-time answers. Since I already know that our destination is Prague, I typed in the first prompt – “I am planning to visit Prague. Tell me more about the city and what I can do there when I visit with some friends in early November 2023.”

I have already noticed that Bard, powered by Google Search, provides images and links to websites for each tourist attraction recommendation. It also searched for me some information about early November events. This is different from ChatGPT’s text-to-text (only) mode. I find the direct links very useful.

My next step is to plan our travel itinerary. The prompt for Bard and ChatGPT is – “Create an Itinerary for a 5-day trip to Prague. Make sure to include the most beautiful yet safe places for girls to explore. We expect a moderate amount of things to do and we are very interested in visiting nice cafes and vintage shops”. As I expected, Bard provides images and links, presumably from Google products such as the search engine and maps. The itinerary that Bard provides is rather basic, not mentioning much about the things that we show our interest in (visiting nice cafes and vintage shops) from the prompt. ChatGPT, however, clearly knows that this is our main interest while visiting Prague.

For ChatGPT, a follow-up question asks for the names of cafes and shops it would recommend, and the result is very different from Bard’s recommendation. I asked ChatGPT and Bard about their criteria for finding these cafes and shops and they gave similar answers about ratings, locations, and variety of choice. ChatGPT seems to pay more attention to letting us taste the local flavor.

Since I think accommodation and flights are much more expensive compared to other expenses in a trip, I moved directly to asking about accommodation and flight information.

“For accommodation, we are looking for options of no more than €50 per person per night for 4 of us. Keep the itinerary in mind and make sure it’s not too far from the things we want to do. You can also show me some low-star rating reviews for my reference along with any of your recommendations”.

It was at this point that the limitations of both AIs were made very clear in answering my questions: ChatGPT got the price wrong (Bard was relatively right about the price per person per night), presumably due to the AI’s limited access to 2023 data, and its recommendations are very limited to hostels. When I asked why they didn’t list women-only hostels as one of the top options, given that I mentioned “safe for girls” in the itinerary-generating prompt, and why they didn’t list any Booking.com properties, both AIs said that 1) they didn’t catch the importance of recommending women-only or female-friendly places to stay, 2) they were trying to keep their recommendations simple without providing a long list of properties from travel platforms like Booking.com, and 3) they wanted to encourage me to prioritize our preferences when choosing places to stay.

I asked both AIs a follow-up question to regenerate their answers and this time base their recommendations on information from Booking.com. The results were a little disappointing for me, as some to hostels and hotels were repeated (as if they were the only places these AIs knew in Prague). To my surprise, ChatGPT clearly still remembers the set budget range ≤ €50 per person per night, and Bard seems to have forgotten it completely.

For the flights, ChatGPT’s answer is very different from Bard’s as I provide the exact dates of our planned trip. Unlike Bard, who uses Google Flights to show me the exact screenshots, ChatGPT can only give a very rough estimate, such as “on average, you can expect to pay around €100-€200 for a one-way ticket”.

My experiment stops here because I have already asked about all the key questions in making a travel plan – how to get there, what to do, and where to stay. As a person trying to add value by making such plans, my human brain work probably only starts here, and I would not rest until I had narrowed down my options for flights, accommodation and food. It is more than clear that these two AI models could not be of much help to me in comparing the choices – since our preferences are unknown to them, and they can probably never “understand” every single requirement that a human would consider important, especially when some of the preferences are hidden and/or culturally specific (such as the connection between “all girls” and “safety”). What I do find is mostly about the differences between how ChatGPT and Bard construct their answers: ChatGPT understands my question better and uses naturalized language to make recommendations, whereas Bard can provide real-time information and illustrate its answers by inserting images and links. This means that even if my question is not fully answered by ChatGPT, which it is quite often in this experiment, our conversation will continue, and the model would make sure to include what I mentioned earlier (e.g. the budget) in its answers. Compared to Bard, which is more like an advanced version of Google Search, in that it can automatically tell me what to look at when I suggest my questions, ChatGPT does more than just respond to each individual question. This ability makes ChatGPT stand out for being more intelligent.

Here are my conclusions:

1. Use Bard to get real-time answers and direct links when you are thinking about booking flights and looking for restaurant suggestions.
2. Use ChatGPT to brainstorm for inspiration, such as what to do when visiting a new place. This can be achieved through natural, conversational questions and answers. For ChatGPT in particular, the model is all about giving ideas that can be used regardless of time.
3. Avoid asking ChatGPT and Bard to give you a list of all the nice places to stay and eat, because that is not how they work. What they can list as names for people to choose from is very limited, and you still have to keep telling them your preferences in order for them to prioritize the list for you, and it usually doesn’t work out the way you want because they can’t really tell the difference between a “good, but…” and a “nice to have” (or maybe it’s just me being picky).

In the end, it’s fun to play with the AI models and get some recommendations on attractions, cafes, and vintage shops from them, but I’m afraid that I still have to make my own plans for Prague and will also be looking for recommendations from specialized platforms like Tripadvisor, Instagram, and Xiaohongshu.

References

Beebom.com (2023). How to Use ChatGPT for Travel Planning. Accessed 2 October 2023, https://beebom.com/how-use-chatgpt-travel-planning/.

Halpern, A. (2023). I Let ChatGPT Plan My Vacation—Here’s What It Got Right and Oh-So Wrong. Accessed 2 October 2023, https://www.cntraveler.com/story/i-let-chatgpt-plan-my-vacation.