Last week, as a class, we created chatbots for a variety of different uses and I personally had created a chatbot for the website itself if a user didn’t want to try and find something and instead simply find out information right there and then. Users were encouraged to submit their questions about me (Steven Grevera) since the website itself is about me. The chatbot was able to do this much, to my surprise, very, very well.
From my experience with AI and I do have some indeed, AI tends to work better the more you provide it. If you give it too much “wiggle room” or not enough information, it tends to fail. AI isn’t actually “thinking” in the same way that humans are, despite what much of the scientific jargon and arguments for AI might say. AI is simply acting upon the information you provide; the larger the sample size, the better.
That’s not to say that AI cannot be used creatively. That is an emerging industry, regardless of people’s thoughts on the topic and while AI creative projects can usually be told apart from human intervention, they have their uses for small things, if not at the bare minimum, as a “starting basis” creatively for an actual human to then go in and touch up on.
I personally did not have to correct anything except for one single issue with the chatbot, which was its inability to provide downloadable PDFs of my resume. I fixed this easily. The reason, however, that this project was not difficult for me was that I have experience working not just with AI but also with chatbots in general. Being a fiction writer myself in my spare time, I’ve toyed around with seeing if AI could play some of my characters for fun, and while it’s not perfect, it clearly goes somewhere if the correct parameters are given. I’ve even gotten ideas from actions the bots might have done that are totally in character for my character itself but I just had yet to utilize due to a scenario that may be presented to it. For example, a type of personality type that I don’t personally know how to write “well,” interacting with my bot and if that information is then provided, allows me to improve not just the bot but also the character itself for my own written stories.
For training data, I wrote into the baseline code to use the GitHub website itself to pull from all the information that was there. Yet for added good measure, I also fed it all the web pages directly in text format with some mild and minor alterations to ensure that the bot could read them entirely and provide users with the best experience. For example, the bot was returning a 404 error HTTML web page for my resume, so I directly fed it my resume itself and wrote the instructions “Provide the following link for users to download it if asked” in that specific scenario. I would do the same as a precaution for some other scenarios, writing out instructions in the actual text documents, such as “If the user asks for x, then present this webpage.”.
I must stress, however, that despite my extra caution, the bot worked extremely well, even doing things I didn’t expect it to, like being able to calculate my age from a birthday I had previously provided. Due to this fact and after much thinking, I omitted my birthday for personal reasons to avoid that exact scenario from being replicated by a potential employer. This was just about the only kind of data that was “excluded” and, after the fact, not in the process of making the bot itself. Other than that, nothing was off-limits.
The future of this technology in terms of advancement is likely bright and thriving. Stronger and stronger models are being made globally. I personally see the uses of AI but feel that creative projects involving AI should be limited in scope if not outright regulated, however, as there are given dangers with such powers. It is a simple fact, especially creatively, that AI is simply replicating human behaviors, or “parroting,” in many instances.
What I mean by this, in the example of art, is that AI trains off of existing artwork as a means to train itself. This means that a human “style” could never be achieved by AI. All that ever could be achieved by AI is what it views as the “highest quality product.” So, for example. If AI was around when Tim Burton was first created, AI could not nor ever would create in Tim Burton’s style for some time. It would take weeks, months, or even years of Tim Burton’s style existing for the AI to be able to replicate it perfectly. The reason being that AI doesn’t “think”; it doesn’t “create”. It’s simply training off the input provided.
Style, in itself artistically, is human by design. While other humans may try and successfully replicate it, sometimes, such things cannot be done without intensive training, not just by AI but even by humans. For example, when one says “Cartoon,” there is no one specific style that fits into that category. Yet on the opposite end of the spectrum, when one says “Anime” style, there are a certain amount of characteristics that have been not only practiced but standardized over the course of many years.
For artists, styles are like language and the first to make them will be perfect in it and eventually, others will learn it as well. AI, in theory, can do the replication much faster than a human can but they could never create “their own style or language”. Not yet. That could change one day. Yet I personally encourage that day to take its time.
Prompt R: After you have completed the Chatbot assignment or any assignment using AI for most of the work, reflect on your interaction with this emerging technology? What does AI do well What does it not do well? What did you have to correct or re-prompt it to modify? What can you know about its training data? What kinds of data was included What kinds of data was excluded? What do you think the future holds as this technology is used and further advanced?
AI Statement: Grammarly was used to spellcheck the writing of this assignment.