Going beyond ChatGPT

Pi515 AI Challenge opens door for exploration of new AI tools

Pi515’s AI Challenge was an event four years in the making for founder and executive director Nancy Mwirotsi.

With the rapid development of new artificial intelligence tools, she wanted to make it easy for high school and college students to learn about the variety in AI that exists beyond ChatGPT, she said in an interview.

“When we think about AI, everyone thinks ChatGPT. AI is more than ChatGPT. … I wanted to just create a space whereby kids can pick different tracks and just learn something. It might not even hit them today but three, five years later they’ll be like, you know, that makes sense,” Mwirotsi said in an interview.

She believes it’s the first competition of its kind in Iowa, she said.

The competition had teams of students tackle an AI solution in one of four areas: agtech, Iowa’s entrepreneurial ecosystem, social impact or a fusion of ancient stories with augmented reality and fashion.

Mwirotsi said now is a time to make space for conversations about AI and learn about its applications collaboratively because it is still “very new for a lot of us.”

“For me, it’s all about curiosity and what are we learning together, and maybe after that we’re able to know what are the gaps and what gaps exist,” she said.

Pi515 received 11 submissions to the AI Challenge, and some chose to compete during the nonprofit’s Day of Innovation event in Des Moines this spring, including a group of Luther College students, otherwise known as Team Viking.

Team Viking’s members are:

  • Jibran Khan, senior, data science and business management major.
  • Huy Nguyen, sophomore, data science and math major.
  • Mat Rhoden, junior, computer science major. 
  • Evan Marinov, junior, computer science major.
  • Diep Le, junior, accounting and finance major.

Team Viking entered the challenge with an AI solution designed to help restaurant owners identify cities where there are opportunities for their business. Khan imagined the idea due to the limited restaurant options in a small town like Decorah.

Their solution first analyzes restaurant review data of individual Midwest cities to find the top-rated restaurants across different types of cuisine. Then, using U.S. Census data and a machine learning algorithm, the team can recommend demographically similar cities in Iowa or another state that don’t yet have the type of restaurant an entrepreneur is opening.

Data analysis techniques helped them understand and interpret the sentiments of restaurant reviews, Khan said.

The team sat down to share more about their experience in the competition and their outlooks on the future of AI. Their responses have been lightly edited for clarity.

Were there unique opportunities to learn about AI during the Pi515 AI Challenge?

Nguyen: For the data science majors at Luther, our highest level class is applied machine learning and only one of us has reached that class so far, so most of us are going above what we’ve learned at Luther. We used a BERT model in the challenge. It’s a pre-trained model, but it’s more often used in industry or in higher-level research. That’s one thing outside of the class that we implemented in our model. One more thing is that the data that we collected, we were searching, browsing the internet, trying to gather data from multiple sources, which we did not get to do in class as well. We would use provided data from class and usually very clean data, and that we have to go through all the cleaning and collecting process. We realized it was a very time-consuming process.

Rhoden: Collecting data was a big challenge for us because we don’t have a class that can guide us on how APIs truly work or what are some tools that we can use for web scraping and stuff like that. We definitely learned a lot from that.

What were your takeaways from participating in the challenge?

Nguyen: Before I started this challenge, all the things that I did related to AI were on my own — going to YouTube videos, seeing new features or new products that come out from Big Tech companies and I was like, “Oh, these things are cool,” but I have no idea how to implement like the application of AI or just the machine learning in general into a real live problem and come up with a solution to solve it using AI. I had no idea how high the impact or the capability of AI was, but taking on this challenge and hearing Jibran’s idea, I was like, “This AI thing can actually solve bigger problems than I thought and also be applied to anything, literally anything in everyday life. Our project can help entrepreneurs actually make good business decisions and make money, so I was fascinated by that. Now I am more open to the idea of using AI on an everyday basis.

What did you have to learn as you went through this project?

Khan: We realized that, first of all, our approach was that we were going to different places on the internet and just finding as much data as possible for the demographics. For restaurants, we targeted Yelp for our data, but for demographics, we were just going all over the place looking for the correct data set, but the problem with that was the data was inconsistent. For some cities, we would get some information that was not present for the other cities. In order to keep the data consistent, we wanted to have something that remains constant for all the cities so we can have good comparison between the cities when we’re applying the machine-learning techniques to it. We just used the U.S. Census. We found out there was some data available, but it was also limited data. For almost a month we were just looking for data. Things could have been better and smoother if you could have just targeted one source, but we were just looking for more and more, and it just wasted some of our time.

What do you see as the future challenges and opportunities around AI as students whose jobs will likely be affected by it?

Rhoden: You know how Adobe has AI implemented and you can actually create art from just a prompt? All that is being trained with a data set. In past years, the data set that the AI is trained on is human-generated data, but now a lot of people are using that and are creating their own art. We’re going to get to a point, at least this is what I believe in, that AI’s not going to be good enough to the point when it’s training on AI-generated data. Right now, at least for the art area, the art is good, but it’s because it’s being trained on human data. Once it gets trained on AI data, it’s not going to be as good. For example, if I wanted to create a poster and the machine-learning code, let’s say they have like 500,000 posters about the same topic that it can be trained on and they’re all human-made, it wasn’t made by AI. But in 10 years, since people are using those AI tools to create AI-generated posters, it’s going to be more trained on AI data than on human data. It’s going to get to a point where the creative content is going to get stuck.

It’s not going to be as creative because it needs data to be trained on.

Nguyen: It’s not that companies aren’t already using synthetic data created from AI to train their own models. Even ChatGPT-4, they’re using a lot of synthetic data in their model right now. I personally don’t agree with the method of using synthetic data in the training process, but I’m sure the CEO and the leadership in those companies have their own hypothesis about that. 

We cannot imagine how different the world will be in the next five or 10 years, how we’re going to use AI in the next five years. We don’t know yet. The publication of the GPT-4o model from OpenAI, that literally just changes everything that we’ve known so far about AI. Hearing AI talking like a human, reacting like a human, interpreting the emotions and the stories like a real human really scares me. That’s an interesting, but very ambiguous future about AI. 

But looking on the bright side, while people are thinking that jobs that humans are doing will be taken by AI, I think that we won’t lose the number of jobs in general, we’ll just lose the jobs that can be automated by AI, and then we’ll become the people who actually use AI to do other things. It’s like another industrial revolution of people changing their job titles, changing the way that they’re using the tools in their everyday life.

Khan: I’ve been interacting with AI since before ChatGPT came out, and when ChatGPT came out, everything just changed. I always think about what the future is going to hold, but I think a positive way that things will change is that people will be entrepreneurs. If you have an idea, you don’t know what to do. You will think I need like 10 people to work for me and, but with AI, with all these large language models that we have — these are two years in and now they are so good. They’re doing most of the data analysis, so in the coming years it will become better and better. 

It will eat a lot of jobs. That’s true, but they will be replaced in a way that we will not even need these jobs. People will have their own way of thinking, they will be more creative. It’s not how good of a coder you are or how good you are with computers, but how creative you can think as a human. Every individual has unique capabilities of thinking; we are just built like that. The more creative mind you have, the more broad thinking you have, the more you have the opportunity to explore AI.