Author: Hailey Gebhart

https://dtc-wsuv.org/hgebhart23/generative-data/index.html

Project Summary

Vision Statement: “Our vision is to revolutionize the review process of Quality Assurance Project Plans by harnessing the power of artificial intelligence, significantly reducing review time and improving the quality of environmental data reporting.”

Project Idea: This project will develop an AI-driven system that evaluates and improves the quality of Quality Assurance Project Plans for environmental measurement. By utilizing natural language processing (NLP) and machine learning (ML), the system will identify high-quality and low-quality elements within QAPPs, accelerating approval processes and enhancing environmental data reliability.

Technologies and Skills:

• Data Collection: Acquiring and labeling historical QAPPs with regulatory feedback.
• NLP and ML Algorithms: Developing models that can understand and evaluate document quality.
• User Interface (UI) Design: Crafting a user-friendly platform for document submission and review feedback.
• AI Integration: Incorporating AI tools into existing regulatory review workflows.

Needs Assessment

Sector Needs: In environmental regulation, the quality and accuracy of QAPPs are crucial. Delays in approval or subpar data collection strategies can lead to significant environmental and public health risks.

AI’s Role: AI can automate the initial review of QAPPs, highlighting potential issues and recommending improvements, thus reducing human error and speeding up the approval process.

Skills and Technology Overview:

• Current: Basic understanding of NLP, some regulatory knowledge.
• Development Needed: Advanced NLP techniques, ML model training, UI development, integration of AI with regulatory systems.

Learning and Development Plan

• Training: Enroll in online courses covering advanced NLP, ML, and software development.
• Workshops: Attend workshops on AI in regulatory technology.
• Self-study: Regular self-learning sessions using resources like papers and case studies on similar AI applications.
• Project-based learning: Develop initial prototypes using smaller datasets.

Budget

• Learning Materials: $500 for online courses and workshops.
• Software Subscriptions: $300 monthly for AI development platforms (e.g., AWS, Azure).
• UI Development Tools: $200 monthly for software licenses (e.g., Adobe XD, Sketch).
• Miscellaneous: $200 for unforeseen costs.

Application and Impact

Application:

• Automating the preliminary review of QAPPs.
• Providing structured feedback to improve document quality.

Impact:

• Regulatory agencies can process QAPPs faster, improving environmental data quality.
• Companies can refine QAPPs more effectively, ensuring compliance and enhancing operational efficiency.

Project Timeline

• Mid-April: Project kickoff, initial training, and data collection.
• May: Development of NLP models and UI design.
• June: Integration of AI tools and initial testing with sample data.
• Early July: Final testing and refinements.
• Mid-July: Project completion and review.

Evaluation and Reflection

Evaluation Metrics:

• Accuracy of the AI in identifying document quality.
• Reduction in average review time for QAPPs.
• Feedback from regulatory personnel and industry users.

Reflection:

• Monthly review meetings to discuss challenges, progress, and learnings.
• Adjustments to the project plan based on feedback and testing outcomes.

This structured approach to your project plan ensures a clear roadmap, defined goals, and a practical framework for achieving significant improvements in the review process of environmental QAPPs.

I am going to de-rail this and talk about how I am actively using AI in my line of work as I think it is a better use of my time for this blog post. In addition, I think it’s cool (although my definition of “cool” may be different than others’ lol). I work for an environmental technology company that specializes in using beams of UV light to measure gases along long paths. We do both community work and industry work. Our industry work includes monitoring emissions along oil refinery fencelines. I work as a data analyst for this company, and we use data in many different ways.

Since we develop this technology and maintain it, QA is a big aspect of data analytics in our business. Our systems operate on a real-time basis, and specifically that means we get new data every 5 minutes from all of our systems at every project we manage. This is a lot of data. As a company, it is important to us that we are producing high quality data. This means the numbers accurately reflect reality and the systems are turned on and operational as much as possible. Since starting at this company, I have learned this is no easy task. and this is where we get to AI integration.

A lot of times there are indications the systems aren’t operating optimally from the data. However, sometimes these can be very subtle, and it has been shown to us very clearly that a human cannot sit there and go through all the data that we’re producing on their own. In order to combat this, we are working on creating machine learning models that can track the system performance as data is coming in, and alert us if there are issues that may need to be addressed.

One way that you can tell a system is operating properly is by comparing to other systems nearby and making sure they match. To do this well, we will often measure additional gases that are always in the air and are not the target pollutants we’re measuring. This means that is something goes funky with the measurement of this extra gas, we know the system isn’t working. We are investigating ways to use AI and machine learning to create correlations between our systems and publically available data (such as data from local government agency sites) to make sure the measurements are similar. If they do not match, then we have an indication our instruments are not working properly.

Another aspect of machine learning we are investigating is natural language models. All of the work we do is based on Quality Assurance Project Plans (QAPPs). If the QAPP is strong and implemented well, the resulting data produced will be the same. Regulatory agencies are required to review these to evaluate the success of a project. We are investigating ways in which an AI can review the project plans and give feedback on what it does well and what it doesn’t. This would allow for a fast way for government agencies to approve/reject QAPPs and for contractors to create QAPPs more easily without as much time going back and forth for review.

Visualization and UI are a big part of these projects as well, and they serve as interesting examples as to what businesses are looking to do with these types of tools, especially in writing and data analytics.

I want my final project to focus on investigating various avenues of generative art. This is a part of the class that has captivated me, and I can relate it to my degree more closely. I want to try to attempt some sort of visualization using a real data set. I want to see if I can use some form of an analytical tool to process a data set to generate an interesting artwork. I haven’t figured out the data set or the medium in which I am going to display it yet.

My essay will revolve around a similar topic. I am interested in doing an investigation on the value of this type of art, and perhaps a comparison of different techniques, including the physical ones. In addition, I would love to explore how AI generative art can aid in data visualization.

https://dtc-wsuv.org/hgebhart23/interstellar-intruders/

Here is the game I created with Haili Kuanoni and Mercedes Rowland.

https://dtc-wsuv.org/hgebhart23/grah/

I don’t know if the style itself is my best work, but I wanted the site to capture the thematic elements more than I wanted my site to look good lol.

Questions for Happy Finish Guest Speaker:

• What was the best skill or strategy that you used to be successful in this job market?
  • As in, are there any aspects of your career journey that stick out to you as an especially good skill/lesson that helped you be successful
• What experiences/education requirements/certifications do you look for in new hires, and where does AI work come into play here?
• How has AI art changed the competition and your interaction with clients? It is becoming easier or harder to engage with them?
• How has AI art changed your workflow and team dynamics?

I’m not super interested in super gravity-physics based or story based games for this project. I like the idea of puzzle games. In that, they require you to figure out a puzzle using the mechanics of the game. Even simple sudoku type games interest me. I think the opportunity for more generative-based approaches might be interesting in this regard. I would most definitely be a coder in this scenario, as I have some experience in program design with my degree/minor. Games that require a lot of collision detection and\or physics stress me out a little bit because they can get blown up very easily, but I am open to most ideas. In addition, I would find it hard to make a branching story game interesting without having the persistent temptation to go over the top very quickly.

As I am turning in the blog post late, I can also give an update on the game progress so far. It’s going very well, and I am very impressed with what we have put together. On ChatGPT’s coding abilities, it made a product that worked, but it missed the mark style-wise on a few things. One example of this is that I created an abstract class for the falling objects that hit the player, and then I created two types of falling objects that implement the abstract class. However, since we wanted certain aspects of the game to behave differently based on the type of falling object, that means that information has to be changed in methods of the classes themselves, and ChatGPT put them in the game loop. This is a very small example of one of ChatGPT’s big issues. It doesn’t always produce elegant and scalable code.

I am most interested in the processes behind visualizing mathematical ideas as art. When discussing generative art, one of the most intriguing examples in my opinion is the art pieces that use physics in order to create very patterned and geometric pieces. Normally when we talk about natural generative art, there is a chaos and a flow that it’s associated with (branches/leaves on trees, etc.), but art pieces that utilize pendulum swings or even those spirographs you may or may not have played with as a kid. These are phenomena that occur naturally, and are just visualized.

I am interested in the intersection in this kind of art with computerized generated art. I am very intrigued by the ability to visualize very complex mathematical ideas that would not have been possible without computers or complex algorithms. I am also interested in the ways in which computers generate randomness themselves. I think there are artistic ways in which we can show the difference in outcome between true random and pseudorandom beyond the distributions of the processes. In addition, I think there are interesting ways in which pseudorandom can be pushed to its limits and have interesting results.

In addition, I am very interested in how machine learning can integrate with generative art. I am very interested in what patterns and results can come out of algorithms that have processed more data than one human can in one lifetime.

In short, I am interested in the patterns that can come out of generative art, and the insights we can gain from these patterns.

Assignment 1: https://dtc-wsuv.org/hgebhart23/generative-assignment-1/

Assignment 2: https://dtc-wsuv.org/hgebhart23/generative-assignment-2/

Assignment 3: https://dtc-wsuv.org/hgebhart23/generative-assignment-3/

Here is a link to the Feathered Bebop’s promotional website: https://dtc-wsuv.org/hgebhart23/band/

Just some information about the making of this video: I had a LOT of trouble getting the video to do what I wanted. For instance, the poster when the Gardener woman is walking by is supposed to say, “War in Space.” Chatgpt got the text good, but the image I used had the woman blocking the R in “war,” so you see “WAWA” instead. Objectively, this is pretty funny. In addition, my goal was for some of these scenes to not have narration, and let the imagery tell more of the story, however, I could not find a good way to include enough context and get the animations to work with me to achieve this. You might be able to piece together how I planned to do this with the clips I had. The narration, unfortunately, was too succinct and efficient and getting the story I was trying to tell across.

https://dtc-wsuv.org/hgebhart23/fibonacci/

Since I have access to the WSU server from my web design classes, I went ahead and posted what  ChatGPT created for me to the server, so you all can see it too if you would like.

I am majoring in data analytics with a minor in computer science. My statistics, data, and CS classes all heavily incorporate programming. Modern statistics and data analytics have too much data to do by hand, so programming is an essential role of what I do.

In my experience, you have to have some idea of what you want and an understanding of what the code is doing to do anything substantial. Sure, I can create this website with next to no programming knowledge, but if it had generated something I didn’t want or wanted to edit it in any way, I would have to understand how it works to make any changes, and if I make anything in the real world, I will most likely have to edit it at some point in time.

In addition, it tends to struggle with following “good practice.” I have seen it get very over-the-top with simple problems. When I have used it to do programming for more complicated problems, it and myself goes through so many revisions and back-and-forth. If I feed it the result to itself on a blank chat and ask for feedback, it finds all sorts of issues in the code and basically refactors all of it. I have asked for feedback on code I’ve written myself, and know adhere to good practice, and it gives me bad advice.

It is really good at writing common algorithms or simple code. It is really bad about employing practices and logic that actually make good computer scientists good at computer science.

Where I have found a lot of success is making it explain code that I don’t understand and for finding simpler functions or algorithms to a specific, small, goal I have (shuffling a list for a bigger project, for example). If there are computer science concepts I don’t understand, it is phenomenal at explaining them in a way I understand it. In this, it is a fantastic teaching tool, but it only works as a teaching tool if I am spending time to make sure I understand what it’s trying to teach me.

There’s already a sort of shift to “human-readable” programming. It was one of the main draws of Python, which is a widely used language that is on the up-rise for the time being. However, Python is built with another language called C, which is also widely used. New programmers like Python because it is more flexible than a lot of other languages, including C. However, because it is written in C, there are quirks to how certain things function, including how data is stored, that the programmer has to know in order to be successful. I have found that ChatGPT doesn’t  always use these quirks to its advantage (Python/C are just one example). My point here is that even though we are shifting to “human-readable” programming, the essence of what makes a good programmer remains. Someone still needs to make sure the output code makes sense. All it does is shorten the learning curve.

The story/moment I will be telling is about a Gardener woman who takes care of the AI trees. Her job ecompasses bio-technology. She tends to the AI trees that are the cornerstone of Grah’s society. It will showcase her as she goes about her day in this world. The world is ultimately peaceful. However, there are news postings about the advances and dangers of space travel. The news outlines that space is too dangerous and too expensive for humans. As our Gardener goes about her usual day, she gets a request from a certain tree to grab some data from the server room. As she is looking for the data, she stumbles upon a different file by accident. This file contains confidential information on supposed alien “invaders”. There is a massive space society of interconnected planets and societies that have, until now, have been unknown to humans. These records support the notion that the AI forces of Grah may be the true aggressors in the space conflicts. In addition, it seems human safety is not the real reason the AI wants to keep humans on Grah. The Gardener must tread very carefully, as she may be in danger in the AI find out she know. However, was this really meant to be a secret, or was her discover some sort of comic design by her AI counterparts?

The style I imagine for this is an animation with ink/watercolor inspired art. I really liked the look of the server room image I generated, as the roots were indiscernible from wires, which is exactly what I am going for. This will accentuate the natural feel of the world. I also want to maintain the blues, browns, greens, and purples that are in my images that I already generated. I would like the score to be heavy in woodwinds and plucky strings (guitars, harps, etc., not so much violin/bow string instruments). Once again to highlight the naturalistic elements of this worlds. The sounds effects I am thinking along the lines of big groaning for the trees with fast wire-y sounds from the electronics.

The most amicable solution that will come out of Hollywood’s integration with AI is that there will be policy between the unions and studios that a human will have to be responsible for the creative aspects of the filmmaking process, with them being allowed to use AI as they see fit. I don’t see a problem with saying, “OK, you have to have a least one human writer. That writer can use AI how they please.” With restrictions on the size of the project/how many writers they need. There is a big difference between someone writing, acting, producing, and directing an independent film themselves and a big corporate studio cutting writers out entirely in favor of AI. The only way to make it make sense would be to enforce that a project with big enough resources and budget to traditionally hire out at least one writer must hire at least one writer. This ensure AI isn’t “taking over,” but it still gives the artists lease to use all the tools available to them.

In addition, audiences are already tired of formulaic movies that are made to sell. People are starting to see through the factory-produced entertainment that many studios/streaming services are putting out. That sentiment is seeming to become more popular as AI-integrated media becomes more widespread.

On a sort of side note, I can’t help but get the growing suspicion that people are so against AI in general because of the way in which AI has historically interacted in media. It’s very commonly the “mad experiment gone too far into dystopia” story, and I really think those ideas impact how people approach AI. I don’t think this is necessarily a bad thing, it’s good to be careful and cautious, but it also dramatizes and blows out of proportion, instead of allowing people to consider the bigger picture.

From a bigger perspective, the emergence of AI begs the question: Will human creation stop if it’s automated? I say no. You see people spend extraordinary time on projects and things just for the fun of it. Just to see if they can. I hold to the (possibly naive) position that humans will always be curious and creative, and they will put their energy into being curious and creative even if the process becomes automated. AI generated works are increasing accessibility to art, which is never a bad thing in my mind.

https://docs.google.com/presentation/d/1EjG64Qoqzuezp2IIjv6B5yka2wodypgq4UaW0ZsOVdk/edit?usp=sharing