Sep 30

Oct 13 HW

One interesting thing I’ve learned from “Sherry Tuckle’s Reclaiming Conversations” is that technology can actually make people avoid real conversations. She explains how people sometimes choose texting or using their phones because face to face conversations can feel uncomfortable or require more effort. I thought this was interesting because even though technology helps us stay connected, it can sometimes make our conversations less meaningful.

An interesting thing I’ve read from Ted Chiang’s “ChatGPT is a Blurry JPEG of the Web” is his comparison of ChatGPT to a blurry JPEG. He’s basically explains that ChatGPT can take a huge amount of information from the internet and creates something that looks like the original which basically duplicates it, but it’s not the exact copy. I thought this was interesting due to it helps explain why ChatGPT can sometimes give answers that sounds right but are actually incorrect.

Sep 29

Week 2 Readings

Crawford, Atlas of AI (Excerpt) – This reading made me think about how people can create technology without thinking about the people who could be affected by it. I was surprised that some AI researchers can work with information about real people without having to go through the same ethics reviews as other types of research. The example about the gang database stood out to me because the data had many mistakes but researchers still used it to train an AI system. This made me question how useful an AI system can be when the information it learns from is already wrong. I also liked the quote, “The lesson, therefore, is that the scientist and technologist must, by acts of will and of the imagination, actively strive to reduce such psychological distances.” This means that people who create technology should think about how their work affects real people. I agree with this because creating a system should not be separated from thinking about its possible consequences.

IBM What are LLMs – This reading helped me understand that large language models are not thinking like humans when they answer questions. They use patterns from huge amounts of text to predict what words should come next. I was surprised by how much work happens before an AI can answer a simple question. The text has to be broken into tokens, turned into numbers and processed through many layers of a transformer. I also found it interesting that people help improve AI by ranking its answers through reinforcement learning from human feedback. One quote that stood out to me was, “The model does not ‘know’ the final answer in advance.” This made me think differently about AI because it can sound very confident even though it technically is making predictions. It also made me wonder how often people trust AI answers without remembering that the system can still make mistakes or give false information.

Sep 29

On Our Second Readings: Ruha Benjamin and Kate Crawford

This Crawford piece speaks very much to a thing I have personally been saying to a lot of people in my life and it is that we really need to pump the breaks a little on AI sensationalism because it is a marketing campaign, an AI power accelerator, a thing to argue about in the comments and therefor a major distraction.

Both pieces speak to the real-time on the ground effects of technological systems with biases, potentially operating on behalf special interests, almost entirely extra-judicially. Our being sounding-boards for the biggest hype machine around these systems only further draws attention to and justifies what issuing touted as an inevitability.

I am always very wary of these systems. I am seeing a lot of people (including our current president) who are abandoning the term artificial intelligence for AGI or super-intelligence. We still have no good definitions for what define these terms versus standard human intelligence. When we elevate this to a point that lays some kind of claim, because rogue agentic AI’s behave in strange ways we can’t quite predict, let alone curtail, to notions that it will almost necessarily, within a strange ever-changing statistic, that seems like it was calculated by AI, destroy the human species out of some Bostrom-like notion of the paper-clip-machine or whatever, we only help to mythologize and advertise for the seemingly infinite power of artificial intelligence systems. This happens look real good when IPOs are creeping closer.

It’s not to say it isn’t possible just that it isn’t happening yet and may never happen and as Crawford suggests and as is shown by Benjamin, there are real material degradations on human beings and our societies and our health and our access to resources and our potential for a multiplicity of futures.

AI is here to stay, what we, the living humans on this planet, decide it will become and where we fit in relation to it is still up for discussion. As with any major new technological development, it can be liberatory or or it can be oppressive, usually it is some combination of both, but it has to be viewed realistically and wielded accordingly.

Sep 29

AI Paper Proposal

I use three LLMs for entirely different purposes, aside from sandboxed models for messing around with image generation or video. Claude for code and general server development and maintenance. Gemini for administrative and research things related to schooling. ChatGPT for every other stupid thing I can think of and to test the limitations, reactions, evolution, and guardrails of AI more generally. While I think doing GPT has more broad implications I am more interested in discussing Claude for doing project specific work. I have a very different relationship to Claude than other AI and I think it helps me have a more realistic perspective on AI use cases against something like GPT which is floated as a catch-all general purpose AI for everybody and everything. I have done extensive work with both models but the specificity of the work I have done with Claude almost better exposes the possibilities and the limitations of it as a productive model for concept refining, coding, UX/UI, and server architecture. Which I think would make for a better piece of writing.

Sep 29

Week 3 Readings

Crawford, Kate, NY Times, “Artifical Intelligence’s White Guy Problem” – This reading taught me that AI can have the same biases that people have. I was surprised that technology that seems smart can still make serious mistakes like misidentifying Black people or giving some people fewer job ads based on their gender. I also learned that the data used to train an AI system can affect how it works. If the data does not include enough different types of people the system may not work fairly for everyone. The example about predictive policing also stood out to me because the system can use old crime data and send more police to the same neighborhoods. This can lead to more arrests in those areas and continue the same problem. I agree that people need to pay more attention to how AI systems are designed and tested. It also makes me wonder how many other AI problems people have not discovered yet.

Benjamin, Race After Technology (Excerpt) – One idea that stood out to me was the idea of a glitch. I usually think of a glitch as a small computer mistake that needs to be fixed, but this reading explains that a glitch can show a bigger problem in the system. I also found the idea of being watched but not seen interesting. Technology can collect information about people without really understanding them. The reading made me think about how people may trust technology more because they believe computers are objective. A computer can still make unfair decisions if the system was built with unfair ideas or information. I was also surprised by the idea that a technology can be created to help people but still cause more problems. This made me question whether fixing a technology is enough when the larger system is already unfair. The reading showed me that we really need to look at why a system was created and who it was designed to help.

Sep 29

Proposal

For my Critical AI Analysis Paper, I plan to study the Spotify recommendation algorithm. I chose Spotify topic because I use it to listen to music a lot and I really want to understand how it decides what songs and artists to recommend to me. The recommendations can feel personal because Spotify uses information about what people listen to and how they use the app. In my paper, I plan to explain how Spotify collects and uses data to recommend music. I want to look at information such as listening history, songs that users skip, songs that users save and artists that users listen to. I will also explain how algorithms use this information to predict what a person might want to hear next. I also plan to study the work and resources that keep Spotify running. I want to look at the people who help create and manage these systems, the computer systems needed to process large amounts of data and the business reasons behind music recommendations. I will also examine how Spotify uses user data to keep people using the app and how recommendations can help Spotify make money. Finally, I want to study how Spotify recommendations can affect music listeners and artists. I want to find out if recommendations can change what people listen to or make it harder for some artists and songs to reach listeners. I also want to examine whether the algorithm can create unfair results or limit the amount of new and different music that people discover. My goal is to understand how Spotify uses AI to shape the way people find and listen to music.

Sep 29

Readings for September 29

• Crawford, Kate, NY Times, “Artificial Intelligence’s White Guy Problem” 

Artificial Intelligence presents a threat to humanity. It has a history of depicting bad behavior such as racial, sexual, and gender discrimination through various examples. Machine-learning algorithms contain many flaws and errors which can be seen through the forms of discrimination being built into them. Whether it’s misrecognizing a person of color for something else or causing distress to people’s lives through a wrongful arrest, AI has shown its true imperfections. The reasons behind these actions reside in the development of these algorithms. The more an algorithm sees something, the more likely it is to recognize it. Therefore, these examples of discrimination unfortunately make sense as the humans behind these algorithms often tend to be “affluent white men” (Crawford). This reinforces historical social inequalities and stereotypes, an issue that needs to be considered by the tech industry and governmental institutions. For this reason, we need diversity and inclusivity in this field so that everyone can be represented equally because if they are not, then they will continue to suffer from AI’s threats.  

• Benjamin, Race After Technology (Excerpt) 

There are underlying social inequalities embedded in our technologies, specifically in Artificial Intelligence. These algorithms amplify social hierarchies and pose harm to minorities. For example, “Diversity is monetized, while the needs of Black users are widely unconsidered.” These technologies reproduce existing social inequalities while posing as more objective and progressive. They have many imperfections like the inability to recognize certain faces or skin tones such as dark skin. Algorithms begin to recognize familiarity, so if they are trained by white males, they will recognize them more as they tend to favor their biases. Their biases “aren’t exposed to human differences – like race and gender” which makes them even more discriminatory. These technologies hurt minorities and others who are affected by their discrimination such as black neighborhoods that tend to be heavily policed or those with non-white names that end up facing discrimination.  

 

Sep 29

Assignment 1 Proposal

My proposal for the “Assignment 1: Critical AI Analysis Paper” is to compare the ability of multiple AI to come up with statements based on simple topics. Since AI are essentially databases, this will make it easier to see how certain AI “think.” It’s obvious that AI can make mistakes so being perfect isn’t the end goal for these AI I’ll analyze, though I will be able to see in what they excel and struggle with. I’ve used AI in the past for coding (making scripts for video games) and has improved over time. Being able to make a statement is a much simpler task compared to coding, but one must also keep in mind that certain AI are made specifically for writing, coding, image creation and more (and some can do all of that, but how well can they do it).

Sep 29

Proposal

MCS 350 Critical Analysis paper proposal.docx

My proposal for the Critical Analysis paper.

Sep 27

September 15: Response to First Two Weeks

            The progression of AI (artificial intelligence) seems to be without an end. AI has been on a quick and steady path to becoming a major impact on society. Being a useful tool that can be used for good, though can be used for bad if in the wrong hands. The consequences of such a tool, created with human consciousness in mind, has been a popular topic of discussion in recent years. Some may be excited to see what AI can do for us as a society, while others may fear the outcome of having access to something that could potentially surpass the need for human control. 

            The data that AI uses to continue to grow is being provided by us, the people, whose intelligence is being replicated for AI to use. AI has come a long way from what it was just a few years ago, and while it may not be perfect, it is consistent in what it aims to do. Even now, many companies, such as google or microsoft, already incorporate built-in AI assistants into their websites and applications, fully realizing what AI is and what will be capable of now and in the future, respectively. One side of this that many don’t realize is the amount of labor and cost that goes into powering AI and propelling its growth forward. From creating huge data centers to hiring people to feed AI directly (from the source of where it gets its knowledge), it isn’t only affecting people as it is directly impacting our environment as well.

            Sure, AI can have its uses, but in the long run, it may be doing more harm than good. Potentially, there could be ways to preserve the materials being used to power AI around the world, though many do not feel the need nor the responsibility to do such a thing. It is true that cybernetics is the future, and AI is a tool that was waiting to happen for years. Hopefully, when that future is finally reached, it won’t be short-lived by all the damage done to get there. Maybe someday people who work on AI will start to realize that this kind of growth comes with a cost, and that could potentially affect the entire world as we know it.

 

Works Cited

https://www.cs.ox.ac.uk/activities/ieg/e-library/sources/t_article.pdf

https://www.goethe.de/prj/k40/en/eth/arm.html

https://www.ibm.com/think/topics/large-language-models