Monthly Archives: September 2023

Week 6 Blogpost

This week, I initiate the idea of an AI application in game. I come across an experimental game demo developed by Stanford researchers called Smallville. What sets this game apart from the commercial product is that the game’s NPC (non-playable characters) comfortably discuss topics such as local politics and composing music, pulling from ChatGPT’s enormous database” (McCurdy 2023). What is more fascinating is that they even retain the interaction history and reference the earlier given information (McCurdy 2023). They claim to use the human-centered design to prototype simulations of human behaviors as they ‘learn’ plausible sets of behaviors and reflections that may exhibit in life” (McCurdy 2023). However, it poses a threat to replace game writers when in fact, human agents still play a vital role in assessing AI.

With our group’s media sketch this week, we take advantage of our open-ended storyline to create different visual approaches for our trailer. In the meantime, everybody will have the independence to choose their favorite part in the synopsis and create a trailer following their interests. The arbitrary visual functions as art direction ideas, giving us more quality options to pick out. Furthermore, it is important to look back and assess the effectiveness of text input. What does not work is that we have not decided on the characters. As a result, the trailer misses an important element- presenting the main characters. 

In order to back up the deficiency, I feel it is more effective if we can use this brainstorming technique called 6-3-5. Six people will sit around to jot down three ideas for problems – each round takes five minutes and rotates after we get eighteen ideas on each note. After that, we can cross-check the ideas to force association. This might help us subconsciously track unexpected suggestions.

This week’s reading clarifies the ramifications of deep generative models in the breadth of variable media forms, notably the mention of hyper-production. It also demonstrates the association between hyper-production with social-economical thought on the rentier state (Ferrari and McKelvey 2022). Thus, it addresses the challenge of understanding the ethical barriers of AI generation, such as the extensive use of personal information. I am interested in the notion of homogenization and convergence of media production through hyperproduced content (Ferrari and McKelvey 2022). Imagine the possible media outputs generated by the compressed data point, for example, the recreation of Mona Lisa painting in Van Gogh’s style. As a result, it leaves several questions that might impact cross-industrial, such as the copyright issue of archived content.

Reference list:

McCurdy W (2023) ‘No more ‘I took an arrow to the knee’: Could AI write super-intelligent video game characters?’, The Guardian, accessed 1 September 2023. https://www.theguardian.com/games/2023/may/25/could-ai-write-super-intelligent-video-game-characters-stanford-smallville

Ferrari F and McKelvey F (2022) ‘Hyperproduction: a social theory of deep generative models’, Distinktion: Journal of Social Theory, p.1-23, doi: 10.1080/1600910X.2022.2137546

Here is the link to the sypnosys (please use RMIT account to access):

https://docs.google.com/document/d/11X_GoWq4QmpSubxagBaEHwEEeM3ZvHAEldZ8pfCvABI/edit?usp=sharing

Here is the link to the trailer (Shout out to Beila):

Week 4 Blogpost

This week I come across the AI fashion week debut in NYC. Followings the related articles and interview clips, it is fascinating to see every process automated entirely by AI, from cloth designing to stage set-up. The quality is phenomenal to the extent that I initially think it is a Vogue editorial. Personally, it is an interesting observation because this content substitutes the involvement of human labor. For example, it replaces the photographers to capture the atmosphere backstage. You can review the show’s images in the attached link:

Here is the page of the event: https://fashionweek.ai/

FAQ: https://fashionweek.ai/faqs/ 

I look up FAQs and discover two requirements: contestants need to “facilitate the reproduction and sale of your collection” and the final quality is aligned with Vogue’s. Therefore, I guess the contestants extract samples from trendy fashion collections and prompt them in machine learning. They then organize the output images in a linear order that matches the theme of their collection. However, it is questionable whether the use of AI might be considered ethically creative as far as in the garment industry.

We are able to explore new creative directions for our inaugural sketch, from a series of Tweet threads to interactive mediums such as TV news. As far as we develop, it widens the available options with the tools in hand. For example, we can stimulate a podcast focusing on the issue or even recreate it in the form of TV news. However, it strikes a problem that we have not agreed on the medium. Therefore, everybody comes out with different sketches.

We hope that for future work, we get familiar with the technological options provided. For instance, I am interested in the collaboration feature of Reduct where we can create a design-thinking framework. In Eleven Labs, the voice synthesis has a paid subscription feature for integrating real-life voice

For this week’s video, I am particularly enthusiastic about how Simon Willison demystifies the complexity of machine learning concepts. He does not use jargon in explanations but rather compelling language with interactive visuals. For example, when he illustrates how LLMs actually work, he brings out the Apple auto-translator to break down the similarities in AI topics. One key principle that hooks me is when he shows interactive methods with AI. He blatantly zeroes in on the AI’s lacking in which hallucination turns into obviousness. He recommends we understand how the model works and learn the hard lesson that the iterative process will be never polished. As a result, we have to be the deliberate moderator of the companionship with AI.

Here is the link to my Week #4 sketch (please use RMIT account to access):

https://drive.google.com/file/d/1OwNd7RGXziUBp9DQEVcnUN1IU-IUZWu-/view?usp=sharing