Hmmmmmm I for one have come out of this class not knowing if I have the “word of wise” and remain skeptical if I have a grasp on the state of AI at all. One thing I have learned is how rapidly the landscape of this field changes and being surprised by the sudden emergence of AudioLM or ChatGPT. I feel that my concerns about AI have not been assuaged and I’m not confident that enough conversations are being held that challenge the pioneers of these systems to think critically outside of efficiency and optimization. However, I’m realizing that it is valuable to continue questioning these systems and learning more even without being a subject matter expert in the field. I would relay that people working outside the field of AI/ML bring an important perspective that will be integral to making sure the use of these systems is integrated with human interaction. As we’ve seen time and time again, the individuals who craft algorithms and models implicitly affect the outcomes of AI/ML. As a result, having outsiders who are directly being affected by applications of these models can provide more agency to people not privy to all the inner working of these systems. For example, one aspect that I think should more closely be addressed is how AI play as a factor is race and prejudice. One journal in particular, AI and Blackness: Toward Moving Beyond Bias and Representation, addresses this issue through the lens of how investigating anti-blackness can elucidate the status of AI and race and serve to understand the treatment of other marginalized communities. Similar to the discussions we’ve had in 356, Christopher Dancy who is the author of this paper notes how engineering is rooted in designing tools to serve a function and working on how things “ought to be”. As a result, the framing of these functions is directly related to the perspectives individuals designing the system and the source of where these designers are getting their information. The internet and social media in particular have become staples of gathering data to encompass more perspectives, despite being rampant with discrimination, racism, and lack of representation or misrepresentation of people in real life. This results in these models integrating egregious language and misinformation that have had a particular impact on NLP for example. Dancy points out this prejudice that are found in well-known models such as word2vec (which we’ve used in this class) and ConceptNet that complies a lot of other NLP models and is supposedly “debiased”. From there study, they were able to show the correlation of different races and genders having a stronger relation to non-humanistic characteristics especially for the terms “black_man” and “black_woman”. A pivotal revelation of the paper was how Dancy showed terms such as “white_man” and “white_woman” had greater correlations to “human” and “woman” respectively. This is all to show that even in the systems that we are currently deeming efficient models and debiased, we are still seeing clear delineations and prejudice possibly due to the very sources of data sets these models are trained on. One thing I also noticed is the lack of papers addressing these issues. Dancy’s paper also addressed this doing a literature review of the number of papers that addressed antiblackness in 2019, which was zero. I’m really trying to figure out when this will become more of a precedence. When will ethics and the social implications of creations move to the forefront of development as we enter a space that is becoming seamless in our daily use of technology? And will more people from the black diaspora and other marginalized communities be left out and not discussed with each new iteration of innovation? I’m hoping by the time this message is found, there will better answers to these questions and more insight into how design can strive to integrate more than optimization and efficiency. And honestly moving forward, I just hope for more transparency and access to information on how these systems work for all communities.