"This primer seeks to reinsert workers’ current and material experiences into the public discourse and question promotional language about the magical, infallible, and seemingly inscrutable qualities of AI.
In Part One, we define generative AI and discuss its different iterations as work technologies, and interrogate assumptions and claims about its capabilities.
In Part Two, we draw connections between different global dimensions of data labor, and how the underlying data behind generative AI technologies impact workers’ livelihoods.
In Part Three, we situate generative AI in the context of workplace adoption and identify how it is distinct from and connected to other forms of work automation."
https://datasociety.net/library/generative-ai-and-labor/
[#]GenAI #AI #Labor #Labour
=> More informations about this toot | More toots from christinkallama@hcommons.social
"The development of AI relies on a large global network of data workers who have been purposely made invisible in order to make systems appear more autonomous, even sentient. OpenAI flaunts that GPT, its LLM, is trained on millions of data points scraped from the internet, but hides the fact that workers must make this data suitable for training. Workers perform a wide set of tasks including cleaning, labeling, keywording, and semantic segmentation." (p. 6)
[#]OpenAI
=> More informations about this toot | More toots from christinkallama@hcommons.social
This discussion of "tacit knowledge" is 👍 helpful for explaining how #GenAI can seem like it's doing something magical:
"Generative AI models can recognize patterns in data, such as transcripts of call center interactions, to identify the language workers use to build rapport with customers. When prompted, generative AI applications will replicate these elements in outputs, suggesting that some form of human ingenuity or creativity was required to generate these responses." (p. 7)
Human ingenuity was, of course, needed to generate these responses - in the form of the training data, for which the originators received no recognition or compensation.
But customer service is a skilled occupation:
=> More informations about this toot | More toots from christinkallama@hcommons.social
"...generative AI models do not understand the motives, meaning, and context behind interactions, such as easing a customer’s frustrations while troubleshooting a problem.
Workers’ tacit knowledge gained through customer service experience is embedded in the language that AI systems appropriate."
Therefore: 👇
"Reorienting our understanding of the type of labor or skill that generative AI is able to achieve, from creativity ➡️ replication of tacit knowledge, gives us a wider view of how occupations may be affected." (p. 7)
=> More informations about this toot | More toots from christinkallama@hcommons.social
Even in this nuanced and critical discussion of #GenAI, some further clarification is essential, e.g. (p. 7):
"The foundation models at the center of generative AI can be used to accomplish a variety
of tasks, such as content creation, research and summarization, and task automation."
[#]LLMs can produce results that look like research or summarization, but they do not engage in the discovery and assessment of facts or in the discernment of the core arguments and evidence of a written work. In their current form, they cannot engage in research or summarization.
📣 Spread the word 📣
=> More informations about this toot | More toots from christinkallama@hcommons.social
@christinkallama the most accurate name I've seen for AI of the ChatGPT variety is autocomplete on steroids.
=> More informations about this toot | More toots from quixote@mastodon.nz This content has been proxied by September (3851b).Proxy Information
text/gemini