We should stop anthropomorphizing AI. In Kate Crawford’s Atlas of AI, she says, “AI is neither artificial nor intelligent, rather artificial intelligence is both embodied and material”. This idea that AI can think on the same level as the human mind is a misconception. One that exists without its proper context. Crawford delves into the different aspects of AI’s material infastructure, but the one that fascinates me the most is data.
Data
AI models are trained off data/content to make accurate predictions. Whether that is to identify an apple, or to construct a paper in MLA format. AI will follow its data set to produce a desirable answer. With the rise of social media, content has exponentially increased in the past decades. In Kate Eichhorn’s Content, two points are addressed: diversity and accessibility. In short, Eichhorn mentions how social media has made sharing photographs, videos, and posts easier than it has ever been. This content, although great for users to interact across the web, has become an unregulated gold mine for data collectors.
Humans
Photographs of humans, videos of humans, and posts written by humans are what train AI models to “think” like us. It is what it is made for, yet there is no regard to the privacy of all those people. Data and humans might as well be interchangeable terms to AI researchers. Social media practices makes me question if there will ever be a way to return to an ethically researching AI. Crawford uses an example from the 90’s in which subjects would be photographed in different lighting and angles for facial recognition research. It is reminiscent to scientific studies, as “each subject was briefed about the project” and gave their consent. Nothing like this is being done today, and why would it be? Researchers have access to an infinite amount of data.
Takeaway
Crawford’s passage has shed some light on the inner working of AI for me. Returning to the initial quote, AI is often made out to be more than the sum of its parts. Like it’s this futuristic tech that can “learn” from us. Eventually having the intelligence to create something new. However, it really is the individual elements that fascinate me more. What data is collected to inform this AI model, and how was it collected? For what purpose was it made by the researchers? What biases can be seen through the predictions and answers it will come up with?
Loose Ends
This is post disregards all the other aspects of the AI infrastructure that Crawford brings up: human labor and environmental costs to name a few. Sorry if this post seems so disorganized. Crawfords passage was way too vast for me to touch on everything I would have wanted to. So, my main question is whether or not you think AI research can ever become ethically produced again?
Comments
2 responses to “AI Resources”
I really like how you drew out the major key parts of her set on Data. I think Ai is often mistaken for true natural intelligence, just as stated, but rather more is massive engine, generating data. While the human mind is the same, “downloading” key phrases and information to our brains and then regurgitating them later on, AI is only able to access how much the provider allows it to, which is us.
I think the mention of Ai eventually being able to become intelligent enough to create on its own is really big and can be quite scary to think of how it could impact us in the future. I just don’t know if Ai will be anything more than a strong tool, something that while has the ability to generate answers and create projects, it will still be limited to the confinement of the human mind. Unless of course it turns into terminator.
To answer your question, no I do not believe that AI could ever be researched ethically again. Much of the data sets that were gotten have been fed into larger data sets that AI is using. How would we separate these data sets? Moreover, we must consider the context that these were gathered. If every Facebook user agreed to give away their data, we would still have issues. How many photos posted on Facebook are of city shots, of people’s kids, or include people who did not agree. I would guess quite a lot. The issue with using this data is once again that it removes context. The only way to make it ethical is through a proper study (which is already mostly devoid of context). Then we run into the issue of database size. In short, I do not believe it will ever be possible to gain this data ethically in a meaningful capacity.