These are open questions! As you ask them, keep an open mind and be ready to go “off script.” Your students’ answers may surprise you.
All major GenAI creators built the large language models (LLMs) their technology uses by conducting massive “scrapes” of the public internet. They did not ask permission to incorporate and use all of this material, and in some cases, there is evidence that their activities may have sought to circumvent copyright and other laws.
In starting a conversation about this, it may be useful to ask some of these questions:
- Who is responsible for what’s on the Internet?
- Service providers often say they offer only a platform and that individual users are responsible for what they post and for whether it violates laws or moral expectations. How fair is it to expect individuals—or service providers—to take responsibility for what’s online?
- How much of human behavior should be governed directly by laws and the courts, and how much should people be expected to exert influence on their peers?
- Background Resource: A New York Times podcast explored the origins of GenAI LLMs.
In building LLMs, corporations also employed humans at very low wages to filter and screen material, to keep LLMs from containing too much harmful content. These screeners were therefore exposed to the very worst the internet has to offer, and may experience long-term psycho-emotional consequences, for which they were compensated very little.
- Who is responsible when a company uses suspicious labor practices in product design, even though the product is widely used?
- How would you compare the labor practices sketched out here with other practices that have been called into question, such as sweatshops, animal testing, or the employment of undocumented immigrants for dangerous or uncompensated work?
- One traditional answer to unethical product design was the boycott. If we come to believe that LLMs have unethical origins (and this is still being litigated, by the way), then how would a boycott on GenAI work? How effective might it be?
In GenAI conversations, it’s also common to reason in terms of final outcomes. Is this kind of framing useful for this context?
- What’s the difference between using GenAI to generate text, using it to generate an image in the style of a specific artist, and using it to generate an image that looks like a photograph?
- What other technologies might have dubious origins like GenAI seems to have?
- What other technologies might have had honest origins, but dubious uses or effects?