Understand hard truths about generative AI

- In mid 2023, comedian Sarah Silverman launched a lawsuit against two big tech companies, alleging they had used her books to train AI systems. Silverman was not alone. AI vendors are facing numerous copyright lawsuits involving the data used in training AI. AI vendors claim this is fair use, serving the bigger public interest. We talked about that in our discussion of copyright earlier in the course. Yet beyond the legalities, there are ethical questions. There are already stories about AI running out of training data, which speaks to just how much data is needed to train these massive systems. This is clearly different in scale from the traditional ideas of fair use, such as quoting a line or two from someone's book. To have your life's work be used this way and to not be compensated is a big deal. A point that 1000s of authors guild members have made in an open letter. And it's not just text, it's software code, images, audio, video. Basically, every type of digital content is being used as training data. This is far from the only ethical issue in the design of generative AI. It's just one of many. Let's walk through some other ethically challenged design choices. Generative AI is trained with biased unrepresented data, primarily gathered from the internet. Since so much data is required, there is little curation for quality. These biases are reflected in the models themselves, and it's not easy to solve this problem after the fact, with filters or other technical fixes. The use of big data leaves an enormous carbon footprint, because all of that data is housed on servers in data centers, which use water and energy to keep things from overheating. Generative AI can use 10 times more resources compared with regular digital processing. In a Washington Post, article researchers estimate that for every 100 words generated by AI, the equivalent of one bottle of water is used at a data center to keep things cool. In some communities where data centers exist, people are already experiencing the impacts of competing for limited water and paying higher costs for access to power. With increased demand for data centers, some researchers estimate that carbon emissions from data centers will more than double by 2030. There is also an unethical global supply chain used to prepare data for AI systems. This is tedious and often traumatic work, as the work need to review the most disturbing and offensive content, in order to keep it out of the training data set. This work is done by poorly paid people, mostly from the global south. Finally, we need to consider who controls generative AI. It's a handful of companies that have the billions of dollars needed to build this kind of product for further consolidating power and inequality. We've already seen how a handful of big tech players are dominating the AI space, raising issues around monopolies and anti-competitive behavior. So let's summarize. We have a product controlled by a handful of big companies that further consolidates their power. That product is allegedly made with stolen goods, is known to create biased outputs that can cause harm, was built with sweatshop labor and is an environmental nightmare. These are the hard truths about generative AI that everyone needs to know. When we use generative AI, we're complicit in upholding the system of ethical design choices. That's a really hard thing to acknowledge. As an end user, you will not be able to impact those choices directly. However, indirectly, you can raise your voice to demand fairer options. Somewhat like a fair trade coffee version of generative AI. Some organizations are setting up ways to ensure people are fairly compensated for the use of their data and developing more ethical supply chains for data work. Researchers are working on developing models using better and less data, which would still perform at the same level as these bigger models. Other parties are focused on greener data centers, powered by renewable energy. You might be wondering at this point, do I even want to use generative AI as it currently stands? That is a tough question that only you can answer. But before you decide, let's take a look at another set of ethical questions related to the use of generative AI.