Research Guides: AI Literacy: Algorithms, Authenticity, and Ethical Considerations in AI Tools: Emerging issues with AI

Emerging Issues with AI

While AI has a lot of potential when used responsibly, there are a lot of issues that are still being worked out.

Copyright

Because AI was built using lots of textual data from a variety of sources, there have been questions about how copyright applies to both works created with AI and the works that were used to train AI tools. The New York Times is currently suing OpenAI in a claim that their use of New York Times content violates copyright restrictions. They claim that the ChatGPT has occasionally reproduced New York Times articles verbatim and that hallucinations which falsely attribute information to The New York Times are damaging to their reputation. You can read more about this lawsuit here.

Some countries are addressing copyright issues by creating opt-in policies instead of opt-out. A proposed bill in Denmark would automatically give humans the copyright to their own image in an effort to combat deepfakes. Chat GPT has also begun including metadata with each generated image to help users identify AI-created images.

In comparison, Australia is considering adjusting their copyright laws to make it legal for AI companies to use Australian copyrighted materials in training AI models. This move has garnered criticism from author rights groups.

The Value of Information

As AI tools develop, we're beginning to see different tiers of access where some users are able to pay for a better version of the tool. Like so many new technologies before, these pay-to-access models could exacerbate gaps between people who can afford to use them and those that can't. Additionally, much of the world's information is not freely available on the internet. Most academic information requires a subscription or authentication through your university. This means that many AI tools can't use some of the highest quality information when generating responses.

Reproduction of Bias

Because AI tools were trained on data that was created by humans and programmed by humans, there is a high likelihood that many of the existing biases in our world will be reproduced endlessly by AI tools. For instance, AI tools used in health care have been shown to consistently downplay women's health issues when compared with men's. Another study analyzed 133 AI systems across different industries and found that about 44% showed gender bias and 25% exhibited gender and racial bias. UNESCO found that AI reproduces bias in three main ways.

Data selection - When the training data does not include viewpoints from diverse backgrounds, it is likely that these voices will continue to be minimized, marginalized and stereotyped.
Algorithmic bias - Because AI tools use pattern recognition to produce an output, historical biases are likely to be reproduced. For instance, when certain jobs are heavily male or female dominated, AI tools are likely to produce information that reproduces those patterns.
Deployment bias - Researchers found that when AI tools were used in settings which they weren't developed for they began using improper terminology and making associations between specific ethnic groups or genders.

Environmental Issues

AI has also had an impact on the environment primarily in the construction of data centers. Data centers are large temperature controlled buildings designed to house servers. Training and running large language models require enormous amounts of power for these data centers, which increases demand on the energy grid and often relies on fossil fuels. Additionally, data centers consume vast quantities of water for cooling, putting a strain on local water resources.

Overall AI is likely going to radically change how we work, communicate, and interact with the world. That makes it important for you to understand how it works and to keep up with the developing issues as you grow and learn.