Implications of Generative AI on Information and Media

By Ethan Bingemann.

Investment in artificial intelligence has been on the rise since 2013 with a peak in 2021 with 72 billion dollars invested in the technology [1]. With market value of around 276 billion dollars [2]. This investment has unsurprisingly been paired with a huge step in the functionality of AI technology. AI has already been used to do many invaluable things. The ESP (Enzyme Substrate Prediction) model, for example, can predict how enzymes interact with various substrates with 91% accuracy [3]. There has also been a huge rise in chat bots. With most major tech companies coming out with their own form of chatbots. This includes newer chatbots such as Microsoft’s Bing AI, Google’s Bard, Open AI’s ChatGPT, and Meta’s Llama 2 [4]. The older AI Chatbots include Apple’s Siri, Google’s google, Samsung’s Bixby, and Microsoft’s Cortana [5]. It has also been used in personalized education, coral reef restoration, and city optimization [6]. In fields such as medicine, user interfacing, data analysis, and many forms of optimization the true functionality of AI has not fully been recognized. If it isn’t already part of everyone’s daily lives it will be soon. The potential benefits of AI cannot be understated.

The biggest concern with AI technology is that the majority of AI investment is not by governments. Countries like the US lead investment with a relatively small 140-million-dollar investment in the technology [7] which is less than the investment of 16 companies and startups [8]. This represents about .2% of total investment and in turn is likely not to have a huge impact on the market as a whole. The influence is in the hands of 5 to 10 major tech companies competing. This context needs to be considered when discussing any other concern relating to AI technologies. It also explains why government regulation is so necessary.

Misinformation can easily be spread through AI technologies. One form of this misinformation is AI deep fakes. Simply put, a deep fake is the manipulation of facial appearance through deep generative methods [9]. Non-factual information can spread like wildfire on social media because inflammatory content tends to attract engagement and thus gets recommended to more users. AI models are also used in algorithms to better predict content users would enjoy. This makes social media more addictive and potentially unhealthy. Social media is considered highly addictive but, due to its recency, there is only speculation about the long-term effects it will have on youth. With some experts going as far to say “Social media is the new smoking, it’s an addiction” [10] and calling out it’s “dopamine-manipulating features” [11].

Discrimination can also be a large problem in AI modeling. Most AI systems rely on data sets for training and issues can easily arise if these data sets are biased. Amazon Alexa for example misidentified 35% of the words in African American speech compared to only 19% for speech from white people [12]. This discrimination is not always this subtle. Another example being GPT-2 which was trained on BookCorpus (7,000 self-published fiction books) and another dataset of 8 million web pages [13]. Despite this dataset only being about 40 GB, the LLM (large language model) encoded white supremist rhetoric as well as racial slurs [14]. These AI systems do not have any form of consciousness and are not learning to be racist. The biases come directly from biases in datasets.

Government surveillance and transparency has also become a growing concern. AI has the ability to track large amounts of information in very short times. This makes it ideal for analyzing video and tracking people or objects. This raises many ethical questions on how far the technology can go in monitoring people. There is a sense of security that could be violated if AI is monitoring personal actions. One AI bot created by graduate students demonstrates an ability to guess within 25 miles of an actual location given a photo and within a couple blocks in major cities [15]. There is also a concern that AI could “perpetuate existing biases and discriminatory practices” [16].

Along with this AI has great potential for inaccuracy. This ranges from ChatGPT creating fake court cases to use as precedent [16] to nearly fatal self-driving car issues [18]. Relying on AI for policing runs into similar concerns. In terms of moral and ethical issues there are countless. This was perfectly exemplified in Microsoft AI article with a “guess the cause of death” [19] and a New Zealand based AI meal planner recommending a recipe for chlorine gas [20].

Job loss has always been a concern with AI. An estimated 300 million jobs will be lost to AI [21]. This means a large percentage of the workforce could be entirely replaced with an AI system. This leaves world governments and or large businesses with the profit while hurting the workforce. These workers could still find new jobs however they would need new education in a more demanded job market. This demonstrates the issue of structural unemployment. In many fields AI job replacement appears inevitable, and the responsibility falls on the government to ensure that its citizens are not without the ability to work.

The final issue that all world governments need to concern themselves with is AI becoming too powerful. There are beliefs that AI will lead to “civilization destruction” [22] and that AI poses an “existential threat” [23]. These concerns come from the development of self-awareness AI systems. This means that AI systems have knowledge of themselves and their existence in a way that emulates consciousness. No such system exists yet.