TLDR: A recent paper by Dirk H.R. Spennemann explores the controversial idea that the proliferation of generative artificial intelligence (AI), particularly large language models like ChatGPT, could lead society into an ‘age of public ignorance.’ The research, utilizing strategic foresight methodology, examines how these AI models, despite their transformative potential in knowledge transmission, might erode critical thinking, reinforce biases, and centralize ‘truth’ in a manner reminiscent of historical knowledge gatekeeping, rather than fostering broader understanding.
The advent of generative artificial intelligence (AI), epitomized by large language models such as ChatGPT, marks the third major technological revolution impacting knowledge transmission, following the printing press and the internet. This is the central premise of a paper authored by Dirk H.R. Spennemann, titled ‘Will the Age of Generative Artificial Intelligence Become an Age of Public Ignorance?’, submitted on September 21, 2023, and posted on September 22, 2023, on Preprints.org.
The paper argues that while the printing press democratized access to curated knowledge and the internet further expanded this by allowing anyone to publish globally, generative AI introduces a new dynamic. It suggests that this technology, despite its capabilities, could paradoxically lead to a decline in public knowledge and critical thinking.
Historically, knowledge was concentrated among the clergy and guilds, with literacy and professional expertise carefully controlled. Gutenberg’s printing press (1452) enabled mass dissemination, but publishers became new gatekeepers, influencing content based on commercial or political interests. The Age of Enlightenment and compulsory public education broadened access to curated information through encyclopedias and educational institutions. The late 20th century saw a shift from formal outreach to platforms like Ted Talks.
The rise of the World Wide Web (1993) and smartphones eliminated the digital divide in knowledge access. Search engines initially democratized information by indexing content and ranking pages. However, commercial interests soon influenced page rankings, making search engines de facto gatekeepers. Concurrently, social media fostered specialized online communities, leading to the rise of ‘alternative truths’ and the devaluation of academic experts, as seen in ‘anti-vaxxer’ movements and narratives around the January 6 insurrection.
Generative AI models, like ChatGPT and Google Bard, leverage transformer architecture to produce human-like responses based on vast textual datasets. ChatGPT, for instance, evolved from 1.5 billion parameters in version 2.0 (2019) to 175 billion in version 3 (2020), enabling diverse natural language tasks. The current GPT-4 (March 2023) boasts improved responsiveness, reduced harmful output, and greater factual accuracy, though its training data cutoff was September 2021.
Spennemann highlights that while generative AI can be customized for industry-specific applications (e.g., museum exhibition planning, interactive home design, navigating government regulations), these models inherently lack empathy and confine human creativity to user interaction. Crucially, their output is based on statistical patterns, making them prone to ‘inverted logic phenomena’ and incapable of independent creative thought. The perceived creativity of AI-generated content is subjective, residing in the user’s interpretation.
A significant concern is that AI models, especially in public settings, present a ‘single truth’ derived from their finite, circumscribed, and authoritative knowledge bases. The author posits that this will further erode critical thinking, as the majority of users seek quick, convenient answers without in-depth research. This trend is fueled by:
Increasing familiarity: AI’s integration into daily work will extend to non-work settings.
Instant gratification: Public preference for immediate, low-effort solutions.
Technological dominance: Transformative technologies meeting this demand gain traction.
Declining critical thinking: Evidence suggests a near-terminal decline in information literacy, with growing reliance on social media influencers and dismissal of evidence-based research.
The paper also addresses inherent biases in generative AI. Despite claims of neutrality, model specifications, algorithmic constraints, and policy decisions introduce biases. These stem from the quality and selection of source material (primary, secondary, or tertiary sources like Wikipedia) and the subconscious or conscious ideologies of programmers and trainers. Studies have shown ChatGPT exhibiting preferences for libertarian, progressive, and left-leaning viewpoints, with a North American slant.
This raises the ‘spectre of a malevolent actor intentionally influencing the dataset to pursue an ideological, political or commercial agenda.’ Such control, while more likely in authoritarian regimes, cannot be ruled out in other nations or by commercial IT giants. The dynamic nature of AI datasets means future iterations could dynamically acquire new sources, with algorithms determining what is included or ‘overlooked,’ potentially confining news access to selected channels with inherent biases. Disinformation campaigns could inject content into datasets, and targeted external training, or even the removal of ‘undesirable’ material (akin to Orwell’s 1984), could further manipulate responses. This creates a ‘real risk of a future with a single truth presented to a progressively uncritical public.’
Also Read:
- Canada Urgently Needs Coordinated National Strategy for Generative AI in Higher Education
- Unpacking AI Safety: Red Teaming Generative AI in Education and Beyond
To avoid this dystopian future, the paper emphasizes the pivotal role of the public education system. Educators must instill an understanding of evidence-based research, foster critical thinking, and promote information and AI literacy from early schooling through university. This requires significant political will to prioritize these educational goals and provide necessary resources and teacher training. The author concludes that society has reached a ‘Rubicon’ where the emergent generative AI forces a decision on whether to cultivate a critically thinking populace or succumb to an age of public ignorance.
