Language Model AI: Exponential Growth or Flowers for Algernon?

The continued developments in AI are fascinating. With simple language prompts, language models can write articles, essays, poems, and suggest improvements to your LinkedIn profile. As AI capabilities rapidly progress, tech leaders have called for a pause in the AI race due to “profound risks to society and humanity.”

The concerns hover primarily around language models such as ChatGPT and Bard. According to TechTarget:

"Language models determine word probability by analyzing text data. They interpret this data by feeding it through an algorithm that establishes rules for context in natural language. Then, the model applies these rules in language tasks to accurately predict or produce new sentences. The model essentially learns the features and characteristics of basic language and uses those features to understand new phrases."

The rapid growth of these abilities and their use across industry and academia raises multiple concerns.

"Artificial intelligence experts have become increasingly concerned about AI tools’ potential for biased responses, the ability to spread misinformation and the impact on consumer privacy. These tools have also sparked questions around how AI can upend professions, enable students to cheat, and shift our relationship with technology."

I understand the concern. However, will the growth of AI "intelligence" be more like the mouse and Charlie in Flowers for Algernon? If you're unfamiliar with the work, it was first published as a short story in 1959 and later published as a novel in 1960. As a quick summary, researchers conduct a procedure on a mouse named Algernon to increase the mouse's intelligence. Later, they do the same procedure on Charlie, a man with an IQ of 68, and his IQ triples. His new abilities are far beyond what he could do previously.

But, it was all temporary.

Algernon, the mouse, first begins to decline and eventually returns to its former state. During Algernon's decline, Charlie researches why and eventually determines he will experience the same fate. Before he loses his newfound intelligence, he publishes his findings about the flaws in the procedure.

Could language modeling be headed the same direction? Language modeling is highly dependent on the data created by people. These models learn from what has already been created to predict and create text and answers based on that data. As people begin to become more and more dependent on these models for generating articles, stories, documentation, poems, songs, and research, will the volume of human creativity decline?

I believe the answer may be, yes.

As human creativity wanes will language model entropy grow? Will the loss of human content creation be the flaw that causes the eventual devolution of language models and content quality?

Time will tell.