What's wrong with AI?

This essay is not meant to fundamentally alter your perspective on artificial intelligence. If you view AI as beneficial, continue using it; if you find it concerning, acknowledge that you are not alone in your apprehensions. My background is rooted in technology, spanning decades of involvement in the industry, from early computer code programming to current work in music software creation. Despite my long-standing enthusiasm for technology, I now advocate for caution towards AI.

I will focus on two specific groups: those who are skeptical or doubtful about AI, and those who are skeptical or fearful about its current implementation. My aim is not to spark a debate, but rather to elucidate concerns and offer assurance that they are legitimate and proportionate. For the skeptical individuals, I intend to provide reassurance that you are not alone in your apprehensions, and to address all of your worries. For the sceptics, I hope to validate your feelings and provide further points of concern.

To frame this discussion, I will focus on two audiences: the doubters and the sceptics. I do not seek to initiate a debate; rather, I want to provide reassurance to the former and give further cause for concern to the latter. If you disagree with any of my assertions, please refrain from informing me directly. I am willing to provide additional details upon request, but please do not use this as a way to sidestep a disagreement or the beginning of a discussion.

The topic of AI is not inherently problematic. Research on the concept has been ongoing for decades, with advancements in AI technology mirroring the progression of our understanding of artificial neural networks since the 1950s. These improvements have refined systems to better interact with humans using human language, allowing them to take actions that appear intelligent based on vague or incomplete instructions, and even to create entirely new concepts. This progress is remarkable and impressive.

However, there are multiple types of AI. In the past, computers were limited to running straightforward decision trees—sequences of yes/no questions that would lead to a conclusion. As computer power increased, the complexity of AI models expanded as well. Modern AI simulates neural networks found in human and animal brains, a concept that was first proposed in the 1950s. Each subsequent decade has seen further refinement and refinement, culminating in AI models capable of interacting with humans in a human-like manner.

Despite these advancements, there remains a significant barrier to creating truly general-purpose artificial neural networks - the lack of real-world experience within the digital information that feeds these networks. Some projects have been initiated to standardize and digitize basic knowledge, allowing neural networks to learn from what humans already know simply by existing.

A major shift has occurred as a result of the proliferation of digital information; our lives have become increasingly online, and various fields—from physics to biology, sociology to psychology, art, music, and beyond—have transitioned a substantial portion of their work to digital format.