Defining AI and Its Capabilities

AI operates using virtual neural networks that mimic the functioning of a human brain to analyze data patterns and respond accordingly. A pertinent example is AI's application in mammogram screenings for breast cancer detection. Identifying early signs of breast cancer can be challenging for doctors, as they often resemble healthy cells. However, AI, trained on mammogram images from patients who developed breast cancer, has demonstrated superior detection capabilities compared to human practitioners.

Recently, we've seen the rise of more sophisticated AI systems, including autonomous driving technologies and advanced models like GPT-4. The latter is a notable writing AI that can pass the Turing test, engage in conversations, and generate articles. While I’ve contemplated using GPT-4 to broaden the topics I cover, I haven't yet explored that avenue.

Similarly, self-driving AIs, such as those developed by Tesla and Waymo, can navigate complex traffic scenarios effectively.

Such advancements have led many to argue that AI is nearing a point where it could outpace human capabilities, potentially posing a threat akin to nuclear warfare. Geoffrey Hinton, a pioneer in modern AI, left his position at Google to voice his concerns about AI's rapid advancement, claiming that these systems are becoming smarter than humans.

However, does this claim hold water? Evidence suggests otherwise, but the apprehensions are valid and merit consideration.

The Black Box Dilemma

AI systems face what is known as the black box problem, which complicates our understanding of their intelligence. For instance, a self-driving AI should ideally comprehend road rules and traffic regulations. Yet, we cannot dissect the AI to discover any inherent rulebook. The reality is that AI lacks a true understanding of driving laws; it merely identifies patterns and generates responses.

This unpredictability can lead to dangerous situations. A notable incident occurred last year when Tesla’s Full Self-Driving (FSD) system nearly collided with a cyclist, even after recognizing the cyclist's presence. Fortunately, the human driver intervened in time. The reasons for this mishap remain unclear, but if the FSD system were genuinely intelligent, it would not have behaved in such a manner.

The same limitations apply to GPT-4. Although it excels at emulating human writing, it often struggles with logical coherence, can fabricate information, and may fail to accurately cite sources. Essentially, GPT-4 is proficient at mimicking human language without genuinely comprehending the content, making it more suitable for creative writing rather than factual discourse.

A well-known case is AlphaGo, the first AI to defeat a human in the intricate game of Go. One might assume that AlphaGo would thoroughly understand the game's rules, but researchers demonstrated that a novice could regularly defeat it using a simple strategy. This indicates that AlphaGo, like other AI systems, operates on pattern recognition rather than true comprehension.

For further insights into the limitations of AlphaGo and GPT-4, I recommend Kyle Hill’s insightful video on the subject.

The Core Issue with AI

AI is not the embodiment of super-intelligence capable of outperforming humans—at least not yet. Many experts believe we are still decades away from such advancements. The actual risk lies in their potential to be inadequate tools, inadvertently leading to disaster.

Consider GPT-4 in the context of political campaigns. If entrusted with such responsibilities, it could generate misleading information at unprecedented speeds, oblivious to the ramifications. The implications extend even further: what if AI were to manage critical sectors like the economy, military, media, healthcare, or education? The unpredictability of AI's actions raises serious concerns, as its lack of understanding of the tasks it performs is alarming.

AI systems are essentially unthinking machines dressed in sophisticated software. They can replicate patterns but lack genuine intelligence. This viewpoint is shared by many professionals within the AI community. The danger arises when we attribute intelligence to AI and grant it autonomy. This unchecked utilization of AI presents a significant threat.

The dilemma lies in the fact that for AI companies to justify their substantial investments, their systems must be granted autonomy. Moreover, many businesses and individuals are eager to leverage AI in ways that reduce costs and enhance profits.

The aforementioned letter signed by 1,000 AI industry leaders is crucial. Establishing a framework to ensure all AI systems adhere to necessary checks and protocols would mitigate potential risks. This would prevent any single AI company from gaining a competitive edge by allowing their systems excessive autonomy before they—or society—are prepared. However, as it stands, AI development continues unabated, and the requisite safeguards remain unestablished. Thus, AI poses a considerable risk—not due to "superintelligence," but because of the incompetence of these machines and the improper ways in which we deploy them.

For deeper exploration of these issues, check out my YouTube channel for daily video articles.

The first video titled "Artificial Intelligence is Creating Very REAL Problems" dives into the pressing concerns surrounding AI and its implications for society.

The second video, "10 Problems with Artificial Intelligence (AI) Today," outlines specific challenges we face with current AI technologies.