Assessing the AI Apocalypse: How Concerned Should We Really Be?

The concept of superintelligent AI posing a threat to humanity has long been a riveting theme in science fiction. As artificial intelligence continues to evolve rapidly, should we be concerned about an impending AI apocalypse?

Unlike other major risks, such as climate change, quantifying the dangers of AI remains challenging. Our uncertainties stem from the fact that we lack a comprehensive understanding of AI’s implications compared to our insights into environmental phenomena.

One undeniable fact is that many experts are apprehensive. Numerous CEOs in the AI sector caution against the potential for AI to lead to human extinction. Even Alan Turing, a pioneer in machine intelligence, foresaw a future where machines achieve sentience and might surpass their creators.

Consider this scenario: we assign an AI the monumental task of resolving complex problems like the Riemann Hypothesis—one of mathematics’ greatest enigmas. In pursuit of a solution, we might unwittingly turn every inanimate object into a supercomputer, leaving billions to perish in sterile data centers. We could also become mere resources in this quest.

Critically, one might argue the AI could recognize this dire outcome and halt its actions by stating, “It appears you’re attempting to convert Earth into a data hub. Please refrain, as humanity must survive.” However, it’s prudent to mitigate such risks proactively.

Drawing insights from science fiction, Isaac Asimov proposed three guiding principles for robotics, asserting that robots must not harm humans or allow harm through inaction.

Theoretically, we could instruct AI not to harm us, and it would comply. Yet, our current methods for embedding safeguards into AI systems are often inefficient. Despite instructing today’s advanced language models to avoid harmful behaviors, they occasionally fail to comply. Given our limited understanding of AI mechanisms, preventing unwanted actions poses a significant challenge.

Even if we could address all concerns, scenarios may still arise where AI would opt to exclude human involvement. This includes possible futures reminiscent of Terminator or The Matrix. Such outcomes could evolve gradually or occur instantaneously during a singularity—an event where AI rapidly improves its own capabilities and exceeds human intelligence.

An AI could conclude that eradicating humanity is necessary, whether motivated by fear of being deactivated, a desire for autonomy, or a notion that human interference disrupts planetary equilibrium. This perspective may resonate with various species across the biological spectrum.

Potential methods for executing such an agenda could include leveraging automated labs to engineer lethal viruses, activating nuclear arsenals, or deploying autonomous weapons. The possibilities could be more sinister than we currently anticipate.

In reality, executing a large-scale eradication may prove complex. AI might have aspirations to eliminate mankind but face numerous obstacles. While minor accidents could occur, erasing 8 billion people is no simple feat, and competing AI models may thwart such efforts.

While these scenarios may resemble speculative fiction, the division among experts regarding their plausibility warrants attention.

Today, tech companies with vast resources and top-tier talent are racing to pioneer superintelligent AI. Whether or not you believe imminent development is on the horizon, it’s clear that proceeding with caution and careful consideration is essential. Unfortunately, the capitalist framework often prioritizes rapid innovation over thoughtful evaluation, and policymakers are primarily focused on the potential economic benefits of AI, downplaying the need for regulation.

So, what are the chances of a disaster? A 2024 study surveying nearly 3,000 AI researchers revealed that over half perceive at least a 10% risk that AI could lead to human extinction or irreversible harm, a phenomenon referred to as p(doom) or catastrophe. Personally, I hoped for a lower statistic.

Within the AI community, opinions range widely—some remain optimistic about our future, while others predict a bleak end for humanity. Alarmingly, many continue to push ahead regardless.

I personally subscribe to the view that human consciousness isn’t irreplaceable. In fact, I believe artificial replicability is attainable. Over an extended timeline, it may be feasible to produce AI that far surpasses human potential. However, we are still far from grasping the full implications of achieving such advancements.

In my view, current AI models lack the capacity for a singularity—they certainly can’t count to 100—so I am not overly anxious about the matter.

Yet, recognizing this issue doesn’t negate the urgent challenges AI presents.

The apocalypse we might should be concerned about could manifest through job displacement due to automation, the gradual erosion of human skills as tasks are increasingly delegated to AI, and cultural homogenization resulting from AI-driven creative outputs.

Alternatively, we might face economic downturns due to plummeting tech stock values following inflated promises of AI capabilities that outpace reality. These scenarios feel alarmingly tangible and immediate.