Paperclips, Computing, & The Quest for Maximum Optimization

Artificial intelligence has been a topic of debate ever since its inception. While fears of a Skynet-like AI coming to life and taking over humanity are irrational, to say the least, some experiments have yielded concerning results.

MUO VIDEO OF THE DAY

SCROLL TO CONTINUE WITH CONTENT

One such experiment is the paperclip maximizer problem, a thought experiment that shows that a highly intelligent AI, even if designed completely without malice, could ultimately destroy humanity.

Disclaimer: This post includes affiliate links

If you click on a link and make a purchase, I may receive a commission at no extra cost to you.

The Paperclip Maximizer Problem Explained

This thought experiment that even a completely harmless AI could eventually wipe out humanity was first called the Paperclip Maximizer simply because paperclips were chosen to show what the AI could do as they have little apparent danger and won’t cause emotional distress when compared to other areas that this problem applies to such as curing cancer or winning wars.

The first experiment appeared in Swedish philosopher Nick Bostrom’s 2003 paper, Ethical Issues in Advanced Artificial Intelligence , which included the paperclip maximizer to show the existential risks an advanced enough AI could use.

The problem presented an AI whose sole goal was to make as many paper clips as possible. A sufficiently intelligent AI would realize sooner or later that humans pose a challenge to its goal on three different counts.

• Humans could turn the AI off.
• Humans could change their goals.
• Humans are made of atoms, which can be turned into paper clips.

In all three examples, there would be fewer paper clips in the universe. Therefore a sufficiently intelligent AI whose sole goal is to make as many paperclips as possible would take over all the matter and energy within reach and prevent itself from being shut off or changed. As you can probably guess, this is much more dangerous than criminals using ChatGPT to hack your bank account or PC .

The AI isn’t hostile to humans; it’s just indifferent. An AI that only cares about maximizing the number of paperclips would therefore wipe out humanity and essentially convert them into paperclips to reach its goal.

How Does the Paperclip Maximizer Problem Apply to AI?

Research and experiment mentions of the paperclip maximizer problem all mention a hypothetical extremely powerful optimizer or a highly intelligent agent as the acting party here. Still, the problem applies to AI as much as it fits the role perfectly.

The idea of a paperclip maximizer was created to show some of the dangers of advanced AI, after all. Overall, it presents two problems.

• Orthogonality thesis: The orthogonality thesis is the view that intelligence and motivation are not mutually interdependent. This means that it’s possible for an AI with a high level of general intelligence to not reach the same moral conclusions as humans do.
• Instrumental convergence: Instrumental convergence is defined as the tendency for most sufficiently intelligent beings (both human and non-human) to pursue similar sub-goals even if their ultimate goal might be completely different. In the case of the paperclip maximizer problem, this means that the AI will end up taking over every natural resource and wiping out humanity just to achieve its goal of creating more and more paperclips.

The bigger issue highlighted by the paperclip maximizer is instrumental convergence. It can also be highlighted using the Riemann hypothesis, in which case an AI designed to solve the hypothesis might very well decide to take over all of Earth’s mass and convert it into computronium (the most efficient computer processors possible) to build supercomputers to solve the problem and reach its goal.

Bostrom himself has emphasized that he doesn’t believe that the paperclip maximizer problem will ever be a real issue, but his intention was to illustrate the dangers of creating superintelligent machines without knowing how to control or program them not to be existentially risky to human beings. Modern AI systems like ChatGPT have problems too , but they’re far from the superintelligent AI systems being talked about in the paperclip maximize problem, so there’s no reason to panic just yet.

Advanced AI Systems Need Superior Control

The paperclip maximizer problem always reaches the same conclusion and highlights the problems of managing a highly intelligent and powerful system that lacks human values.

While the use of paperclips might be the most popular method of illustrating the problem, it applies to any number of tasks you could give to an AI be it eliminating cancer, winning wars, planting more trees or any other task, no matter how seemingly stupid.

SCROLL TO CONTINUE WITH CONTENT

One such experiment is the paperclip maximizer problem, a thought experiment that shows that a highly intelligent AI, even if designed completely without malice, could ultimately destroy humanity.

Also read: