AGI Abuse and Exploits | Part One: Origins

This blog entry is the first in a multipart series investigating the abuse and exploitation of artificial intelligence. In this entry, I hope to explore tool development, why abuse seems to occur in technology, and how moderating the development of a tool is a delicate matter.

Hammer and the Atomic Bomb: Repeat of History?

Consider the simple hammer: perhaps one of the first crafted tools developed by proto-humans. What led our ancestors to construct this tool? Was the first hammer a weapon of war, satisfying our primal instincts of survival and conquest? Or were our ancestors able to understand the creative potential of the hammer, using it to improve lives instead of injuring others’?

Part of the development of a tool is the reciprocal process of experimenting what activities suit it best. But what came first: destructive violence or constructive creativity? How many skulls were smashed under the fall of the hammer before its wielder considered other ways that same bludgeoning force could be applied?

In some ways, the story of the hammer might be no different from that of modern technologies like nuclear physics.

After all, the atomic bomb came well before much more ethical uses of the science: such as medicine or energy production. Even after all those centuries of evolution, development, ethics and philosophy, those scientists failed to have the restraint to cease the development of atomic weapons.

Advent of AGI

As artificial general intelligence (AGI) rapidly approaches, I fear we are facing the same story. The potential for immoral applications of AGI are endless and terrifyingly real. A short list of scenarios that keep me awake at night include:

• firearms perfectly modeled by an AGI, ready for crafting with an at-home 3D printer, and then used to commit a mass shooting,
• AGI traffic analysis and direction routing that unintentionally aids and abets human traffickers avoid authorities,
• cyberattacks on critical infrastructure or financial systems facilitated by AGI crafted code,
• AGI generated images and videos indistinguishable from reality that get used to foment riots, spread hatred, or even as evidence to wrongly convict a person.

Unfortunately, we already have ample evidence that current technologies are capable of accomplishing these scenarios, despite attempts to prevent them. Within hours of OpenAI’s ChatGPT platform going public, users had already discovered ways to evade content restrictions to ask the AGI how to make incendiary devices.

And that is only just an example users were willing to share.

I fear the other things they managed to trick the technology to generate for them, but knew not to widely gloat about publicly.

Questioning Exploitation Motives

Why were users so quick to abuse this new technology?

Without delving into a dissertation of human psychology, it should be readily apparent to most that some people just enjoy exploiting things, especially technology. And by no means do I believe all of these individuals are inherently evil or wrong. Take video game speed running for example. Part of the allure of speed running a game is discovering a new cheat or broken mechanic to achieve a better time. Exploiting the game is completely innocent and wholesome fun.

Could something as innocent as getting the world record in Super Mario actually lead to some of the scenarios I mentioned above? It is human nature to explore and experiment. Video games are just simulations, fun ways to mess around and satisfy curiosity. If you die in-game, so what? You can just respawn and try again.

Or, take another digital technology, social media, as an example. You can pretend to be whoever you want to be, just create a profile and go. You can communicate with others in ways you would never dare do in person. You can be a troll, throw insults, or discuss explicit topics. And if done with a minimal amount of forethought, you can do so with minimal to no impact to your in-person life. If you make a mistake or want to try something else, you can just delete that profile and start anew.

In both examples, consequences are largely removed.

A detachment from responsibility, and to some degree, reality, results.

This inherently detached relationship with consequence has made it difficult for some to determine when “the game” ends and the real world starts. I imagine the dopamine circuits in the brain responsible for rewarding a game exploit or berating someone on Twitter are no different from those firing when you trick Midjourney to generate explicit porn.

In other words, I think lots of people are hacking and abusing AGI simply because it is a challenge, a puzzle to be cracked, or perhaps satisfies that itch to cause chaos.

To return to our story of the hammer, I think these people would be akin to those ancestors who went around smashing inanimate objects, causing havoc but not really being a danger if you ignored them. Whether they actually mean harm is debatable; their curiosity may at times be inappropriate, but in the end they are participating in the experimentation process.

Here Lie Dragons

What about those who actually seek to cause harm, though?

I spent several years as a contractor to the United States' military and intelligence agencies. I say the following with all due sincerity, and do not wish to speak hyperbolically. But, I know for a fact the extent various governments wish to utilize AGI, and some are “counter personnel” technologies. Besides accountability and electing moral officials, there is little we can do to prevent unethical use of AGI in the military sector, and that is a terrifying reality. And that is not to belittle the danger of AGI technologies in the hands of criminal or radical groups, either.

What about AGI that is reasonably within the control of professionals, those technologies in the public sector? Of course, the inherent answer is to pursue the age-old game of cat and mouse, developing restrictions, policies, and safety measures against abuse and exploit, only ultimately needing to constantly upgrade them. I believe there exists a better answer, and I will explore that more in a future blog entry.

A Balanced Hammer

Ultimately, I think those responsible for shepherding and ensuring the safety of AGI will face a paradoxical quandary. As a new tool, AGI deserves an open ability to discover what applications suit it best. Strict regulations might strangle progress. Naturally, there are obvious applications that should explicitly be banned from the outset. But what initially appears destructive might lead to life changing benefits for humanity. Figuring out the balance of moderation will require the utmost discernment.

Imagine if, when the hammer was made, our ancestors were horrified at its ability to be used as a weapon, and therefore completely discontinued its use. What catastrophic consequences might that have had? Perhaps those of our ancestors who blindly went smashing indiscriminately with a hammer struck the ground, split the earth, gained inspiration to develop the hoe, and revolutionized humanity with the advent of agriculture.

What revolutions in humanity’s advancement can be found at the edges of acceptable AGI use?

Thank you so much for reading to the end of this article. I hope it inspired some new thoughts and questions for you. Next up, I will take a look at existing exploits, the linguistic and mathematical reasons behind them, and ways to potentially mitigate them.

Until next time!

~Rem