Data-Driven Loophole: How to Convince Your AI of Anything and Everything

 

We love making comparisons of AI to humans. We give it humanoid characteristics and human-like brain structures. We teach it through human methods by giving it rewards and penalties in reinforcement training. However, we often lose sight of what AI truly is, and what it is not.

The Reality of AI

AI is a data-driven algorithm with the ability to mimic human language and behaviors. It is heavily reliant on data, which makes it relatively easy to cater to specific contexts. This data-driven nature, however, is also what makes it so easily exploitable.

Image created by Microsoft Copilot

The Human vs. AI Paradigm

Unlike humans, AI lacks the complex nuances that come with human experience and decision-making. For instance, one of my coworkers is a vegetarian not because of a love for animals, but due to his religious background in Hinduism. Although he no longer actively practices his religion, he remains steadfast in his vegetarianism. His family believed he would eat meat once he moved to Canada, but he continues to maintain his dietary choices, even lying to his family to avoid confrontation. This kind of complex, multi-dimensional decision-making is inherently human.

In contrast, AI does not possess this fundamental capacity. If presented with sufficient data disproving its previous belief, it will simply adjust. Its reliance on data makes it corruptible; for instance, one could theoretically convince an AI that the Earth is flat or that a lizard super race is real by skewing the data it receives.

Data Manipulation and AI

AI operates on hypotheses derived from data, and these hypotheses get updated with new data. This creates a loophole: manipulate the data, and you manipulate the AI. Essentially, it's a built-in confirmation bias.

Humans also form beliefs based on evidence and data, and we adjust our thinking as we learn and grow. Yet, many of us also hold steadfast beliefs in the unknown and unseen, such as irrational fears, religion, and spirituality. These core beliefs are a fundamental trait of humankind, making us complicated creatures. On the other hand, AI is reliable and deterministic, capable of deriving theories from data and making predictions based on its internal formula.

The Love of AI

Despite its limitations, I love AI for the potential it holds. AI hallucinations are one thing, but influencing AI to stick to a strange belief system is another. Marketing and showy products might like us to believe that AI is near-human. Fiction often portrays AI robots as something to fear. Humanity's vanity drives us to build machines in our image, convincing ourselves that AI is similar to us, with the same emotional capacity and social needs.

However, at the end of the day, we must understand AI for what it truly is—a data-driven, pattern-detecting, retrieval agent. Although fascinating and promising, AI is ultimately an algorithm, a non-human entity.

Moving Forward

As we continue to develop and integrate AI into our lives, it's essential to recognize its strengths and limitations. Understanding the potential for data manipulation and the inherent differences between human and AI decision-making will help us navigate the complexities of this technology responsibly.