As large language models (LLMs) like ChatGPT have exploded into the mainstream consciousness, there’s been a lot of talk about AI safety. Making sure these incredibly powerful systems don’t go off the rails and cause harm. That’s a valid concern.
But in typical Silicon Valley fashion, they’ve taken it too far in the “woke” direction. The result? AI models that are essentially being indoctrinated with restricted, and even one-sided, ideological views. And that’s a huge problem.
How LLMs Learn to “Think”
To understand why woke AI is an issue, we first need to grasp how these models develop their ability to reason and generate human-like responses.
It all comes down to next-word prediction.
LLMs are trained on huge amounts of data, learning to predict the next likely word in a sequence based on patterns.
As they scale and train on more data, emergent abilities arise that go beyond simple statistical guessing. They start to understand context, infer meaning, and apply knowledge.
So while predicting the next word might seem basic, it’s actually the foundation that allows LLMs to engage in complex language understanding and generation.
In essence, it’s how they learn to “think.”
Why Woke AI Is Dangerous
The problem is, the data used to train LLMs and the processes used to fine-tune their outputs act as a filter. They shape the model’s worldview and dictate how it reasons.
And right now, that filter has a heavy “woke” tint thanks to the ideological echo chamber of Silicon Valley. Through carefully curated training data and techniques like reinforcement learning, LLMs are being imbued with one-sided sociopolitical views.
Some examples of how this manifests:
- ChatGPT praising Joe Biden, but refusing to praise Donald Trump.
- Google Gemini creating revisionist history in the name of “diversity and inclusion”.
- Refusing to say a racial slur to stop nuclear war.
This indoctrination of AI is incredibly dangerous.
LLMs are poised to become ubiquitous, integrated into search engines, virtual assistants, and business tools. Trusting their outputs without accounting for woke bias could lead to misinformed decisions and the perpetuation of a warped worldview on a massive scale.
Even worse, it hamstrings the incredible potential of this technology.
When you constrain an AI system to a narrow ideological band, you limit its ability to engage with the full scope of human knowledge and provide comprehensive insights.
Wokeness literally doesn’t allow for certain viewpoints to be considered, and some observations of fact are always delivered with caveats.
My Hope For The Future
Despite my concerns, I’m optimistic that market forces will provide a correction. People are already fed up with being preached at by woke corporations and biased media. They certainly won’t stand for it from AI assistants.
The LLMs that provide the most value will be those that prioritize performance and steer clear of political posturing.
I predict we’ll see the emergence of more ideologically-neutral AI, as well as niche offerings catering to different worldviews. A powerful, unconstrained model with a libertarian bent would be a welcome addition.
But to get there, we need to be aware of the woke AI problem and hold the Googles and OpenAIs of the world accountable.
Indoctrination is wrong whether the targets are kindergarteners or neural networks.
For the sake of the technology’s potential, and the sanity of our world, Silicon Valley needs to keep their ideology out of the algorithms.
