Let's stay aware AIs developmental bias, especially when they're "right" 🫣

It is well known that ChatGPT, Gemini, Claude, Grok and all the other LLMs are extremely good at presenting information that seems real but is completely made up, fake, and often dead wrong. Even with loud disclaimers, they’re going to get smarter and we’re going to forget this! Let us remember! They’re often wrong! 

There’s something even more pernicious and unconscious that is much harder to spot, and much harder to correct. The LLMS are developmentally biased. If this doesn’t change, the more you use the AI, the more you put a developmental cap on your own maturity of meaning-making and self-identity.

Almost all of the conversation about bias is happening within a pluralistic frame—a “Green” set of values hyper-focused on demographic, racial, political biases, power imbalances, and egalitarianism. The issue is not about facts, it’s about the way facts are presented, such that even a “correct” piece of information can be presented in a misleading way. 

Here’s one of infinite examples: Someone in a recent conversation thread of mine quoted ChatGPT’s definition of “colonialism”: 

“Colonialism is the practice or policy by which a nation establishes and maintains control over foreign territories, exploiting them for economic, political, and cultural gain. It often involves the settlement of the colonizing population in the territory, the imposition of the colonizer’s systems of governance, and the subjugation of indigenous populations. Historically, colonialism was marked by the extraction of resources, suppression of local cultures, and the justification of domination through ideologies such as racial superiority or “civilizing missions”.”

It’s not wrong, maybe it’s subtle, but it sure has a different feel from this Oxford English dictionary definition: 

noun: The policy or practice of acquiring full or partial political control over another country, occupying it with settlers, and exploiting it economically.
"the state apparatus that was dominant under colonialism" 

The AI definition includes postmodern critique (green), the dictionary reflects a modern (orange) bias, presuming there’s an objectivity that exists outside of meaning-interpretation-coenactment. But we know the epistemological stance of the dictionary, we expect it, and we don’t give it more authority than it deserves (and ironically it usually presents multiple definitions). Let us know this about AIs; even when we use developmentally signified training data, let us include this knowledge in our awareness.

I’ve written about this AI developmental bias before, but it’s important enough that I want to say it again and try to say it more clearly: AI is developmentally biased*, this shapes how information is framed and understood in a way that is mostly invisible to most people—emphasizing pluralism, egalitarianism, and relativism, while often missing or even ‘dumbing down’ higher-order complexity (e.g. meta-systemic, integral). We should be taking that into account.

To be clear, I like AI—I used AI to help me clarify this piece of writing. We just have to stay frosty.

*This isn’t just my opinion—in late 2022 Integral Life put GPT3 through world-renowned developmental Loevinger/Cook-Greuter’s Sentence Completion Test and found its responses smack-dab in the middle of orange-green—which is exactly what I would have guessed from my interactions, and predicted from the training data coming from the median internet user!  Although LLM capabilities have dramatically increased, I don’t think developmental maturity has at all.

With love, Jordan