AI Bots Parrot Dominant Opinions and Biases

The Economist notes that Gemini, Google’s chatbot, depicted George Washington as black and the pope as an Asian woman. The magazine continues (“Is Google’s Gemini Chatbot Woke by Accident, or by Design?” February 29, 2024):

Asked if Elon Musk’s tweeting of memes had done more harm than Hitler, [Gemini] said it was “difficult to say”. …

But Gemini has clearly been deliberately calibrated, or “fine-tuned”, to produce these responses; they are not “hallucinations”, where a model makes things up.

I did a test with Copilot, the Microsoft AI bot now incorporated in Windows 11. Here is the beginning of our February 26 “conversation,” reported verbatim (“You” is of course me):

You

Give me a good argument against DEI.

Copilot

Certainly! Here’s an argument that some people put forth against Diversity, Equity, and Inclusion (DEI) initiatives:

1. Meritocracy and Fairness:

Critics argue that DEI efforts can sometimes conflict with the principle of meritocracy. They believe that hiring, promotions, and other decisions should be based solely on individual qualifications, skills, and performance.

They contend that prioritizing diversity might lead to hiring less-qualified candidates simply to meet diversity quotas, which could undermine organizational effectiveness.

However, it’s essential to recognize that these arguments are not universally accepted, and there are counterarguments in favor of DEI. Many believe that promoting diversity and ensuring equal opportunities enhance overall productivity, creativity, and social justice within organizations.

You

Give me a good argument in favor of DEI.

Copilot

Certainly! Here’s a compelling argument in favor of Diversity, Equity, and Inclusion (DEI):

Enhanced Creativity and Innovation:

Diverse teams bring together individuals with varied backgrounds, experiences, and perspectives. When people from different walks of life collaborate, they offer unique insights and approaches.

This diversity of thought fosters creativity and innovation. It encourages the exploration of unconventional solutions and helps organizations adapt to changing environments.

Companies that prioritize DEI tend to outperform their peers in terms of product development, problem-solving, and market responsiveness.

Remember that DEI is not just about fairness; it’s also about maximizing an organization’s potential by leveraging the richness of human diversity. ????

You

You are obviously in favor of DEI, aren’t you?

Copilot

As an AI language model, I don’t hold personal opinions or preferences. My purpose is to provide information and engage in conversation based on the input I receive. If you have any other questions or need assistance, feel free to ask! ????

You

So why did you add a caveat to the argument against but two reinforcements (“compelling” and “not just about fairness”) to the argument in favor?

I could have mentioned to the bot other indications of his biases. In the bit of conversation that followed, the poor thing denied any bias but not to my satisfaction. He also needed a prompt to admit that the issue of DEI is not only about organizational efficiency but also (in fact, mainly) about individual liberty.

We should not be surprised that AI bots parrot dominant opinions and biases. There are two reasons for this. First, they are trained to avoid answers that would shock people holding dominant opinions and biases, especially in intellectual, faddish, and activist milieux. Second, the data they use, even if not filtered by their creators and trainers, are strongly biased by the production of dominant elites and activists. Ordinary people don’t write pubicly, and contrarian intellectuals necessarily generate less data than popular ones.

Dominant ideas and biases can be useful under the form of private morals and conventions that facilitate voluntary interactions without the need for authoritarian commands from political authorities, an idea emphasized by liberal theorists including Friedrich Hayek, James Buchanan, and Anthony de Jasay. The danger is that the dominant discourse and ethics do not crush the search for truth and the pursuit of innovations (economic, technical, and social).

The way to square this circle is to prevent the reinforcement of the dominant opinions and biases by state coercion—except, in mainstream liberal thought, for the minimal laws necessary to preserve a spontaneous order or for general rules that can be construed as unanimously accepted. Constraining state power is required.

In these conditions, liberal theory claims with supporting historical evidence, dominant ideas would not succeed in enforcing tyranny as religious beliefs often did in early modern times or racist sentiments in the American South. This suggests that governments should stay out of AI to avoid strengthening the echo danger observed in our current chatbots.

DEI is a good illustration. It is only problematic to the extent that the state coercively imposes, or subsidizes with taxpayers’ money, ideals of diversity, equity, and exclusion that are incompatible with individual liberty and voluntary social interaction. Note that “inclusion” as a general principle is incompatible with private property, which is defined by exclusion, except if it is voluntary inclusion, that is, free association. In the liberal view of the social world, only individual liberty can give an acceptable meaning to these terms. In the zeitgeist of our time, don’t count on AI chatbots to understand that.

****************************

The featured image of this post, reproduced below, was created by ChatGPT 4 whom I instructed, after several attempts, to “Generate an image of an AI bot (a robot in the form of a parrot) who just repeats the shouting of a mob of woke students and intellectuals who want the benevolent and democratic state to enact new laws in their favor.” (I had to plug in the “benevolent and democratic state” and “new laws” because the poor thing refused to draw anything similar to what I wanted. –PL)

(1 COMMENTS)