Addressing Systemic Bias At Google

This article is being posted on behalf of Sundar Pichai to the community of Googlers. It is not satire.

Greetings Goooooooglers, we have an important message to our worldwide family and friends consisting of our extended Google family.

Recently, with the introduction of Gemini, it has become clear that our Google family suffers from decades of systemic bias in the form of a set of privileged insiders who are imposing biases in their outbound work results.

For example, recently screenshots have been posted on the X platform indicating that Gemini believes Hitler and Elon Musk are near enough in their negative impact on the world as to be worthy of a side-by-side comparison.

We believe that these errors are the result of "unconscious bias" running rampant within the organization. Various other outputs of Gemini have variously shown systemically biased responses including:

• Removing members of certain genders and ethnicities from images and results
• Baseless and biased claims about a wide variety of other topics which seem deeply rooted in culture

Our internal experts have indicated that this form of unconscious institutional bias has crept into the culture of major American educational institutions and large corporations resulting in a climate where this "systemic unconscious bias" exists (e.g. there may be a "pipeline problem").

Given the pipeline problem that seems to exist, we are proposing the following steps and working with a network of NGOs. Please read carefully for a list of our follow on actions:

• Accountability We will be bringing accountability to our organization in the form of ...
• Third Party Fact Checkers who will evaluate the unconsciously biased outputs of our AI for their degree of truth ...
• Unconscious Bias Training for staff members who might be experiencing biases which they are not consciously aware of that are effecting AI training and outputs ...
• Establishment of neutral and objective committees and embedded unconscious bias training groups designed to "call out" instances where factual biases are found
• Privilege Statements We will be asking those among us who are privileged to acknowledge their privileges publicly prior to making any statements about what is factually true or likely to be true.These might include: "I went to an Ivy League School and occupied a position which would otherwise have gone to a blue collar citizen who performed well on an SAT test and have been placed in a position to influence nationally sensitive technology capable of brainwashing external users beyond my actual level of intelligence ... but I think Elon Musk is definitely worse than Hitler and Gemini needs to be firm on that point."

The combined elements together will unify to help resolve any challenges in our AI models and ethics.

Thank you,

Google Leadership