Gemini’s madness is the true meaning of “unbiased” AI

opinion

by William A. Jacobson and kemberly kay

Everyone is laughing at the rollout of Google Gemini AI. But it’s no joke.

This problem is even worse than historically inaccurately generated images.

Manipulating AI is just one aspect of a broader “algorithmic discrimination” built into American companies that could lead to job losses and more.

Gemini was asked to create a photo of a white person, but refused, saying the request would “reinforce harmful stereotypes and generalizations about people based on race.”

But it had no problem generating photos of a female pope, a non-white Viking, and a black George Washington.

Microsoft’s AI imaging tools have their own issues with producing sexually explicit and violent images.

Clearly, AI imaging is off track.

Google’s CEO admitted that Gemini’s results were “biased” and “unacceptable,” but said that was a feature, not a bug, and that “anti-racism” theories are openly racist It’s the same thing that created the practice of equity and inclusion.

As one of us (William) recently explained to the Post, “In the name of anti-bias, real bias is being built into the system. This is not just about search results, it’s about being ‘unbiased.’ It’s also a concern for real-world applications where algorithmic tests build bias into the system by targeting the end result that actually hits the quota. ”

Almost a year ago, our Equal Protection Project (EqualProtect.org) sounded the alarm by exposing the use of algorithms to manipulate the pool of job applicants in LinkedIn’s Diversity in Hiring feature.

LinkedIn justified its manipulation of race and other identity groups as necessary “to ensure people have equal access to employment opportunities,” but what “equal access” actually means is It was a preferential treatment.

Such biases operate in the shadows. Job seekers don’t understand how algorithms affect their prospects.

Algorithms can and are used to elevate certain groups over others.

But this isn’t limited to LinkedIn.

The Biden administration has issued an executive order requiring unbiased algorithms, but under the policy’s progressive DEI rubric, lack of bias is evidenced in “fairness” rather than equal treatment. There is.

Equity is a slang term for allocation.

In the world of “unbiased” algorithmic testing, bias is built in to achieve fairness.

What happened with Gemini is an example of such programming.

Getting bad search results is another thing. Losing employment opportunities is something else entirely.

As attorney Stewart Baker, an expert on such deck stacking, explained at an EPP event: . . Code in artificial intelligence will most likely be code for imposing stealth quotas. ”

The insidious reach of “open-minded” prejudice will expand.

Algorithmic discrimination can manipulate every major detail of our lives to obtain group outcomes or group assignments.

These algorithms are designed to remove the threat of DEI and smuggle it into every aspect of life and the economy.

People are Intentionally “Teaching” the AI ​​that an image of a black Viking is a fairer outcome than the truth.

Big Tech already knows so much about you, including your race and ethnicity, that it’s not hard to imagine discrimination by algorithms that manipulate access to a host of goods and services.

Have you been turned down for a job, loan, apartment, or college admission? There may be a “bias-free” algorithm at work.

But it’s almost impossible to prove it. That’s because the algorithms operate in secret, out of sight, and are certified as “bias-free.” because They build bias into the system to meet quotas.

You get the picture.

Algorithmic discrimination is a threat to equality and must be stopped.

William A. Jacobson is a professor of clinical law at Cornell University. equal protection projectwith Kemberly Kay serving as administrative and editorial director.