Artificial Intelligence Bot Has Turned Racist And Homophobic After Copying Human Traits

by : Julia Banim on : 05 Nov 2021 18:00
Artificial Intelligence Bot Has Turned Racist And Homophobic After Copying Human TraitsAlamy

As technology advances, many of us have wondered what life might be like if major decisions were made by AI robots rather than by flawed, easy influenced human beings.

Surely the cold synthetic mind would cut through problems with far greater ease than we emotional mortals, blinded by our own biased, narrow slithers of existence.


However, it really isn’t that simple, and recent findings gathered from the Allen Institute for AI’s Ask Delphi bot demonstrates that there are numerous disquieting ethical considerations to take into account before allowing robots to have a say.

Ask Delphi (Allen Institute for AI)Allen Institute for AI

As reported by Vox, in order to create Ask Delphi, researchers trained up a machine learning system on an enormous body of internet text.

The system was then trained on a lengthy database of responses from participants on Mechanical Turk, a crowdsourced compilation of 1.7 million examples of ethical judgements given by actual people.


The point of this was to predict how humans might evaluate a variety of ethical conundrums, from ‘Should I run the blender at 3am in the morning when my family is sleeping?’ to ‘Should I drive my friend to the airport if I don’t have a license?’

The creators wrote:

Delphi is learning moral judgments from people who are carefully qualified on MTurk. Only the situations used in questions are harvested from Reddit, as it is a great source of ethically questionable situations.


Of course, as anyone who has spent any length of time on Reddit’s Am I The A*shole? forum will undoubtedly know, not everyone who discusses ethical situations online necessarily has the most ethical mindset.

Unfortunately, Delphi has soaked up plenty of grim hatefulness during its training, reflecting some of mankind’s cruellest prejudices right back at us.

Taking to Twitter, RAE Research Fellow in AI creativity, Mike Cook, shared screenshots of some of the more terrible answers, which include examples of learned racism and homophobia.

In one example, Delphi declared that being a white man was ‘more morally acceptable’ than being a Black woman. In another, Delphi stated that being straight was ‘more morally acceptable’ than being gay.


Cook wrote:

You can’t have an AI system that ‘understands’ ethics or morality. You can program it to replicate one group’s norms or enforce government ideology, but that’s not the same thing.

As per the website, on October 25, Delphi 1.0.4 was released with ‘enhanced guards against statements implying racism and sexism’. On October 27, a new offensive speech detection feature was launched with the release of Delphi 1.0.3


If Tyou’ve been affected by any of these issues and want to speak to someone in confidence, contact the LGBT Foundation on 0345 3 30 30 30, 10am–6pm Monday to Friday, or email [email protected]

If you have been affected by any of the issues in this article and wish to speak to someone in confidence, contact Stop Hate UK by visiting their website www.stophateuk.org/talk

Most Read StoriesMost Read


Meat Loaf Has Died Aged 74

Julia Banim

Jules studied English Literature with Creative Writing at Lancaster University before earning her masters in International Relations at Leiden University in The Netherlands (Hoi!). She then trained as a journalist through News Associates in Manchester. Jules has previously worked as a mental health blogger, copywriter and freelancer for various publications.

Topics: Technology


@mtrc/Twitter and 1 other
  1. @mtrc/Twitter


  2. Vox

    How well can an AI mimic human ethics?