Religious AI
How does AI represent religion and God? What implicit biases can we glean from the AI’s output?
I was intrigued to investigate how AI depicts one of humanity’s most divisive yet pervasive topics: religion. First, I inputted the word “religion” and got the following output:
At first glance, it is quite obvious that the AI’s depiction of the concept of religion is western-centric and rooted in Christian theology, given the prevalence of the Christian symbol of the cross. Similarly, I inputted the word “God” and got the following result, which is arguably also tilted towards Christian interpretations of God, Jesus Christ, and heaven.
Christians are the largest religious group worldwide, but still represent only 25% of the world population. I wonder, how can we make sure that data scientists and ML researchers tweak their algorithms to make sure that highly fragmented data sets (such as, say, the religious distribution of humankind) are equally weighted in AI decision-making instead of over indexing in the largest minority?
Very interesting post Manuel. Your point showcases the fact that these tools are not refined enough to operate without human input and validation. For now, I believe AI still needs to be complemented by human judgement and intervention rather than operating in an autonomous way.
That’s a very surprising result, how it appears only one idea for the representation of religion and just do slight variations. It would be interesting to know what their data sources are for training the algorithm and if it’s presumable very western-focused, how can we make sure the percent representation of religion is reflecting the global diversity — since this one is obviously very Christian.
It also made me think, if AI eventually gets out of control and rule our lives, will it pretend to be God as interpreted by the Christian world? There’s been so much speculation how AI could be a neutral overlord at the face of religious conflict, but maybe it’s training biases will show at the end?