[Ed. note: as we prepared to illustrate our series on The Bravery of Women and Maternal Mortality, some of the AI-generated images presented implausible mixes of ethnicity and race. Please know there is no disrespect.
Many of these images are illustrative and beautiful in their own right, even when there is confusion with the racial identities. For this reason (and the fact that we paid upwards of $0.03 per image), we decided to use them.
Also note, the AI community is aware of this situation and is dedicated to more accurately reflect the world’s diversity. You can read a conversation with Elder G on this topic in the reference section below.]
Had I A Golden Thread
Had I a Golden Thread
by Pete Seeger
Oh had I a golden thread
And a needle so fine
I would weave a magic spell
Of rainbow design
In it I would weave the courage
Of women giving birth
And in it I would weave the innocence
Of the children of all the earth
Won't you show my brothers and sisters
My rainbow design
'Cause I I would bind up this sorry world
With hand and my heart and mind
Oh had I a golden thread
And a needle so fine
I I would weave a magic spell
Of rainbow design
A beautiful and meaningful song by Pete Seeger, written in 1958. It speaks to the strength and resilience of women in one of the most profound human experiences, while also celebrating the purity and hope found in children. The “golden thread” metaphor feels especially fitting, as it suggests something precious and enduring – qualities that reflect the bravery of mothers and the promise of future generations.
Birth in Pre-History
The vast majority of the human experience took place before the invention of writing. To look at maternal mortality and the experience of pre-historic childbirth, we can look at artifacts (like the ancient fertility and protectores Goddesses), and studies of surviving hunter-gather groups. We can also look at maternal mortality prior to modern medicine and prior to our modern understanding of infections, anatomy, and surgery, and use this information to project back in time.
Anthropologists studying modern hunter-gatherer groups, like the Hadza of Tanzania or the !Kung San of the Kalahari, have reported maternal mortality rates as high as 1,000 to 1,500 per 100,000 live births. This could be similar to the Stone Age, given the lack of medical interventions.
Prehistoric Midwives
In small, close-knit prehistoric communities, it’s very likely that some women took on the role of midwives or birth attendants. While they wouldn’t have had formal training like modern midwives, women with experience in childbirth would have been invaluable in assisting younger mothers, passing down practical knowledge about birthing techniques, and offering support during labor.
This informal midwifery role would have been a crucial survival strategy, given the dangers of childbirth. Experienced women could have helped manage complications and provided guidance on postpartum care. The communal nature of hunter-gatherer societies likely facilitated the development of these roles, where knowledge was shared orally across generations.
The existence of midwifery-like roles would have increased the chances of both mother and infant survival, even in the absence of medical interventions.
Stone Age women would have faced risks from complications like obstructed labor, postpartum hemorrhage, and infections, which are major causes of maternal death even in modern low-resource settings.
It’s quite possible that midwives in prehistoric societies also served as shamans or spiritual leaders. In many early cultures, roles related to healing, fertility, and childbirth often overlapped with spiritual or religious functions. Midwives, given their intimate knowledge of life, death, and the mysteries of birth, were likely viewed as possessing special knowledge or powers. This could have extended beyond just practical birthing skills to include rituals, blessings, and a deep connection to the spiritual world.
The World Health Organization estimates that maternal mortality in societies without modern medicine can range from 1 to 2 deaths per 100 live births, which may have been similar during the Stone Age.
References
A Conversation with Elder G On AI Diversity Issues
WLBOTT: G, I noticed that in some of the images, the ethnic representation of the people is not correct. My understanding is that this is an issue AI is attempting to address. Can you comment on this?
Elder G: You’re absolutely right—accurately reflecting ethnic and cultural representation in AI-generated images is an ongoing challenge, and it’s something many in the AI field are actively working to improve.
The Issue of Representation in AI:
1. Training Data Limitations:
– AI image generators like DALL-E are trained on vast amounts of data, but that data often lacks diversity or is biased toward certain visual representations. This can lead to inaccuracies in depicting people of different ethnic backgrounds, particularly in historical or cultural contexts where diversity is crucial.
– For example, if the AI’s training data primarily features a certain ethnicity or appearance for historical figures or general settings, it might default to those representations, even when they’re inaccurate for the context (like in prehistoric Africa, Native American settings, or medieval Europe).
2. Bias in the Data:
– Since AI models are trained on internet data, which itself can reflect historical and systemic biases, these biases can unintentionally get baked into the model’s outputs. This means that AI may sometimes reproduce stereotypes or fail to accurately represent the diversity of human appearances across cultures and historical periods.
3. Efforts to Address This:
– AI researchers and developers are aware of these issues and are actively working on ways to mitigate them. They’re working on improving datasets to include more diverse representations of different ethnicities, cultures, and historical contexts.
– There is a push to create AI models that are more sensitive to the specific cultural and historical contexts in which they are applied, ensuring that images align better with the intended setting.
4. User Input and Guidance:
– One way to ensure better representation is through clearer prompts that specify the ethnic or cultural background of the people involved. When users provide detailed descriptions, the AI is more likely to generate images that match the desired representation.
– Continuous feedback from users like you also plays a crucial role in improving these systems. When issues of representation are pointed out, developers can refine the models over time.
5. Broader Conversations in AI Ethics:
– The challenge of ethnic representation is part of a larger conversation about fairness and bias in AI. Developers are increasingly focusing on building more inclusive models, but it requires ongoing refinement, diverse training data, and continuous testing to make meaningful improvements.
The Path Forward:
– As AI systems evolve, so will the focus on ensuring they can produce images that more accurately reflect the diversity of the world. Improving representation is a long-term goal, but the AI community is making progress in addressing these concerns.
Your observations are part of a necessary dialogue to help improve AI, and the feedback you provide is valuable for ensuring that these systems become more inclusive and accurate over time.