Are AI deepfakes a distorted reflection of social bias?

As we near International Women's Day, I reflect on the dichotomy of progress and peril in our digital world through the prism of non-consensual deepfakes. These AI-generated synthetic media can replace a person's image or voice, and while they demonstrate AI's leap forward, they also reflect and intensify social biases.

In 2019, I initiated a consumer survey on deepfakes to gauge their reach and social impact. At that time, awareness was limited to celebrity impersonations and harmless pranks. Fast forward five years, and we're seeing a spike in malicious uses, with deepfakes becoming alarmingly sophisticated.

See my comments on this BBC article - "Deepfakes take this [women's worth equated to beauty standards and female bodies being objectified] further. The non-consensual nature of deepfakes denies women the dignity and autonomy over the depiction of their bodies. It takes away the agency, and puts power in the hands of the perpetrators."

Non-consensual deepfakes are a twofold problem:

1. They violate privacy and consent, disproportionately targeting women. With the power to craft false narratives, they erode trust in media, deepen the digital gender gap, and challenge online and offline safety for women.
2. Deepfake technology magnifies existing AI system biases. From facial recognition that fails to identify women and people of colour to natural language processing models that encode stereotypes into their word associations, biased training data breeds biased AI, exacerbating gender discrimination in the virtual world.

Addressing this calls for a multi-faceted approach:

1. Elevate AI and Digital Literacy: Public education on deepfakes is critical. Awareness campaigns can help individuals critically evaluate online content.
2. Implement Robust Legal Frameworks: We need laws that address deepfakes, with penalties for misuse. Protection of rights and dignity, especially for women and minorities, must be a priority.
3. Technical Safeguards: Advancements in detection, digital watermarking, and information source verification are key to combating deepfakes.
4. Promote Diversity in AI Development: Inclusive teams in AI can help counteract built-in biases, ensuring a variety of perspectives shape technology.
5. Be Critical Consumers: Scrutinising media authenticity, particularly if the content seems controversial, will help curb the spread of synthetic content.
6. Advocate for Change: Support movements and policies that combat digital violence and advance gender equity in tech.
7. Educate and Empower: Share knowledge about the effects of deepfakes and AI biases, fostering community action towards a fair digital environment.

An informed dialogue and proactive engagement are vital as we explore AI's vast potential while upholding best practices, gender equity, and digital safety.

For a deeper dive into this pressing issue, read the full article on my website: Gender Equity in AI: Are Deepfakes A Distorted Reflection of Society's Bias?