Identity talk lends itself to essences - is there a 'real' me that remains the same from when I was born to when I die (or if you believe in reincarnation, in the next life as well)? You've likely posed the question to yourself: “who am I?” Perhaps it was during a late-night discussion with friends in college, or within a religious context that encourages such self-inquiry. In today's digital age, there’s an app (or many) to help explore that query, with even Google offering a program named ‘search inside yourself.’
To be frank, 'who am I' is mostly a pseudo-question. It's rarely posed with a genuine desire to uncover the truth. And sincerity is no protection, for serious pursuit of the answer reveals that asking this question skews you towards a specific type of answer, suggesting a true self lying in wait to be discovered. Alternatively, one might gravitate towards the Buddhist perspective, embracing the notion of self as no one. I am increasingly inclined to dismiss the question altogether.
Unfortunately, who-am-I-tis is so robust that even when we talk about technological change, we remain fixated on a human essence as if the human whose intellect is being augmented by print media is the same as the human who is being augmented by AI. Of course, there's continuity, but the human is as much the product of the times as our gizmos. There's a techy angle to all these questions of continuity and difference. In the early days of the internet, it was more about making sure devices could talk to each other than people. Imagine you and I both had email IDs. How do we ensure that an email sent from my ID to yours gets to you? There needs to be a way to send that bunch of information from my computer to yours, right? And for that to happen, both our computers must have their own identities.
Now, think about today. The internet is way bigger, and we all have loads of devices doing all sorts of things for us – shopping, dating, moving, and more. It’s like, is the same person doing all these things? And how are all these activities linked together? Like, how can a website know to suggest you buy flowers before a big date? There’s a big deal in knowing if two users are the same person or if they’re friends or related. It’s all about connecting the dots between people and their actions online.
Very interesting, but not what we are going to talk about in today's episode about identity
What we do talk about are:
1. We start with the problems of automated face recognition systems. They're struggling to recognize people with darker skin tones. It's mainly because the training sets used are biased towards light-skinned people, causing both false positives and negatives. The chat also touches on biases in image generation and selection, where the default images are usually young, attractive, and often have Western features. This bias isn’t just in facial recognition systems but is everywhere - in media, scientific literature, film, TV, and ads, where attractive people are overrepresented, just making these biases even worse.
2. We then address concerns about biases in large language models because certain texts are overrepresented in their training data. This overrepresentation just keeps on reflecting and perpetuating social, class, and geographical biases. These models usually default to an upper-middle-class, college-educated persona in their responses. Public voices, and so text producers, are generally from middle or upper-class backgrounds, making the representation in language models even more skewed. Exceptions like voices from trade unions are rare. The AI, trained on such biased datasets, doesn’t mirror all our identities but rather reflects the identities of the more privileged and text-producing parts of society.
3. We then talk about how AI might let people explore and experiment with their identities. Starting with the early internet era, people could present themselves differently in text-based forums, highlighting the disembodied nature of digital media that lets people adopt various identities in different contexts. AI, we argue, just boosts this capability. AI can be used to simulate conversations with different versions of oneself, like a teenage self, and the potential for people to explore various identities systematically. This exploration is compared to the developmental stage in youth, where people try on different identities, just like trying on clothes.
4. The talk goes deeper into the impact of AI on personal identity, with Rajesh sharing a personal story about buying sports coats to fit into an academic role, symbolizing a change in identity. AI could provide even more opportunities for people to adopt different personas. by merging with the metaverse, where people can adopt various avatars, playing with aspects like gender, age, and even species, offering a more visual and embodied experience compared to text-based chat rooms. That portion of the conversation concludes with a discussion about the potential of AI to let people experience the sensory inputs of other species, fundamentally changing their sense of bodily identity, and the ongoing efforts to use machine learning to communicate with other species, like whales.
5. We then move to a discussion of social class and the evolving definition of maturity. There's an ongoing shift in class markers, such as the unremarkable clothing of tech billionaires as a reflection of a new dominant social class characterized by cultural fluidity and the adoption of new technologies. This shift prompts a discussion about the perception of maturity in the age of AI. Traditionally, maturity has been associated with age, dependability, stability, and knowledge. We suggest a redefinition of maturity to include adaptability, the acknowledgment of one's lack of knowledge, and the ability to learn from failure, especially in a rapidly changing technological landscape.
6. The conversation also touches on the fears and challenges brought about by AI, including job loss and the risk of societal irrelevance. We emphasize the importance of a social safety net to allow individuals to adapt to these changes without devastating consequences. The discussion concludes on a hopeful note, citing a recent agreement between the Writers Guild of America and film companies as a positive step towards defining workers' rights in the age of AI, reflecting an optimism for the future of work and society in the AI era.
Humainity Episode 5: AI and Identity