I was invited by the legendary Shahidul Alam – one of the most influential photographers alive – to give two lectures in Dhaka, including to all the students of Pathshala Institute, one of South Asia’s pre-eminent photography, film and media schools.
I had a wonderful time engaging with the students, and what was billed as an hour long lecture, went on for around twice that long because of the questions I got, and the conversation that ensued.
I started the lecture by flagging examples of generative AI use in Bangladesh, including in the recent election campaigns. Coincidentally, I spoke a day after Ukraine introduced its AI generated foreign ministry spokesperson.
I flagged how generative AI is becoming increasingly democratised, with industry-scale production (the vast reach established on social media, now combined with unlimited production) and a lack of accepted guardrails leading to a potential inflection point in truth decay.
Flagged a couple of key issues arising from the inability to establish provenance or determine veracity of AI-generated images, which can be indistinguishable from content with epistemic authority. For photography (and photo journalism), I said this challenges the reliability of captures, and undermines witnessing through acts of journalism (i.e., by citizen journalists), as well as professional photo journalism. I said these are issues that are already challenging institutions, society, shared realities, and democracy.
I then talked about four foundational concepts around the presentation(s) of self online, splintered realities, making false connections, and how populists get away with lies. I used a story by Hans Christian Andersen to illustrate how vastly different our online representations are from our real selves, and identities (and how generative AI can significantly mess around with how we are perceived by the state, security agencies, at borders, or around college admissions, and job interviews etc). I talked about irrealism, and what Nico Frijda calls the law of apparent reality, to explain how a subscription to shared facts, realities, and values are being eroded at an increasing pace through divisive generative AI content that adds to, and entrenches splintered realities. Connected to this, I talked about apophenia, and the erroneous perception of patterns in totally unconnected phenomena.
Finally, and in relation to the kind of government Bangladesh enjoys, along with many other states in South Asia, I spoke about the “Liar’s dividend“, and how using deepfakes can erode trust and cast doubt on factual accounts.
I also noted how generative AI presentations can widen and deepen the so-called “Devil Shift“, where actors perceive their opponents as more powerful and malevolent than they truly are. This, I said, was especially a risk in the (ab)use of generative AI by political campaigns in deeply divided societies, and contexts marked by a significant democratic deficit.
I went on to talk about how the democratisation of content production allows anyone to generate content using free or cheap access to generative AI tools, either cloud-based (with some guardrails) or locally (with zero restrictions or guardrails). I provided examples using Stable Diffusion, highlighting the potential for creating realistic portraits with specific prompts in real time.
Noting the line between image enhancement and AI manipulation is blurring, with on-device, cloud-based, core OS, and app-based tools already the norm, I said generative AI brings both new creative potential and harmful uses, redefining essential photography skills, and pedagogy. I stress Pathshala needed to reflect on how quickly the landscape was changing.
Speaking to the importance of media literacy in general, and visual media literacy in particular (given the nature of the class I was speaking to), I flagged various measures to identify synthetic media such as hashing, watermarking, cryptographically-signed provenance information, and AI detectors. I also spoke in some detail around how none of them really address what’s a growing issue.
I ended the lecture by holding up a 100 taka currency note, and asking the students to reflect on a world where you couldn’t tell if it was a fake or not. I connected this back to a world overrun by generative AI content, and what a seminal essay on generative AI published in the New York Magazine noted, featuring Emily Bender, and written by Elizabeth Weil.
All the lecture slides can be downloaded as a PDF here.
Sanjana Hattotuwa 