Masoumeh Iran Mansouri

I love going to conferences, seeing old friends and colleagues, having nice local foods or late drinking in local bars. This time was different, however. We were one year into a live genocide, and as much as life goes on as normal here in the West, I struggle to clear my mind of the last picture of horror on the TV or social media.

It was pouring rain on the first day in Santiago de compostela so I took a taxi, arriving at the conference centre right before the first keynote talk. It is the European Conference in Artificial Intelligence, relatively high-profile thus not so easy to get a paper accepted here. I had so many expectations, AI cannot be more in the news than now. It is only two weeks since Geoffrey Hinton, the so-called father of AI, received a Nobel prize in physics, and in his acceptance interview warned, concerningly, that AI is “getting out of control”. All these have been making headlines: EU regulations around AI; the new UK government’s wooing of big tech, with consequent excitement and anxiety around building new data centres; using AI to help solve the epidemic of mental health problems in the UK and cut costs in an ever more expensive NHS. And last but not least, there has been a flurry of articles about the AI system called ‘Lavender’, used by the Israeli Defence Force to identify the location of what they consider as Hamas supporters.

I was thinking surely the conference will be full of panel discussions and topical debates around AI and society. Excited to know the opinions of the AI experts without the pressure of usual journalism click-bait styled interviews. Yet, to my great disappointment, it became apparent to me that the conference exists in a parallel world.

The first keynote was on misinformation. The speaker started with several examples including false news about the vaccine and how in the old days of AI, researchers would automatically fact-check a text. Then, she moves to the challenges Large Language Models pose, and how it is not any more straightforward to automatically fact-check both pictures and texts. I was following her talk carefully and enjoyed seeing that an attempt was being made to address one of the greatest threats AI poses to the world. I was drawn to the technicality of the algorithms she was describing until she hit my head. Among the examples of faked images, she mentioned an image of a dead child, ostensibly killed during the current war on Gaza which turned out to have been taken in Afghanistan. I could not believe what I just heard. Surely there is misinformation on this war, so as others, but why does she choose this example? She could bring any number of examples which are more egregious, i.e. sought to portray an alternative reality. But this is not an example of a malign actor trying to tar the innocent. Sixteen thousand children have been killed in Gaza, at the last estimate. Is a single proven case of misattribution really going to tip the scales of public opinion, or negate the fact that many real Palestinian children are being slaughtered, largely by high-yield bombs dropped from thousands of feet above them??

I was stunned and enraged by the sound of clapping in that big auditorium. Frantically asking people around me, did you see what happened? Some hadn’t noticed, as she did not show the picture and gave the example as a passing remark. Some pretended not to hear me and walked away, and those trapped by my animated outrage showed sympathy but were made uncomfortable by the discussion and a bit confused as to why the example matters this much. That was the moment I felt I was in a parallel world. It was not parallel because it was governed by ignorance, these are smart people after all. The disconnect grew from the fact that the AI people, those creating the algorithms that will supposedly change the world, seemed entirely uninterested in any question beyond the technical details. Surely not all the attendees could be so apathetic? I determined to find out, and began canvassing opinion about the keynote talk and more importantly, the Lavender system itself. All I had seen so far was silence and discomfort.

On the second day, there was a panel of experts on the future of AI. I missed the history session, which apparently was all personal anecdote, not the social history from which we could potentially learn lessons. But I was in the ‘future’ session. The panellists discussed the challenges AI imposes on us today and laid out some research directions, none of which was that interesting. From what I gathered lack of understandability, explainability, black boxes, and safe guarding were the themes. There was some disagreement here and there, but nothing that 5 minutes’ googling on this topic cannot tell you about. When it came to the audience questions, the first was on climate change. One panellist refused to engage with the question as he considered it a problem created by capitalism, thus a social, not a technical question; more efficient data processing would not solve anything. There were no questions on the usage of AI in war or killer robots, nor did the panel mention the topic, other than a vague mention of the need for regulations in the future!

Back in my hotel room  and checking emails, I saw a message from a former AI student, distressed about what is happening in Palestine, who asked whether we as AI researchers have reflected on what we can do in this situation. I went to bed with hope that night.

After two underwhelming days, I attended the session on AI ethics and fairness which featured some genuinely interdisciplinary research studies with insightful results. One paper studied the impact of humans ‘in the loop’ of AI assisted decision-making systems, and pointed to the fact that this does not automatically make a decision fair, and in some domains, makes it worse. Of course, I immediately thought of Lavender, where there is a human in the loop, approving new targets produced by the machine every couple of seconds without further analysis or checks. Effectively just giving a rubber stamp, or human stamp, to each ‘targeted’ operation.

The unintended highlight of the fairness session, for me at least, was the presentation of two papers from researchers affiliated with investment bank JP Morgan, one about taking into account the fairness of a history of past decisions. One can hardly imagine a better example of ‘corporate responsibility’ ethics-washing than JP Morgan discovering a passion about fairness in society!

Then there was a banquet, full of delicious Galician seafood, drink and music. We laughed and got drunk, checking in at local bars till very late. There was no mention of Lavender, genocide, AI replacing jobs or serious discussions on climate change. I am now on the train back, writing this while reflecting on the conference, and mulling the question my student asked: what should we do as AI researchers?