As Europe burned from extreme weather, yet hundreds of new energy-hungry data centres popped up around the world; as AI threatened to destroy the job market in which I was exploring my options; as Open AI, Anthropic and Meta agents committed cyber-crime; as hyperscalers created a trillion dollar financing bubble, and even as human existence in the AI era started to be questioned, I was busy being harshly critical of a small 10 person team building an AI tool which helps deliberative democracy.

Working at the Humboldt institute for Society and Internet me and fellow scholar Ben Berman supported the ‘Impact AI’ research project. Their aim is to create an impact assessment framework that can be used to guide ‘AI for good’ projects. This is in order to ensure that they remain sustainable and in the public interest, with the ultimate wish being for the framework to be adopted by more powerful regulatory or governance bodies. Still in an early stage, me and ben helped produce some of the first impact reports on the projects the team has been assessing, namely, Fullfact AI (an AI for assisting fact-checkers), Talk to the City (TttC - the aforementioned LLM powered public consultation tool) and Burokrat (the Estonian governments chatbot bureaucrat). These projects were all noble in their intentions, sometimes impressive in their achievements and the worst they could be accused of would have been being slightly ignorant or ineffectual. A far cry from the enormous ethical, leadership, sustainability and public interest questions facing frontier labs, governments or big tech. Thus, being so hard-nosed about the fact that Talk to the City didn’t answer our interview questions properly felt a tad unfair (considering for example that Anthropic won’t even let the UK’s AISI look at its Mythos 5.1).

Despite this, we made good progress on our tasks. We grammar checked two papers that were due to be published, we qualitatively coded interviews for TttC and Burokrat, we pointed out the need for, and started work on, a full guidebook to their methodology, we started writing the reports which would eventually be published and I conducted some literature review on factchecking for another report. The feedback from HIIG was that it was useful for them to get fresh eyes on their newly developed and deployed method, and of course for us to grind away at the grunt work. The work did indeed feel valuable; it was shocking how little some of the initiatives under review had thought about what seemed like obvious questions. There was, in my opinion, often little thought payed to the sustainability of the tools they were producing, to the harms their tools could exacerbate and whether the tools achievements were really having a meaningful effect. Importantly, the feedback from the initiatives being assessed was positive, they reported that the interviews and reports had opened their eyes to some weaknesses and could be used to improve in future. In this way, it did seem like Impact AI had a real sense of direction, and that it was having, itself, a promising impact, one which should only scale up by the end of its expected runtime in 2029.

Inspired and in reaction to our work, me and Ben became increasingly concerned with AI governance, as well as with what our futures would hold. I spent too much time reading FT pieces about the complicated financing and economics of the AI build out, worrying about how greed may very well lead to enormous economic turmoil. Ben became worryingly interested in Alex Karp.  We were disheartened at how the projects we were assessing, full of good intentions, were having such meagre impacts, and by worried how much impact one can have in think-tank or policy roles. We lamented how many of our peers want to go into banking, consulting and big law, with little regard for ethics. Karp exemplified in many ways what we were thinking about. An academically minded student who gained a PhD in Frankfurt school critical theory, but who has ended up as the CEO of Palantir, a company which will if “necessary, scare enemies” of the institutions with which it works “and on occasion kill them”.  The job market in the state that it is, and AI threatening to make it worse, made me feel quite lost at what I ought to be doing. We looked around, got quite into LinkedIn, and even consulted Effective Altruism initiative 80,000 hours’ AI career advisor. This didn’t help much. Besides, conversations with project lead Dr. Züger about her dislike of EA, put me off some of the chatbot’s suggestions. I wish I could say there was a remarkable wave of realisation which stuck me at some point, where it was revealed what I needed to do, but alas. No, the LIA certainly reaffirmed by desire to work in the pursuit of social impact and re-opened my eyes to the horrors that may proceed from the tremendous lack of leadership we are witnessing globally, but personally, I remain at a loss on what to do about it.

Amongst the doomerism, there were some smaller, concrete lessons learnt. I enumerate a handful below.

 On leadership – The change in pace, energy and direction when Dr. Züger was present versus when she was on holiday was palpable. To feel such an obvious change in the presence of a senior was one of the first times I have really felt the effect of leadership that the Laidlaw programme has talked about. In a discussion with Dr. Züger she shared her own perspective on a good test for if one wants to be a leader. In her view, one wants to be a leader not if one is happy to be in charge while all goes well but if one is ready and willing to deal with the pain, embarrassment and responsibility when things go wrong.

On creativity - Ben and I, spurred on by one another, took our weekly blogs in unorthodox directions. What started as an attempt to write amusingly spiralled into a Bill and Ted spoof script, a cowboy song, an archive footage collage and comedy spoken word. This was not just an attempt to avoid writing a serious blog, or a diversion, but became central to our week. We realised how crucial creating art (if you’ll allow me to call our blogs that) is for coping, digesting, feeling and enjoying what goes on around you. The Oxford workload has not encouraged me to take time to create inane songs and creative experiments, so I have relished the chance to rediscover these little, essential joys again.

On monotony – Gosh working in the same place every day is boring. As a first foray into the workplace, we benefitted from the novelty, but looking back, I can hardly separate week 2 from week 6. The office lifestyle, the monotony of work and the fact that in a real job you have to do that for a lot longer than six weeks hit quite hard. At least there was a coffee machine.

On Berlin – What a city. Human, cheap, easy-going. Excellent public transport (with no ticket gates?!) more spacious and less stressful than London. Bars that exuded coolness, whilst remaining cheap and relatively welcoming and staying open late. Beautiful public spaces and museums tying the city vibe together. I feel it is the city one might design if given control over a very realistic city-builder simulation (complete with buildings that feel like they come out of a badly rendered video game).

In reflection, I have thought a lot about responsibility (both ethical and personal), work (at the career level and at the quotidien). My experience at HIIG was extremely valuable forcing me to open my eyes to the reality of the impact of both the ‘AI for good’ and the ‘AI for bad’. I hope to have contributed fruitfully to their work, and I believe, unless the letter of recommendation from them is very exaggerated, I did. Beyond HIIG, living with Ben and marauding around Berlin with him and Oscar (a non-Laidlaw but fellow Oxford student and HIIG intern) was a delight for which I owe the Laidlaw programme much credit for facilitating. We were together inspired to think about the impact we will have with our futures, excited by what lies ahead, and grateful for what we are able to enjoy in the present. Lessons on leadership, creativity, appreciation popped in many a surprising place and I am glad to have come away from my LIA feeling however so slightly more complete as a person than when I set out. Frankly my resolution in the face of a 19-hour return train journey ought to speak for a newfound maturity. Of course, I remain, as always, intentional collaborative and transcendent as I finish my second Laidlaw summer and move on to my future.