Researchers at Apple published a paper about LRM’s (large reasoning models) and further validated a hunch I had during my thesis when I was writing about generative AI. At its core, there are many indicators that heavily imply that AI, in its current form, has no intentionality. It’s fundamentally heuristic, admittedly mind-boggling, but to say it is actually thinking like we do is not only a discussion for neuroscientists and philosophers, but very likely also premature.

This gamescom I often said things like “Yeah sure, that sounds good!” and “Of course, I’m thinking about it.”. This was merely an illusion. I tricked the clerks at the venues into giving me free swag and goodies by pretending to think about what they were saying. However, I was actually pondering if the Danish folks in the business area will offer free beer before the Belgians like last year (They did). Similarly, even if we stumble upon behaviour that mimics thinking in AI (Such as Chain of Thought (CoT) in the current DeepSeek iteration https://arxiv.org/abs/2203.11171.) we also fumble with the question of whether it’s thinking or acting as if it’s thinking. This is just another addition to the pile of problems such as the Chinese room, an AI becoming truly sentient and intelligent enough to still fail the Turing Test on purpose to avoid detection, or (See Searle) understanding that you can’t just transfer systems from one medium into another (Can a brain be transferred from flesh to circuits? If yes, a motor made out of steel surely works constructed from wood only too.).

While evaluating the question how much wine and beer can fit into a single person on the business mixer event next to the gamescom, I started wondering, if a human without eyes, ears, really any sensation, could develop intelligence or any form of consciousness. While also hard, if not impossible, to prove, it raised the question for me, if multimodal AI is the first step into a direction that could scratch the surface of consciousness or true intelligence, given that it has access to a stream of data it can interpret (interpret being one of these words that is equally hard to define as intelligence), comparable to senses in a human. Or, circling back to the Chinese room, this threshold has already been surpassed if we consider our interaction and its learning based on this data a rudimentary form of sense or even communication.

Aside from the hangover, gamescom was pretty sweet. It is always a joy to say hello to old friends and make some new. Path of Exile 2 had a cool booth that turned a selfie of you into a corrupted orb, and I think the way they solved it was through an AI they trained, as simple manipulation and projection through an engine or 3D wouldn’t cut it. Besides that, the winner, as every year, were the indies. Unless you’re into standing in line for hours. Then I recommend the AAA section. Still no Astolfo figure though. (They had one on sale, but it was too expensive)