How machines (might) take over the world
The state of AI in 2025
TL;DR
This newsletter is about AGI and the current trends in AI development. It looks at why we might not get AGI, what we might get instead and the current barriers to wider AI adoption.
Here are the key takeaways:
Some remain very confident about achieving artificial general intelligence (AGI). As soon as 2027 we could apparently have machines exhibiting a level intelligence akin to that of humans.
However, it is most likely the case that we will not get to AGI. Simply put, the kind of intelligence required for which it is impossible to build mathematical models for that can be executed by algorithms.
As developers continue to try to get to AGI nevertheless, the following trends will continue:
Developers continuing to be data-hungry
Developers continuing to spend on infrastructure
Developers continuing to release new models with (allegedly) better capabilities
The cumulative result of these trends may be AI diffusion, in particular:
Enterprise adoption will see more companies trying to build more AI-powered applications, deployed for both internal and external use cases.
Consumer adoption will see multi-modal foundation models underpin a new means for navigating the digital world, challenging search in particular.
But the main barrier to AI diffusion remains alignment, in both a local and global sense:
Locally, can model developers and companies building on top of models successfully control their AI systems to behave in the desired fashion?
Globally, can developers convince the wider populace to adopt these AI systems?
The hype is still very much alive around AI, and in particular around artificial general intelligence (AGI). The exact definition of this still somewhat shifts over time, but generally AGI refers to an artificial entity capable of a level intelligence exhibited by humans.
Recently, Daniel Kokotajlo (formerly of OpenAI), Scott Alexander (writer of ACX on Substack), Thomas Larsen (AI Futures Project), Eli Lifland (AI Futures Project) and Romeo Dean (formerly of the Institute for AI Policy and Strategy) published AI 2027. This elaborate blog post sets out predictions for how AI will develop between 2025 and 2027. The authors contend that:
...the impact of superhuman AI over the next decade will be enormous, exceeding that of the Industrial Revolution.
Further on they state:
Over the course of 2027, the AIs improve from being able to mostly do the job of an OpenBrain [placeholder AI company used for illustrative purposes] research engineer to eclipsing all humans at all tasks.
And so there it is, AGI in 2027.
The main propeller of the predictions here is that advanced AI models can be used to develop even more advanced models. This is the positive feedback loop that can be observed in technological development whereby good technology leads to better technology (or at least makes it easier to create better technology). Accordingly, "this process is constantly accelerating: the returns of each generation of technology layer on top of the previous generation's, and even feed into one another."1
For AGI 2027, as Nathan Lambert of Interconnects explains, the authors rest "on a few key assumptions of AI research progress accelerating due to improvements in extremely strong coding agents that mature into research agents with better experimental understanding." In particular, stronger models "enable AI progress to change from 2x speed all the way up to 100x speed in the next few years."
The hype remains around AI because of a firm belief in this. In fact, some proponents of AGI are even of the opinion that naysayers are in the minority who bear the burden of showing why AGI will not happen:
If you want to argue that human-level AI is extremely unlikely in the next 20 years, you certainly can, but you should treat that as a minority position where the burden of proof is on you.
Keep reading with a 7-day free trial
Subscribe to The Cyber Solicitor to keep reading this post and get 7 days of free access to the full post archives.