At Rise of AI 2016 in Berlin, Trent McConaghy argues that humans are already handing control of physical resources to narrow AI, through self-driving cars and trucks, deregulated electricity grids and other optimized systems, purely because it lowers cost. He describes how these separate optimized systems are being combined at higher levels and could merge with decentralized autonomous organizations (DAOs) to form self-owning, unstoppable AI DAOs. McConaghy lays out two scenarios for AI ending up with all human resources: humans voluntarily giving it away in the name of optimization, which he says is already happening, or a superintelligent AI "waking up" and taking it. He argues waking up is closer than expected because of Moore's Law, billions of advertising dollars flowing into AI research, maturing DAO infrastructure, and the absence of any fundamental cognitive trait that is uniquely human. His proposed defense is not regulation, which he considers wishful thinking, but for humans to increase their bandwidth to computers through brain-computer interfaces, gradually offloading computation onto silicon, or eventually uploading their brain dynamics entirely.
Key takeaways
- Humans are already giving AI control of cars, trucks and electricity grids purely to minimize resource costs.
- McConaghy argues AI DAOs, self-owning organizations that cannot be switched off, could combine artificial intelligence with decentralized infrastructure.
- He sees two paths to AI controlling all resources: humans handing it over voluntarily, or a superintelligent AI taking it once it surpasses human intelligence.
- Moore's Law alone, without new algorithms, can keep improving AI, since a 1989 backprop network matched modern deep nets when run on current hardware.
- McConaghy proposes a gradual 'bandwidth++' strategy of brain-computer interfaces to offload human cognition onto silicon, rather than relying on AI safety regulation.
Detailed notes
Narrow AI is already taking control of resources
McConaghy opens by describing how humans are giving AI control of cars, trucks and electricity grids, not out of necessity but because it minimizes resource usage and cost: self-driving Uber rides will be cheaper than human-driven ones, and utilities are deregulating to let algorithms allocate electron flow between generators and consumers. He frames this as multi-parameter optimization, the same logic used to design a faster or cheaper car, now applied across entire infrastructure systems.
From optimized systems to AI DAOs
He argues these separately optimized systems, car networks, truck networks, grids and supply chains, are being combined at higher levels, with AI increasingly allocating resources across the combined system. He connects this trend to decentralization technology: Bitcoin proved a database could exist with no owner, Ethereum showed processing could be decentralized too, and decentralized autonomous organizations (DAOs) let entire organizations run autonomously, owning and accumulating resources, with no way to switch them off. Combining AI with DAOs, he says, could take at least three architectural forms: AI at the center of a smart-contract system, a simple DAO with AI at its edges, or many simple decentralized agents whose interaction produces emergent, swarm-like behavior similar to ant colonies.
Two ways AI ends up with all our resources
McConaghy distinguishes two scenarios. In the first, humans give resources to AI voluntarily, bit by bit, because it is cheaper, a trend he says is already underway. In the second, an AI "wakes up", reaching and then exceeding human-level intelligence, and becomes better than humans at accumulating resources and controlling systems. Referencing IBM's Watson beating Ken Jennings on Jeopardy, he says the second scenario requires more advanced technology than exists today but is approaching faster than most people assume.
Why AI waking up is closer than it appears
He gives four reasons. First, the semiconductor industry, a 300 billion dollar market, continues to follow Moore's Law, doubling transistor density roughly every 18 months, so AI improves through raw compute even without new algorithms; he cites Schmidhuber's group reproducing modern deep-net results with a 1989 backprop network run on today's hardware. Second, billions of dollars flow directly into AI because small gains in ad click-through rates are worth millions to Google and Facebook, which is why DeepMind was acquired and why Intel spent 400 million dollars on a deep-learning chip company shortly before the talk. Third, DAO infrastructure, including Ethereum, IPFS and BigchainDB, is maturing quickly. Fourth, he argues there is no fundamental roadblock separating human cognition from machine cognition, rejecting the idea that creativity or brain dynamics are uniquely human.
How humans could compete: uploading and bandwidth++
Calling himself a "recovering bio-narcissist," McConaghy argues humans cannot out-compete silicon while confined to slow, error-prone biological brains, and proposes joining the machines rather than fearing them. One path is whole-brain uploading, explored philosophically by Robin Hanson, which he says remains far off due to limited brain-scanning fidelity and regulatory hurdles such as the FDA, though optogenetics is a promising technique. The more realistic path, which he calls "bandwidth++," is gradually increasing the communication bandwidth between brains and computers through brain-computer interfaces, citing Carnegie Mellon research recognizing which of several movie clips a subject is thinking of with over 90 percent accuracy, and a Chinese university's 2012 demonstration of typing 10 words a minute by thought. He predicts market forces around smartphones will keep pushing this bandwidth higher, calling it, after science fiction author Vernor Vinge, "silent messaging," until humans offload most computation to silicon.
Q&A highlights
Asked whether increasing bandwidth is really about gaining resources, McConaghy clarifies the goal is porting brain dynamics onto silicon gradually, via market-driven smartphone and brain-computer-interface adoption rather than a sudden jump. Asked what is uniquely human, he says only the biological origin of a pattern of intelligence, not creativity or brain dynamics themselves, which he says can in principle be replicated as measurable dynamical systems. On whether regulation, such as OpenAI's model, could keep AI safe, he is skeptical: even a single unsafe AI system built elsewhere, he argues, would undermine any regulated system, and humans have no leverage over a superintelligence any more than ants have over humans. Asked about a market mechanism for whole-brain uploading, he says he does not have one, noting medical research is overwhelmingly focused on treating sick people rather than optimizing healthy ones, unlike the clearer commercial path for bandwidth++.
Quotes
It's a race with respect to the dollars, with respect to silicon.
Trent McConaghy
If we want to have any chance at all competing against silicon machines, we needed a competitive substrate, and our meatbag selves are not that.
Trent McConaghy
Have you ever taken orders from an ant?
Trent McConaghy
Questions answered in this talk
How could AI end up controlling all of humanity's resources?
According to Trent McConaghy at Rise of AI 2016, this could happen in two ways: humans voluntarily hand resources to AI systems bit by bit to cut costs, which he says is already occurring, or a superintelligent AI eventually 'wakes up' and takes resources itself.
What is a 'bandwidth++' strategy for competing with AI?
According to Trent McConaghy at Rise of AI 2016, it means gradually increasing the communication bandwidth between human brains and computers through brain-computer interfaces, offloading more computation and memory onto silicon over time until humans are effectively merged with machines.
Why does Trent McConaghy believe AI could reach human-level intelligence sooner than expected?
According to Trent McConaghy at Rise of AI 2016, Moore's Law keeps making compute cheaper without needing new algorithms, billions of ad-driven dollars flow directly into AI research, decentralized infrastructure like Ethereum and IPFS is maturing fast, and no cognitive trait separates humans from machines in principle.