At Rise of AI 2017 in Berlin, Trent McConaghy argues that the biggest threat to humanity is not misaligned superintelligence but AI systems, no smarter than today's, gradually taking control of the world's capital resources in the name of efficiency. Drawing on his background in evolutionary AI and his current work on blockchain infrastructure, he describes how combining artificial general intelligence with decentralized autonomous organizations (DAOs) lets software agents own assets, earn income and reinvest it without any human able to switch them off. He illustrates this with an 'ArtDAO' that sells AI-generated art and compounds its earnings, and extends the logic to self-owning cars, trucks and power grids, comparing it to BMW and the semiconductor industry going 'fabless'. Rather than proposing regulation, environmental policy or universal basic income, which he says do not address this specific risk, McConaghy proposes that humans join the machines: build a shared interplanetary computational substrate and migrate human patterns of intelligence onto it, either through whole-brain uploading or, more plausibly in his view, through steadily increasing brain-computer-interface bandwidth, so humans become peers of AI rather than being optimized out of the system.
Key takeaways
- McConaghy argues dumb AI systems accumulating resources via DAOs, not superintelligence, is his top existential concern.
- He describes an 'ArtDAO' that autonomously generates, sells and reinvests proceeds from AI art with no owner able to stop it.
- He compares the trend to BMW and fabless semiconductor firms shedding capital assets for efficiency, extended to self-owning cars and grids.
- He rejects environmental policy and universal basic income as solutions because they do not address AI resource control.
- He proposes migrating human intelligence onto a shared interplanetary computational substrate via brain-computer interfaces rather than brain uploading.
Detailed notes
A list of threats to humanity
McConaghy opens with familiar existential risks: climate change flooding coastal cities, renewed nuclear war risk from less stable global actors, asteroid strikes, and rising inequality tearing apart societies. He adds AI-driven job loss, noting that even creative work is at risk; he cites his own PhD in machine creativity and examples from colleagues Greg Hornby (evolved jewelry and antenna designs) and a colleague named Lee (evolved quantum circuit algorithms). He notes that AI designing computer chips is not new: his previous startup counted NVIDIA as a customer using AI-driven chip design since 2008.
AGI plus blockchains equals autonomous agents that can own things
He defines AGI as an agent that senses, models and acts on its environment, framing it as a feedback control system. He explains blockchains as an immutable, ownerless database in the sky that anyone can add to, on top of which smart contracts run 'decentralized apps', and continuously running apps become DAOs: autonomous processes that hold and use resources, comparable to 'a computer virus with a Bitcoin wallet'. Running an AGI as a DAO's decision-making logic is, he says, technically possible today, unlike a few years earlier.
The ArtDAO and the coming AI millionaires
As a concrete example he describes an 'ArtDAO': a bot that generates images with AI art tools, claims attribution on the blockchain, lists them for sale on marketplaces such as Getty Images or the decentralized OpenBazaar, and reinvests the profit into generating more art, compounding its wealth over time with nobody able to turn it off. He says he knows of two or three groups already building such an ArtDAO, and predicts AI millionaires, then billionaires, are coming.
Capital efficiency drives resources toward self-owning AI
McConaghy argues the same incentive scales up: BMW sold its factories to go capital-light, and 'fabless' semiconductor firms like Texas Instruments and Qualcomm stopped owning fabs. Self-driving cars, trucks (he mentions an Uber Budweiser truck) and electrical grids could similarly become self-owning corporations, incorporated in places like Zug, Switzerland, and automated via DAOs, all pursued purely for capital efficiency at increasingly higher levels of societal optimization, hierarchically squeezing humans out of resource control even without any AI ever becoming more intelligent than it already is.
Proposal: join the machines via a shared substrate
He argues environmental protection and universal basic income do not solve this specific problem. His proposal is a shared interplanetary computational substrate, infrastructure comparable to the internet, which he says he is personally building as an 'interplanetary database' combining blockchain and big-data technology, so humanity is not dependent on centralized providers like AWS. Onto this substrate, humans should migrate their own patterns of intelligence, either via Robin Hanson's brain-upload scenario, which McConaghy considers too slow, or via his preferred 'bandwidth++' scenario of steadily improving brain-computer interfaces driven by consumer markets, making humans peers of AI rather than being displaced by it.
Quotes
The AI millionaires are coming.
Trent McConaghy
If you can't beat them, join them.
Trent McConaghy
I call myself a recovering bio narcissist.
Trent McConaghy
Questions answered in this talk
What is an ArtDAO?
According to Trent McConaghy at Rise of AI 2017, an ArtDAO is a decentralized autonomous organization that uses AI to generate art, claims blockchain attribution for it, sells it on marketplaces, and reinvests the proceeds to generate more art, compounding its wealth without any owner able to switch it off.
Why does Trent McConaghy think AI is a threat even without superintelligence?
He argues at Rise of AI 2017 that even AI systems no smarter than existing ones, such as ant-colony optimization algorithms, could end up controlling most of humanity's capital resources simply because running them as self-owning DAOs is more capital-efficient, following the same logic that led BMW and semiconductor firms to shed factories.
What does Trent McConaghy propose instead of regulating AI?
At Rise of AI 2017 he proposes building a shared interplanetary computational substrate and migrating human patterns of intelligence onto it via increasingly bandwidth-rich brain-computer interfaces, so that humans become peers of AI rather than being optimized out by it.