I was going to write earlier about the dangers of centralized AI and the temptation of governments to use it to control the population, such as the bipartisan bill to nationalize TikTok and the more recent Facebook settlement of $17 billion ($6 per user) for making its app too addictive, but I got sucked into two hours of scrolling stupid videos of Chinese robots doing acrobatics and how AI is turning our brains into shit.
Anyway, yes, I could vibe code my 2008 distributed AGI by handing Claude the protocol specification that I already wrote, but that's only a minor hurdle. At the time I thought the biggest obstacle would be getting people to run peers on their personal computers. You need a computer that's always on and connected, a fixed IP address, and a firewall that allows incoming requests. At the time we had Napster and BitTorrent running on home desktop PCs, showing that it was at least possible. Those services were still vulnerable to being shut down because they depended on centralized index servers to find bootleg music and movies. So I focused on distributing the index across millions of peers, with each specializing on a single topic and knowing the locations of related peers so no peer needed a complete copy of the index. The first true peer to peer network, Bitcoin, was introduced a year later, but my algorithm would not work on it because a list of transactions (X sends N bitcoins to Y) has no ontology. Every peer needs to have access to every transaction that ever took place. This O(n^2) storage cost is the reason for the high transaction costs that makes crypto impractical for small payments. Banks don't have this problem. Running a P2P network today would be vastly more difficult. Not only do you not have access to your phone's firewall or IP address, but any app has to be approved by Google or Apple. Even if they allowed it, a server would quickly deplete your data limits and battery. Almost everyone who owns any crypto today depends on a server owned by someone else to manage transactions and often their private keys, even after massive frauds like Mt. Gox and FTX. But that's not the biggest obstacle. My proposal was basically a social network with both humans and bots, with the bots getting smarter and gradually taking over. That was fine in 2008 when social media was mostly USENET news groups and mailing lists, as long as there was a compelling (and probably illegal) reason to use it. Otherwise, nobody is going to use it unless everyone else is already on it. This was the reason that Google+ failed to compete with Facebook and Facebook's Threads failed to compete with Twitter/X. If trillion dollar companies can't create new social networks, what hope do I have? In 2008 and 2013 I estimated that AGI would cost $1 quadrillion, conditional on Moore's law making hardware costs negligible and on people willing to make all their personal data public. I didn't anticipate the end of Moore's law as transistors are now as small as possible with features only a few atoms across. I didn't anticipate the irrational objections to data centers. Why did nobody object to Bitcoin mining using 1% of the world's electricity? It seems more likely to me that people are objecting to AI in general turning the Internet into shit, a world controlled by trillionaires where nobody knows what's real. -- Matt Mahoney, [email protected] On Thu, Aug 27, 2026, 2:57 AM Quan Tesla <[email protected]> wrote: > The issue of content origin relates to authorship and IP. AI have been > absorbing public and private data like dedicated industrial vacuum cleaners > for years. It appears to be data insatiable. Did it watermark all the > original sources of all that content? I know for a fact their owners did > not. Trying to now slap a property label of ownership on that content seems > vontrived and typical to "Open X" practices we've observed for decades in > the IT industry. It's a strategy of hostile data acquisition. > > IMO, human content producers stand to forfeit what may amount to > collective thousands of years of productive manpower. One cannot put a > price on the colkective knowledge repository of a civilization, and should > not be allowed to either. > > If watermarking, then all AI-assimmilated content should first be source > watermarked, or class-action compensated for. Again, this Watermarking ploy > should be rethinked. Lawsuits are rapidly increasing against AI companies. > Human content creators will fight back. > > As matters stand now, society cannot trust the integrity of AI-product > owners and their insatiable machines. Why trust the grave robbers with > custodial rights to the cemetary? It's no secret that profit and greed and > a quest for world power generally drive AI developments. > > On Thu, 27 Aug 2026, 02:15 John Rose via AGI, <[email protected]> > wrote: > >> On Tuesday, August 25, 2026, at 4:38 PM, Matt Mahoney wrote: >> >> It looks a lot like my 2008 proposal that I wrote before smart phones, >> social media, block chain, or AI existed, but I think it has the same >> problem that you can only trust big players, which leads back to >> centralization. https://mattmahoney.net/agi2.html >> >> >> You do know that this is much easier now to build then in the past. You >> feed the doc into AI and it generates whole projects then you hand tailor >> them to build prototypes. Servers, websites, mobile apps, with devops. I >> would do all of it in C# personally, every project in the solution shared >> code. Dot NET and MAUI is very good for everything. One person could do it. >> Wouldn't that be funny. >> > *Artificial General Intelligence List <https://agi.topicbox.com/latest>* > / AGI / see discussions <https://agi.topicbox.com/groups/agi> + > participants <https://agi.topicbox.com/groups/agi/members> + > delivery options <https://agi.topicbox.com/groups/agi/subscription> > Permalink > <https://agi.topicbox.com/groups/agi/T08974f0b616697ac-M4167a6c52b8fc21cbabd2f5a> > ------------------------------------------ Artificial General Intelligence List: AGI Permalink: https://agi.topicbox.com/groups/agi/T08974f0b616697ac-Mdbbf1e5b4ab58dd262b49d0d Delivery options: https://agi.topicbox.com/groups/agi/subscription
