Sunday, June 17, 2018

the Internet as the AGI hardware substrate

What type of hardware substrate do you envision for the first artificial general intelligence (AGI)?

Many scenarios seem to envision AGI emerging in a particular computation system, and/or perhaps in a robot or robots. Many scenarios involve the system being connected to the Internet, as a crucial aspect.

What I tend to envision is that when an AGI first emerges, the substrate will be millions or billions of Internet-connected devices, along with their human operators. Not only will the collective computing power of all these Internet-connected devices dwarf the computing power of any particular system, but the inclusion of millions/billions of humans in the mix may also be crucial.

The software platforms through which billions of people currently interact -- Google, Facebook, Twitter, etc. -- are (compared to what seems possible) fairly stupid/limited, focused significantly on goals like maximizing users' engagement/attention in order to more profitably sell their attention to advertisers. But creating more intelligent ways of connecting these existing biological intelligences (ourselves), to each other and to software and information, could well help us congeal into a more effective, coherent global mind, which would begin to have many of the capabilities that we expect AGI will have.

Such scenarios have been envisioned by science fiction authors and others. The salient strategy that seems to suggest itself is for AGI projects to be structured as social media platforms. I've blogged about some of what I'd like to see such platforms do in the early stages:

  • focus on post-verbal communication, using graphics and nonverbal sounds instead of words, pervasively and increasingly.
  • focus on giving users the greatest possible control over the behavior of their interfaces.

This can facilitate vastly more efficient cross-pollination of information and of programming components: We'll see, in this graphical environment, visual representations of information, including the tools that others are using to filter/curate/manage their info-streams. We can take those tools and plug them into our own control panels, which regulate the behavior of our interfaces/info-streams, allowing us to more efficiently locate and create additional information (including additional tools for info-stream filtration/curation/management), and so on.

This strategy seems like a great way to optimize our screen time and speed us toward optimal cultural evolution, if nothing else. And I suggest that through it could emerge something we would feel compelled to recognize as AGI.

This is not to say that existing and future AGI projects with complex cognitive architectures running on particular systems are misguided strategies. It's just to say that such architectures can - and for maximum impact, should - be deployed in an open, distributed, social media context, contributing maximally to the creation of optimal experiences for Net users and to the growing intelligence of the Network (all interconnected devices and people) as a whole.

1 comment:

Josh Maurice said...

I also posted this in the "Real AGI" Facebook group, and there were some interesting comments. [Link]