February 5, 2011

SAI in the material world

Mondolithic Studios
A meme that gets bandied about by Singularity denialists is the assertion that a rogue super-artificial intelligence (SAI) won't be able to act on its desires and make a real impact on the physical world. It's just a really sophisticated computer, goes the thinking; it couldn't actually reach out and touch someone.

This runs contrary to the concerns of those in the Singularity camp who are gravely concerned that an SAI will be both uncontainable and capable of manipulating physical space in a non-trivial way.

I'd like to present a pair of  arguments that will serve as a warning to those who would like to dismiss this possibility. The first is based on a recent technological breakthrough, the second being more of a thought experiment.

Robotic networking and self-replication

RoboEarth is a system that's allowing robots to build on and learn from the experiences of other robots. Think of it as an internet for robots. As it stands, robotics engineers have to teach their bots to navigate and function in the real world. RoboEarth, on the other hand, collects and centralizes information on objects and navigation, and in turn shares this information with other bots. What this means is that any new robot that's connected to this system will have immediate knowledge of its surroundings.

But it doesn't stop there. A recent breakthrough has endowed the TechUnited AMIGO robot with the ability to download all the information it needs for a specific task and then carry out that task. Check out the video below of AMIGO at work:

If this doesn't blow your mind then you're not paying attention. While the task was simple enough, that of autonomously picking up and serving a bottle of water to a person, the potential implications of this are huge. As Joris Peels of iMaterialize clarifies,
If you would combine Robo Earth, with genetic algorithms that automatically design robots and 3D printing you have a very powerful combination. It would be a system that could design a robot based on its experiences, then give that robot all the information it needed to navigate the world and carry out tasks. Anyone could then 3D print this robot anywhere around the world. And the system would be one of continuous learning and itteration with better robots being made every second. We’re still very far away from this but it is these kind of ongoing developments that make me think that I live in the future.

I think we should really consider the implications of this. I know, it sounds a bit sci fi and off piste. But, we will develop a Skynet at one point and we should consider the implications before we do so.
The scenario I'm imagining, as I'm sure are other Singularity-concerned futurists, would see an SAI co-opt this system (or create versions of its own) and begin to fulfill its intentions through a myriad of self-designed, recursively improving, and remotely controlled agents disbursed around the world.

Plenty of room at the bottom

Okay, so there's that example. The next consideration is something a bit more fantastical (relatively speaking): the potential for an SAI to reshape the planet (or significant portions of it) from the molecular scale upwards. Before you tune out, watch this video, Molecular Visualizations of DNA:

What you're seeing in this video is a very small sampling of the kinds of molecular machinery that's capable of arising through the processes of natural selection. What you're not seeing here, however, is the space of all possible molecular machinery that's capable of arising through intentional design. And what you're definitely not seeing here is the space of all possible molecular machinery that's capable of arising through intentional super-intelligent design.

The kinds of molecular machinery that we're familiar with has come about solely for the purpose of maintaining and propagating complex organisms. We're only beginning to imagine the kinds of molecular-scale processes and devices that might be designed to perform other kinds of functions; the design space is massive.

And this is where an SAI comes in. It's easy to imagine a system similar to RoboEarth in which an SAI can design and disburse both macro and micro scale devices. The only limitations facing such a system would be inherent energy and material constraints, other human or SAI-driven countermeasures, and the laws of physics itself.

Okay, what exactly am I imagining? Given free reign, an SAI could potentially re-arrange all matter on the planet. One possibility is that it could turn the Earth into computronium or anything else it wants. Or, it could remove all toxins and other pollutants from the surface and atmosphere. It could turn the planet into a Venusian hell, or a verdant Utopian paradise. Whatever. In all honestly, I can't even really begin to speculate without knowing the intentionality of a Singularity-surviving intelligence. But suffice to say the scope of its impact on the material world needn't be subtle.

For those of us engaged in foresight activities, the risk is in thinking too small on this matter—or in denying it altogether.


Sean Strange said...

Aren't issues of SAI risk so speculative and unknowable that it's almost a waste of time worry about them? The way I see it, a Singularity is by definition unpredictable; if it happens, all bets are off anyway, and if it doesn't then who cares about it? Humanity has to continue to live and progress now, but if we took the idea of an imminent Singularity seriously, it seems to me there wouldn't be much incentive to do anything, since anything we do now will quickly be surpassed and made absurd by these coming god-machines. So my attitude is to proceed as if SAI is a fantasy, and start worrying about it if it begins to look like a serious possibility.

Mike Treder said...

But is ‘computronium’ really ‘unobtainium’?


ZarPaulus said...

For some reason RoboEarth brought this to mind:

Martin Andersen said...


We could get information about what would happen in a singularity, if we could observe other civilizations in our galaxy. Maybe some of them would send out a kind of distress signal or warning to others, just before they reach the singularity.

The problem is we need very advanced technology to catch these signals, so when we have it, we are probably very close to the singularity.