New Scientist quotes me on the simulation argument:
Although we are unlikely to get proof, we might find some hints about our reality. “I think it might be feasible to get evidence that would at least give weak clues,” says Bostrom.
Economist Robin Hanson of George Mason University in Fairfax, Virginia, is not so sure. If we did find anything out, the operators could just rewind everything back to a point where the clue could be erased. “We won’t ever notice if they don’t want us to,” Hanson says. Anyway, seeking the truth might even be asking for trouble. We could be accused of ruining our creators’ fun and cause them to pull the plug.
Hanson has a slightly different take on the argument. “Small simulations should be far more numerous than large ones,” he says. That’s why he thinks it is far more likely that he lives in a simulation where he is the only conscious, interesting being. In other words, everyone else is an extra: a zombie, if you will. However, he would have no way of knowing, which brings us back to Descartes.
The reporter gets this a bit wrong. If I’m in simulation, I’m more likely to be in a small than a big simluation, but that is not to say I’d be “the only conscious, interesting being.” I’d guess most small simulations with any conscious beings have more than one – humans are social, and most of the interesting things to simulate about humans require more than one of them.
Robin, have you worked with simulations? My own experience with them suggests it is much more of an open question whether the being simulated will ever have consciousness. ESPECIALLY since we are essentially clueless about where our own consciousness comes from.
In a sim, you tend to build in an 1) approximation of the rules that 2) you know about. Consider the lowly neuron, we know it fires in a randomish way with increasing rate under some circumstance, with decreasing rate under other circumstances. In the sim, do we just put in the fixed rates? Do we use a random number generator to populate a randomish variation which we think has some of the same statistics as the real neurons?
The problem is that what seems randomish may, and often does, reflect some more underlying rules. What if that randomish variation is not random at all, or not completely random, but rather is part of some mechanism associated with consciousness? We believe consciousness is weakly connected to the brain. We know we can focus our consciousness on some things the brain is dealing with, the text box on the screen in front of me, the part of my brain holding these ideas and transforming them to language. What if that wandering weak focus is mediated through some mechanism that shows up, in our state of ignorance, as randomish variations in neuronal firing rates? Then our sim would MISS this entirely.
As with any complex functioning system, it is much easier to build something that doesn't work than something that does. The naive engine builder puts all the pieces together and goes to fire up the engine, and it won't even rotate forget-about firing up. The programmer is constantly building something that doesn't work and then working hard to figure out why it doesn't work.
I realize I am sounding just pessimistic. Of course work should go forward on this, and will anyway as coding increasingly complex processors is one of the best ways to create valuable capital these days (in addition to being unbelievably fun). But I would expect brain sims to "go wrong" in a lot of ways for a long time. And that consciousness with shifting focus and motivation and humor and insignt and the ability to invent and triage inventions... I wouldn't expect that to show up in the sims until somewhere between pretty late in the game and never.
When I first started forming my thoughts about sims, I recognized two things. 1) in a simulation of a nuclear explosion, nothing is destroyed, no energy is released. It is gigantically different from an actual nuclear explosion. 2) We already simulate people. In Vice City, I shoot cops and passing civilians all the time. The sims of people are way simpler that real people and I think beyond reasonable doubt they have no more consciousness than do the cars or the buildings in Vice City. Does consciousness just pop in by magic at some point as we make the sim more complex? Or do we actually have to understand how consiousness works, and put the necessary physics to support it into our sim, whereas we would currently just use random number generators which reflect not the physics, but our ignorance of the physics? Having worked hard to get things to work, I am pretty sure it is the latter. We will not have conscious sims until we have some idea how consciousness works, and is connected wtih brains and neurons.
I'm totally serious. The real year is 2101, and the real original Robin recently turned 142. I'm his psychiatrist and I've been running this em simulation of all the original folks connected with AI - it's a 'software diagnostic' sim for the purposes of verifying friendliness of the final AGI design. Nah, just kidding ;)
But how do we know it's not true? Pick three items of very high scientific significance. (1) The number of days it takes the moon to orbit the Earth. (2) The number of bones in the human hand, (3) The circumference of the LHC tunnel (km). Look up the answers on wikipedia. I kid you not. The '27' does look eerily like an Easter egg in a simulation....