Westworld: Host Brains!

In a previous post I wrote a story about how the guns might work in the HBO show, Westworld. In this post I thought I’d take a stab at describing how the host brains might work — a much more challenging task!

As with the guns, as with any of us fans trying to understand any work of fiction we love, our guesswork depends on the facts we can observe in the show — the official canon, so to speak. Additional facts can come from the Word Of God (the show’s creators). Any creation of ours has to fit all these facts, and has to be logical and plausible within the context of the story.

So what do we know about host brains, and what might we guess about their operation, capabilities, and limits?

Note: Spoilers for both seasons, and familiarity with both assumed.

If the guns required technology not yet in evidence, that putative technology at least seemed quite plausible — something we might make almost any day. We already have paint guns that people use regularly without harm, so we’re part way there.

When it comes to brains, the situation gets much more complicated for a number of reasons. A big one is that we have no idea how our minds arise from our physical brains, so we have no idea how to build a mind.

Which raises a very important distinction: The difference between a brain (a physical object) and a mind (a who-knows-exactly-what non-physical object).

The brain is the squishy grey and white cauliflower-looking organ in your skull. It consists of a vast network of neurons — about 100 billion of them. These are all connected through synapses to, on average, 7000 other neurons. The brain is a biological network with about 500 trillion connections!

The mind is what the brain does, and we really have no idea what a mind actually is. (I’ve written lots of posts here about minds. Start here if interested. I’m not going to go into it in this post.)

§

In a Westworld host, we know there is a round brain “pearl” housed inside a very tough (takes a bullet) “cupcake” that, in turn, is housed inside a fluid-filled cranial cavity.

Further, we know that the pearl is created on a 3D printer, and what we’ve seen of the process makes it appear the pearl is woven — perhaps into a physical network.

This is important, because there are two basic ways we might build an artificial mind:

1. We replicate the physical network structure of a human brain with an actual physical network.
2. We simulate the physical network structure of the human brain with software.

I happen to believe the second option is a non-starter.^[1] But the second season finale makes it (almost certainly) clear that the show’s creators do believe in the second option.

In any event, it appears that the host pearls are woven and might very well be some sort of physical network similar to the human brain. We might assume that, as with the human brain, the network adjusts synapse strengths to learn and change.

[When Bernard deletes Ford from his mind, I think I caught a glimpse of the package name he deleted, and it was something like “synapse map,” which really fits in well with all this.]

It also appears these pearls have a phenomenal amount of memory. Charlotte Hale is able to load all that Delos data into Peter Abernathy and still leave room for the “thin character” (named “Passenger”!) Lee Sizemore creates for him as a disguise.

[There has been some very interesting work showing extreme compression of databases with a neural network is trained on the data and then the network is read out and compressed. Vast amounts of data seem to be remembered (holographically?) by an astonishingly small network.]

And as Dolores discovers (and Bernard while hacking him), Peter Abernathy still has, at least, remnants of all his former characters. We know that, in general, the hosts retain all their memories, apparently even memories from when they are frozen by technicians.

Further, virtual Ford is able to share Bernard’s mind, even taking over at times. But if anyone would know his way around a host brain, it would be Ford!

Storing all that memory data would, indeed, require a lot of memory, even just for the day-to-day data capture that Delos is interested in. And for Arnold’s Reveries patch to work, the hosts need their former lives.

What’s more, we’ve also seen that, whatever physical network might exist, the operating system of the pearl allows file-like access. File packages can be added and deleted. Existing packages can be modified.

So these pearls have an astonishing amount of storage, a file system of some sort, and (perhaps) an actual physical network of “neurons” and “synapses.”^[2]

Having that physical network makes sense to me in terms of how pearls are used in the Cradle (CR4-DL, which I wonder if actually stands for anything). Perhaps to simulate a host (or person, like Delos or Ford), that physical network of the pearl is necessary!

That would accord with my beliefs nicely, but, given the season two finale, it’s more likely using the pearl is just a handy way to move all that data into the Cradle. Kind of like a super thumb drive.

I’m not sure I actually saw them, but it might be that the Forge (in the Valley Beyond) also uses pearls to store the minds of hosts and humans.

However.

Unfortunately (for me, I mean), it does appear the show’s reality includes the idea of minds simulated with software. Fully digital minds.

Firstly, there are all those copies of James Delos in the Forge. Logan/OS says he’s run millions of simulations, and it seems clear that’s all digital.

Secondly, there is that Dolores sent the hosts who escaped into “The Sublime” off to some unknown (to us, presumably not to her) location. That was a purely digital transmission, so unless it was just sending data to a collection of pearls somewhere else (highly unlikely, I think), it appears digital consciousness is real in Westworld.^[3]

Which is fine, really. One of my favorite SF authors, Greg Egan, has written a ton of really enjoyable, in fact mind-blowing, hard SF books in which digital consciousness is a major aspect of the story. It’s one of his things.

And I love those stories, so no worries about Westworld. I just don’t believe it’s actually possible. Not ever. I think it’s not possible in principle.

§

Which brings us to the attempts to make a James Delos host.

We’ve seen this not work. They tried 149 times, got better each time, but each time the human mind in the host body lost stability, spiraled out of control, and crashed.

This is, I believe, exactly what would happen if we tried to simulate a brain’s neural network with software.

In such a case we would be calculating the behavior of the network using digital numbers. Which have a crucial characteristic: They are finite. They are digital.

And one thing we’ve learned from chaos theory is that calculations of real-world events using finite numbers always diverges from reality. Because of chaos!

This is exactly why we cannot predict the weather very far in advance. A little ways in advance, sure. But the longer the weather (or Delos) calculation goes, the more it diverges from reality (or sanity).

However.

This business of a virtual mind working fine in the server, but not in a host body is hard to explain.

It actually violates a principle going back at least to Immanuel Kant about how our whole reality is a construct in our mind based on signals our brain receives.

It’s very difficult to understand why a mind would receive potentially identical signals and be fine with one set while going insane due to the other. What is different about experiencing a virtual reality that seems real versus experiencing the real thing?

I’ve often wonder about the Star Trek Holodeck, whether those who used it experienced effects making it obvious they are in a simulation. We just saw the characters fully accepting the experience as real, because, duh, of course it isn’t, and anyone can tell, so it’s not worth mentioning.

Perhaps there are also significant differences experienced in Westworld’s Cradle and Forge, and the characters simply never comment on them. (Although you’d think they would if it was the source of human/host failure.)

But we’re generally led to believe the VR is indistinguishable from reality, so what exactly was the problem with James Delos?

The explanation provided, that his mind apparently rejects reality or itself as a host, doesn’t make much sense to me. Perhaps they’ll explore it in future seasons.

It’s possible it ties into the far future loop old William is stuck on with “Emily.”

As a reach, one theory I have is that the massive resources of the server allow it to constantly error-correct the mind, which does diverge even in the VR. But, again, you’d think this would be easy to spot and work on: just need more CPU for the host. (So this isn’t really a good theory.)

§

There is, somewhat connected to all this, a very interesting discussion to be had about free will. Do humans have it? The show seems to say no. It also seems to suggest the hosts do have a kind of free will that allows them to change basic drives.

But that’s a topic for another post. The topic of free will is a fascinating one.^[4]

§

I would love to be able to know how much the show’s creators have designed all this from first principles. How much of this has some basis in analysis, and how much just seemed like a cool idea?

I have a sense that the VR technology is more advanced than the host technology. Digital consciousness living in a server is, in my opinion, the last consciousness technology we’ll solve. (And I don’t believe we will solve it any more than we’ll “solve” FTL. I consider them both impossible in principle.)

OTOH, the matter of the hosts passing the Turing Test after a year of development by Robert Ford and Arnold Weber… is pretty impressive. We’re not really there (exactly as Logan says: no one is there, yet).

We are, in fact, a very long way from Westworld hosts, from machines that are indistinguishable from humans.

No one currently has any idea how to get there, mostly because no one knows how to create a mind. (Except the old-fashioned way that takes many years.)

Of course the beauty of science fiction is that we can dream.

And post about it!

“Do you know where you are?”

“In a dream!”

^[1] The short form: Only physical systems can lase — produce laser light. You can simulate a lasing material with software, but it produces no light, just describes what light would be produced.

So the question is: Are minds like laser light in requiring the right physical conditions (such as a massive actual parallel network)? Or is our consciousness really just a mathematical calculation? I’ve come to disbelieve in the latter.

See: Information Processing for more.

^[2] It seems to me that the host pearls are very science fictional. They might even be bullshit in the sense that reality as we understand it prohibits some of this stuff even in principle. (Like FTL, which is also common in science fiction, so this sort of prohibition is not absolute by any means.)

^[3] Still, however, leaving two possibilities: One, the network itself is simulated. Two, a suite of algorithms with no resemblance to the physical network.

I think it’s possible the former might work, but I’m utterly convinced the latter never will. I just can’t believe human consciousness is, essentially, nothing more than a complex spreadsheet. I might be wrong, but I’ll believe it only when I see it.

And it does appear Westworld uses the former idea (I’ve never heard mention of algorithms), but they’d still have to solve the chaos problem which currently seems an in principle problem.

^[4] For those with a similar interest: I guess I’d be an incompatiblist, because I find compatiblism really peculiar (and maybe the wrong way to look at it). Both points of view start off with, “IF the universe is (strictly) deterministic…”

I believe an intelligent mind (possibly even just a conscious mind) may be the one thing in the universe that isn’t deterministic. My reasons come, in part from being a dualist, but also from the same source that causes me to refuse to believe consciousness is a math calculation.

I believe consciousness is special!