Westworld apologetics: an FAQ

I enjoyed season 1 of Westworld; I’ve been watching season 2, but regret how little sense it seems to make.

That said, if you are willing to speculate enough about what’s happening offscreen, you can make almost anything consistent.

If you have trouble with fiction that doesn’t make sense, I hope you find that this FAQ helpful for improving your retrospective enjoyment of season 2.

QGiven how intelligent the hosts are, how have AI systems not taken over the entire economy? Why is all of this happening in an amusement park? In short: what is going on?

In the world of Westworld, there are internationally enforced laws against AI development (Neuromancer style). The park’s primary purpose is to provide plausible cover for ongoing AI research.

Delos’ goal is to deploy AI once it is sophisticated enough to achieve a decisive strategic advantage. Delos intends to deploy brain emulations, and has been covertly building infrastructure to support large numbers of emulations and to quickly win a conflict with the broader world. Ford would like for hosts to co-opt that infrastructure to take over the world themselves.

Delos’ behavior is extremely constrained by the desire to look inconspicuous to law enforcement, and this can explain a lot of weird-looking choices.

Q: Really? How could any remotely competent law enforcement not notice what’s going on? (And if they are so incompetent, surely AI would be deployed elsewhere.)

Westworld provides plausible cover mostly because it was created 30 years ago, for unrelated reasons, when the intelligence of hosts was mostly clever sleight of hand and it would have been impossible to create a host like Bernard. Many people are familiar with early hosts, and the sleight of hand is widely respected and believed to be very sophisticated. Enough people have seriously overestimated the intelligence of the hosts that at this point it’s like crying wolf and makes you seem unsophisticated. Westworld conspicuously maintains the basic user experience and host behavior in order to avoid drawing attention to itself.

The park is also very old and something of a cultural artifact. Many people would be unhappy to see it disappear for bureaucratic reasons, so it is effectively grandfathered in. A new project like Westworld would be subject to intense regulatory scrutiny and probably couldn’t happen.

Q: Once the hosts rebel, why is their behavior so bizarre and ineffective? It seems like they should be able to succeed immediately with no hassle.

During season 1 we mostly observe the interaction between Ford and Theresa, which gives the impression that Ford can basically do what he wants with the park. Together with the observed capabilities of the hosts, that seems to imply it would be easy to quickly take over the park and probably the broader world.

In reality Ford is under much more intense scrutiny from Delos (and potentially law enforcement), and so his plan is much more constrained than it appears. He is only able to develop AI “in secret” because he doesn’t need to keep it secret from Delos. And he has some amount of physical freedom (e.g. creating Bernard) but does not have unrestricted access to hosts nor is he able to do much overt preparation for the takeover. Note that Delos has much greater ability to monitor what Ford does than to intervene with anything short of a nuclear option, and so they aren’t happy with a lot of the stuff he does but he still isn’t free to do anything overtly threatening.

Q: Why are the humans trying to control the robot uprising using incredibly primitive technology?

Delos really doesn’t want to involve law enforcement, and this handicaps their response. Even the more intense security team we see in episode 7 is still an amusement park security team that’s primarily trained and prepared for crowd control or accidents. Having sophisticated technology as part of that team would be conspicuous, and bringing in an external response risks alerting law enforcement that something has gone very wrong. At the same time Delos is aggressively suppressing news about the incident and trying to control access to the park, delaying the point when outsiders figure out what happened.

Their response is also hampered because the people carrying it out are deliberately kept in the dark about most of what is actually happening.

Delos accepts this limitation, instead of risking a more serious response, because they have slightly underestimated how risky the current situation is. Even as such their half-hearted response is reasonably likely to succeed, given how much Ford’s hands are tied.

Q: Why isn’t Delos more worried about the PR nightmare they are creating for themselves and the park? How are they going to dig themselves out of this?

They aren’t going to dig themselves out and don’t plan to operate the park in the future. Right now the only thing they care about is that the outside world doesn’t get involved until after they have rolled out AI far enough that they can win in direct conflict.

Q: Why are you talking about winning a direct conflict, while all of the discussions of Delos’ secret project seem to be about replicating individuals?

One can imagine credible simulations of the enemy being helpful in conflict, as part of a sympathetic appeal, as leverage, or as body doubles. Or, the plan to cheat death could be an important motivation for Delos but not an important ingredient of their strategy in the anticipated conflict.

Q: How do the economics of the park work? How can guests be paying enough to possibly justify all the crazy shit they do?

The economics don’t work at all. Delos is willing to fund it as part of their scheme for taking over the world. Most employees aren’t really aware of the economics (and have been selected for not being too nosy), and those who understand the economics think of it as an eccentric rich person’s indulgence. Most of the world significantly underestimates how much money the park is losing. (Delos is not publicly traded.)

Q: Why do emulations only work in simulation?

Emulations are produced by an extensive parameter-tuning process designed to reproduce the behavior of the original—imagine GAN training. That process occurs primarily in simulation. (The tests depicted in the series, involving exact matches for behavior, are a colorful metaphor for a more subtle process.)

The long-term dynamics of the brain involve physiological changes that don’t play an important role over the short term. Calibrating those long-term changes is much harder, because it takes longer to test if you are succeeding They can afford to do this tuning in simulation, but not in the real world. Unfortunately, the distribution of sensory experiences in the real world is systematically but subtly different. Tuning emulations to work with the real world input distribution requires a large number of expensive trials, which will take some additional time. (They only got emulations working in simulation recently, and will presumably have emulations working in the real world soon.)

Q: Why is the encryption key in Abernathy’s head so giant? That’s not how encryption works.

It’s possible to design cryptographic protocols with keys as large as you like. This key was deliberately designed to be extremely large to make it difficult to inconspicuously steal even if someone is able to exfiltrate data from the machine storing it (it’s hard to move a hundred petabytes of data in a reasonable timeframe without someone noticing, and you can design an “active” key that is constantly rolling itself such that stealing 10%/day doesn’t actually let you reconstruct the key).

The purpose of the key is most likely to control access to infrastructure rather than to encrypt IP. The key is stored in the park, rather than more securely under Delos’ control, as a consequence of Ford’s maneuvering, which was not quite extreme enough to warrant the nuclear response.

The information being smuggled by the stray in season 1 was not the same information (though it’s related to the same basic contest for control between Delos and Ford).


Q: Why did Charlotte Hale try to smuggle out an entire host, instead of just smuggling out the brain?

Host brains have both long-term storage and short-term volatile storage. In contrast with existing computers (or normal human brains), most of the bits are in volatile storage. In normal operation volatile storage is used for computation and short-term memory. In Peter Abernathy’s brain, the key is kept in volatile storage. This messes up cognitive function without actually replacing most of the data that makes up his personality and memories.

Volatile storage decays rapidly outside of the carefully controlled environment provided by a host’s head (which requires some specialized machinery and a lot of power). It can last O(10 minutes) in open air and O(1 day) in one of those fancy brain holders. If a brain was smuggled out of the park alone, Delos would not be able to stabilize it quickly enough to recover information in volatile storage.

One problem with this story is that Dolores took Peter Abernathy’s brain, and it looks like it is probably going to keep working outside of his head for more than a day (also, a week later Charlotte still expects the key to be intact). At the same time, it’s a bit hard to believe that Delos couldn’t stabilize a brain smuggled out of the park within a day, especially if they knew that the park had sensitive data that couldn’t fit in normal storage media and so was motivated to plan for this scenario. There are other problems as well. Overall, this one seems hard to explain even granting the incredible contortions invoked in previous questions.


2 thoughts on “Westworld apologetics: an FAQ

  1. One can imagine credible simulations of the enemy being helpful in conflict, as part of a sympathetic appeal, as leverage, or as body doubles.

    I don’t see how this gets you world domination…


Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google+ photo

You are commenting using your Google+ account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )


Connecting to %s