The AI Surveillance State

The AI Surveillance State depicted in Terms of Service represents the greatest threat to freedom the world has ever seen, an engine of oppression beyond the wildest dreams of Big Brother or the darkest nightmares of George Orwell.

The Three Pillars of the Surveillance State

The power of the surveillance state rests on three capabilities — the unholy trinity of oppression.

• Surveillance. The ability to collect data on individuals and communities.

• Power. The ability to punish, reward, and coerce.

• Analysis. The processing of surveillance data to direct the application of power.

Without surveillance, the state is blind. Without the power to act, it is impotent. Without analysis, the it has no way to identify its enemies. Analysis is often the weakest link in the chain, and this is particularly true today: the internet companies collect such vast amounts of data that it is literally impossible for them to keep an eye on everyone and figure out what they are up to. Facebook, for example, has only 15,000 content moderators to keep an eye on a user community that numbers in the billions. 

As a result, the surveillance state must focus its efforts on those who are so unwise as to draw attention to themselves, by speaking out too loudly, rebelling in too obvious a fashion, or corrupting others with their unwelcome thoughts. Big Brother can’t watch everyone, so most of us have little to fear.

This is where artificial intelligence comes into play. People are scarce and expensive. AIs, once created, are cheap and abundant. They may be replicated and deployed without limit. They can watch everyone, all the time, and bring coercive power to bear upon the rank and file of the opposition, not just on its leaders. To put it another way, with AI every individual can receive the same intense scrutiny as a prominent politician or a suspected terrorist.

How do you feel about that?

How Does AI Surveillance Work?

A lot of this is illustrated as part of the story in Chapter 1 of Terms of Service, but the basic idea is fairly simple. You take a data set, such as a bunch of photos, and you show them to a large number of humans and ask them to determine whether or not a particular object or activity is contained in that photo. It’s similar to what happens with a CAPTCHA. For example, in the image below you’re supposed to identify storefronts. Click! Click! Click! Some are ‘yes’, others are ‘no.’ Its the same process when you’re training an AI, except that instead of trying to prove that you’re not a bot, you’re creating a bot to perform a recognition task. There is a lot of complicated technology underlying this process — things in the realm of “Data Science” where you might, for example, take a photo and pre-process so that the AI isn’t working with the raw data — but the basic idea remains simple: you collect a large number of test cases along with ‘yes/no’ tags created by a human, then create a neural network or some other similar artifact that replicates the decisions of the humans. You can then take the neural network (or whatever apparatus you’ve created) and deploy it across an entire surveillance network. With enough compute power, you can run billions of images through the trigger daily.

As an experiment, I created a meme that mentioned COVID-19 without discussing it in any substantial way. I wanted to see if it drew the now-infamous ‘More About This Post’ popup that Facebook has been sprinkling around any time someone mentions COVID-19. Keep in mind that I’m small fish; there’s no way that any human manually scanned my photo to see if it talked about COVID. The popup started appearing, proving (to my satisfaction, at least) that the AIs really are looking at every photo you post, looking for what it contains. If they can find mentions of COVID, they can find other things, like American Flags, firearms, and ANTIFA banners, to name just a few examples.

Below is another example, perhaps a bit more frightening. In Australia, they have trained AIs to look into your cars from traffic cameras and determine whether you are wearing your seatbelt or talking on a cell phone. The program was highly successful; it works. But you have to ask yourself the question: If they can do that, what can they do? And what will they be able to do in the future?

There is more about this particular project in another essay: The AIs are Watching You!

Integrating Data

To take it to the next level, the AIs can start making use of auxiliary data to further refine their determinations. This is where things get truly scary. The Internet companies have a tremendous amount of information available, and all of it is potentially useful to the AIs. They know which web sites you’ve visited. They know what you’ve bought. They know what searches you’ve made. They know which postings you’ve liked and shared. They know who your friends are. They know your affiliations. This gives them tremendous capabilities.

To illustrate the process, suppose someone is interested in identifying birdwatchers. They might trigger the identification process based on identifying a bird in a photo someone posted on their Facebook feed. That is going to generate a huge number of hits, far too many for humans to look at, so they could train up a second layer of AI to conduct a more thorough investigation once they’ve spotted a bird photo. Are your friends birders? Are you a member of a birding club? How often do you post pictures of birds? Have you recently bought a pair of binoculars? They can compute a score, based on these factors, and come up with a pretty good guess as to whether or not you are a birder.

Nothing sinister there, but the same techniques can be applied to any group that might be of interest to those in power. Those who are part of the political right are very worried about this capability, and feel that it is heavily deployed against them at this very moment. It also gets used against the left, through to a lesser extent, but don’t worry: sooner or later it will be your turn. What goes around always comes around. No exceptions.

Oh, wait. Maybe you should worry.

Back to the Unholy Trinity

In order to understand how these techniques apply to the present day, we need to go back to the three pillars of the AI Surveillance state noted above. In this regard, I am somewhat concerned about the government, but the government has some very significant constraints on its abilities to use these techniques. Although weakened, our constitutional rights still provide a degree of protection from surveillance, and somewhat restrain the power of the government; they have to prove that you broke a law in order to put you in jail, which remains a time consuming and difficult process at best — far too cumbersome to apply to the average citizen.

These constraints do not apply to private corporations. There are few limitations on on the ability of a private company to collect data on their customers. There are no limitations on their power to withhold services: if Facebook wants to silence you, they can do so. If they want to ban you, they can do so. If they want to interfere with your ability to communicate or to limit your reach, they can do so, and you are powerless to assert your rights. You have none.

To understand the importance of AI to Facebook’s surveillance and control program, you have only to look at some statistics on the company:

There is literally no way that 15,000 moderators can keep an eye on billions of users, and so of course they use AI to assist them in moderating their users. It is the AIs that give them the targeting ability to perfect the Orwellian trinity.

By the way, they have given their AIs the power to silence humans.

Cue the scary horror-movie music, but it’s happening … at least I think it is; there is no transparency and there is no way to know what they’re actually doing, and therefore no reason to trust them.

It Gets Worse: The Weaponization of Corporate Power

We are now starting to see something which I feel is truly terrifying: the government is beginning to use its allies in the corporate world to bypass the constitution and the bill of rights. The government can’t spy on you (usually), and it can’t arbitrarily silence you or put you out of business, but their friends in the corporate sphere are under no such constraint. But that is a topic for another day.

Interested? Want to know more? Read Terms of Service

All of this — and more — forms the backdrop of my science fiction novel, Terms of Service. In this book, I examine the convergence of corporate power, AI, and surveillance, painting an entirely too plausible picture of where we may be heading. Read the book! It’s a great story, with compelling characters and an amazing world to explore. You will enjoy it — almost everyone who has read it has liked it — and you will learn something important about the unprecedented risks we are facing if we allow private companies to perfect the Orwellian state, as they are currently in the process of doing.

Don’t let it happen.
You cannot stop what you do not understand.