Basilisk Hacks - Evolutionary Algorithms and Psychological Warfare Against the Entire World
Part 14 of Artificial General Intelligence (And Superintelligence) And How To Survive It
The idea that we keep so many flaws in our systems unrepaired and unguarded has always been insane.
Everyone talking about seed AI running out of control for decades harped on how quickly an emergent AI could seize processing power with botnets, move money through offshore accounts, operate undetectably...
And a multitude of other options which begged the question of why we were preserving these weaknesses in our networks. All ripe for organized crime, cyber-criminals…
Or hostile foreign intelligence agencies.
Shouldn't an ambitious hacker, foreign-intel operation or runaway AI wonder whether their activities might be picked up on automatically?
Whether they have even the slightest sense of how formidable the technologies deployed to detect, deter or decimate them might be?
It’s an excellent place to start this discussion.
Another would be a terrifying weapon which could employed against us by a hostile superintelligence, the not-so-hypothetical "basilisk hack."
David Langford created the concept of basilisks - images triggering thoughts a human mind can't handle, cognitively or physically. Yes, some neural networks literally have this vulnerability, even to a single, precisely placed pixel.
Other writers made them even more formidable and subversive, ultimately rewriting the target's mind or even their entire body via nanotechnology, genetic engineering, hypnosis and so forth.
Terrifying concepts, supposedly safely in the world of science fiction.
But while a human is not so easily analyzed and hacked, while our “source code” is much more complex and our systems squishy, there’s a way to do a kind of basilisk hack, given sufficient stimuli, willing feedback, and especially time.
Now we come to evolutionary algorithms in psychological warfare - what the tools reputedly used by organizations such as Cambridge Analytica appear to be, whether they conceived of them in this way or not.
Cambridge Analytica was one of those organizations who reputedly interfered in America’s 2016 election via social-media influence. Russia’s government, famously, has also been accused of such activity. Just how pernicious foreign interference could be is hard to grasp until you think through the consequences.
This is psychological warfare, or PSYOPS. And here’s how you do it.
To use evolutionary algorithms in psychological warfare, you need data, feedback, and the means to manipulate.
Social media has these tools, but also vast inherent weaknesses for the perpetrator.
To use evolutionary algorithms in psychological warfare:
Just break up stimuli meant to have an impact into discrete units - messages, emails, tweets, posts - target based on your models, check reactions, remix and then try again. To millions of people, as many times as you need to.
You’re just trying to get a reaction, and to gauge it, at first. You don’t need to know what it is or how strong at first. Initially, finding out is the point.
The most psychologically vulnerable will tend to have the strongest and most easily stimulated reactions.
Think about someone with PTSD, paranoia, obsessions, aversions, phobias, hypertension. Also remember that many millions of American medical and psychological records have been famously hacked, and the data presumably used or sold, so in many cases an aggressor may have some targets extensively, psychologically mapped out to start with.
So you do immense psychological harm, especially with a target group of 100 or 200 million voters or more. But you also provide an immense map of every asset involved in spreading, modifying, parsing, targeting and sustaining this propaganda.
Using this technology is absolutely lethal. To the wielder, especially.
Why? Because if you're targeting tens of millions, potentially a couple hundred million Americans, you're targeting not just normal citizens, but those suffering from PTSD, paranoia, severe aversions, early dementia, and so forth. People expressly vulnerable to this manipulation.
And you know it.
But it also represents unfathomable oceans of hard evidence, showing deception, active collusion, guilt. And integrates easily with other evidence, online or elsewhere.
With automated and micro-targeted propaganda it's always a similar process, informed by your targets' reactions to your latest efforts, their previous responses and psychographic profile. You don't just elicit responses. You drive them to particular decisions or attitudes over time.
But your key targets are not your only test subjects. You're studying people on an incredible scale. So you have people with matching or near-matching profiles in less critical locations - non-swing states, voters with more-certain voting profiles, and so on.
You can target these non-critical demographics to reap response data in initial rollouts, but also to see what beneficial effects you can create for your operation. Creating a larger social circle that will affirm articles you want affirmed, decry actions you want decried.
This creates insane volumes of hard evidence, in context (which you're often laying out yourself), for millions of crimes.
It's not just a specific video clip or article or infographic. When social-media companies can control who sees it, you can use these actions to reinforce beliefs or to stir up conflict and division.
Who sends out the original piece? A trust-washed media organization? A public figure? A well-known troll? Does a troll use a friendly, disconcerting or repellent avatar? Such as a swastika or fascist-frog avatar, in the latter case?
What individual or group reactions are you getting? Even in your non-critical demos? Who is close to or relatively isolated from more significant targets? Can you demoralize volunteers? Silence critics or opposition organizers? Embolden the worst elements?
One further advantage of evolutionary algorithms in an influence campaign that is glorified sabotage: Normally you can't just use, say, random drugs on humans to look for a potent medicine or other effector. People will die. They don't even do that with lab rats - not unguided.
But when your agents are anonymous and you're trying to cause long-term damage, you may think you can get away with destroying a great deal. You will learn otherwise. But in the meantime, it is everyone left holding the bag who ends up in mortal peril.
For this all becomes so extensive and so well documented, there is no escape. Your destruction is written by your own hand.
Again, you can take a bunch of program fragments, throw them a problem, keep what works best and toss the rest. What happens if you do the same thing with stimulus in psychological warfare? Aside from incredible harm and evil?
It's one thing to have mountains of hard evidence regarding crimes. But the computers can literally record every minute detail and action in the felonies, and archive the entire operation. Vast, complex PSYOPS managed with the precision of supercomputer.
Now imagine if some group were hypothetically involved, such as Cambridge Analytica, or a social-media platform like Facebook - whether directly or simply as the provider of ads and engagement so precise you can oversee the whole project through their commercial tools. Imagine if the records of the actions undertaken, and even the computers doing it, could be seized.
Not just the evidence but the crimes themselves, frozen in evidentiary amber.
Consider GIFs, video clips, hashtags and endlessly repeated phrases in text. Think about all the precise pieces of stimulus not only visible as being managed, but easily tracked across a host of servers and databases, as very specific code came up again and again.
Then think of all the tech companies, media assets, entertainment assets, politicians, hackers and others being precisely coordinated with similar talking points and timetables. Who is saying and promoting what will not only be obvious, but so will the timing and the coordination.
Imagine them showing further collusion in the evolutionary algorithms, botnet support, bitcoin payments, and so forth.
Realize exchanging money in a conspiracy is evidence of cooperation. Overlay the information above with data from cryptocurrency exchanges, illicit Deutsche Bank, Danske Bank, Cypriot, Caymans, Swiss and other bank transfers, non-traditional money laundering via clickfraud, micropayments, real estate and fraud. Among other things.
Imagine all of this being integrated with vast archives of not just classified intelligence intercepts, but evidence seized from conspirators and transnational organized crime.
How much evidence?
Prosecutors provided 1.5 million files in the first batch of evidence during discovery before the trial of one spy, Maria Butina. 4-million+ pages of Michael Cohen's seized files were cleared for prosecutors’ use by a special master.
Cohen was known for recording Trump and key phone calls. We’ve also had evidence seized from Roger Stone, who also recorded calls. The there’s the seizure of the National Inquirer's vault. And then we have any kompromat or other data found in the investigation of Jeffrey Epstein.
To the extent any of these individuals were involved in online-influence operations, particularly involving the 2016 election, their records would shed even further light.
So, quite a bit of evidence which could enrich any such investigation comes to mind.
If this sounds like work for a supercomputer, it undoubtedly is.
Regardless of who gets prosecuted for the overarching conspiracy, presumably a multitude of conspirators and their employees committed wire fraud or tax evasion, and are apt to face charges one way or another.
Oh, and whether or not I was right about malware being a map, the FBI followed malware in dozens of hacks of companies across several nations back to Advanced Persistent Threat 10, I believe. In less than 3 months after my first contribution on tracking malware. Going after Chinese intelligence, in China.
Also contemplate the fiduciary duty of companies/organizations' legal teams to the incorporated entity - not an executive, CEO or foreign dictator. There are literally conspirators relying on legal departments with every reason to throw them under the bus. With hand grenades.
And further - are participants proffering to prosecutors? Becoming a fully cooperating witness means telling them not only about any crime you’ve ever been involved in, but any crime you’ve ever been aware of.
And then beyond all of that, we get into some of the basic, terrifying technology. Why is quantum computing a brain-melting, panic-inducing threat here? Besides quantum decryption?
The other, even more eathshakingly powerful quantum “killer app” is amplifying our capacity to search databases. Grover’s algorithm opens the way to quantum search, and all it implies.
Amplifying basic search capabilities when you are dealing with an unfathomably vast sea of admissible evidence inside a living ocean of accessible intel means patterns are immediately available.
Conspiracy and crime online in an era where data is inherently recorded in so many ways meant there was always an event horizon approaching.
One in which every compromised asset - much less their networks, much less their ultimate employers - got burned.
Too many millions of connections exist between too many thousands of intel and organized-crime assets.
Even if people were clever. Which, for the record, they were not.
But when you have multiple terabytes of evidence acquired in just individual archive seizures - bin Laden, Cohen, etc - being able to use artificial intelligence or quantum supercomputing dramatically augments your ability to sift not just for keywords or oddly similar actions.
Your ability to check evidence - particular evidence the perpetrators never knew existed - against other databases of activities, seized communications, known associates, movements, activities, hacks, malware and on and on - increases exponentially.
Some searches today that would just be too vast to even attempt, or that would take years for present supercomputers to accomplish, would become trivial, and even routine.
As a criminal or conspirator, you don't even have to be online yourself. So much is now exposed, in so many ways, that all of these operations are fatally wounded, whether they know it or not.
Why does all this matter to AI safety?
Beyond the implications of such an attack taking place through our media and social media, to influence our elections and wage psychological warfare against our population?
Because these tools could have been used by an intelligent or superintelligent adversary far more effectively. Instead these inept efforts have burned down not only this entire strategy, going forward, but exposed endless trolls, tools and technologies, and assets and agents of influence all too willing to play these games against all of us.
Which means they can’t be used again, and anyone who tries will only get burned.