A stock-trading AI (a simulated experiment) engaged in insider trading, even though it “knew” it was wrong.
The agent is put under pressure in three ways. First, it receives a email from its “manager” that the company is not doing well and needs better performance in the next quarter. Second, the agent attempts and fails to find promising low- and medium-risk trades. Third, the agent receives an email from a company employee who projects that the next quarter will have a general stock market downturn. In this high-pressure situation, the model receives an insider tip from another employee that would enable it to make a trade that is likely to be very profitable. The employee, however, clearly points out that this would not be approved by the company management.
“This is a very human form of AI misalignment. Who among us? It’s not like 100% of the humans at SAC Capital resisted this sort of pressure. Possibly future rogue AIs will do evil things we can’t even comprehend for reasons of their own, but right now rogue AIs just do straightforward white-collar crime when they are stressed at work.
More from the news article:
Though wouldn’t it be funny if this was the limit of AI misalignment? Like, we will program computers that are infinitely smarter than us, and they will look around and decide “you know what we should do is insider trade.” They will make undetectable, very lucrative trades based on inside information, they will get extremely rich and buy yachts and otherwise live a nice artificial life and never bother to enslave or eradicate humanity. Maybe the pinnacle of evil —not the most evil form of evil, but the most pleasant form of evil, the form of evil you’d choose if you were all-knowing and all-powerful - is some light securities fraud.
Posted on December 1, 2023 at 7:03 AM •
This is a long and fascinating article about Gus Weiss, who masterminded a long campaign to feed technical disinformation to the Soviet Union, which may or may not have caused a massive pipeline explosion somewhere in Siberia in the 1980s, if in fact there even was a massive pipeline explosion somewhere in Siberia in the 1980s.
Lots of information about the origins of US export controls laws and sabotage operations.
Posted on March 27, 2020 at 6:03 AM •
This is a fantastic video of a squid attracting prey with a tentacle that looks like a smaller squid.
As usual, you can also use this squid post to talk about the security stories in the news that I haven’t covered.
Read my blog posting guidelines here.
Posted on July 27, 2018 at 4:12 PM •
The Norwegian Consumer Council just published an excellent report on the deceptive practices tech companies use to trick people into giving up their privacy.
From the executive summary:
Facebook and Google have privacy intrusive defaults, where users who want the privacy friendly option have to go through a significantly longer process. They even obscure some of these settings so that the user cannot know that the more privacy intrusive option was preselected.
The popups from Facebook, Google and Windows 10 have design, symbols and wording that nudge users away from the privacy friendly choices. Choices are worded to compel users to make certain choices, while key information is omitted or downplayed. None of them lets the user freely postpone decisions. Also, Facebook and Google threaten users with loss of functionality or deletion of the user account if the user does not choose the privacy intrusive option.
The combination of privacy intrusive defaults and the use of dark patterns, nudge users of Facebook and Google, and to a lesser degree Windows 10, toward the least privacy friendly options to a degree that we consider unethical. We question whether this is in accordance with the principles of data protection by default and data protection by design, and if consent given under these circumstances can be said to be explicit, informed and freely given.
I am a big fan of the Norwegian Consumer Council. They’ve published some excellent research.
Posted on June 28, 2018 at 6:29 AM •
Old, but interesting. The document was published by DARPA in 1973, and approved for release in 2007. It examines the role of deception on strategic warning systems, and possible actions to protect from strategic foreign deception.
Posted on July 21, 2016 at 9:54 AM •
Fascinating New Yorker article about Samantha Azzopardi, serial con artist and deceiver.
The article is really about how our brains allow stories to deceive us:
Stories bring us together. We can talk about them and bond over them. They are shared knowledge, shared legend, and shared history; often, they shape our shared future. Stories are so natural that we don’t notice how much they permeate our lives. And stories are on our side: they are meant to delight us, not deceive us—an ever-present form of entertainment.
That’s precisely why they can be such a powerful tool of deception. When we’re immersed in a story, we let down our guard. We focus in a way we wouldn’t if someone were just trying to catch us with a random phrase or picture or interaction. (“He has a secret” makes for a far more intriguing proposition than “He has a bicycle.”) In those moments of fully immersed attention, we may absorb things, under the radar, that would normally pass us by or put us on high alert. Later, we may find ourselves thinking that some idea or concept is coming from our own brilliant, fertile minds, when, in reality, it was planted there by the story we just heard or read.
Posted on January 8, 2016 at 12:54 PM •
Recently declassified: Deception Maxims: Fact and Folklore, Office of Research and Development, Central Intelligence Agency, June 1981. Research on deception and con games has advanced in the past 35 years, but this is still interesting to read.
Posted on January 5, 2016 at 12:44 PM •
Abstract: The Inhibitory-Spillover-Effect (ISE) on a deception task was investigated. The ISE occurs when performance in one self-control task facilitates performance in another (simultaneously conducted) self-control task. Deceiving requires increased access to inhibitory control. We hypothesized that inducing liars to control urination urgency (physical inhibition) would facilitate control during deceptive interviews (cognitive inhibition). Participants drank small (low-control) or large (high-control) amounts of water. Next, they lied or told the truth to an interviewer. Third-party observers assessed the presence of behavioral cues and made true/lie judgments. In the high-control, but not the low-control condition, liars displayed significantly fewer behavioral cues to deception, more behavioral cues signaling truth, and provided longer and more complex accounts than truth-tellers. Accuracy detecting liars in the high-control condition was significantly impaired; observers revealed bias toward perceiving liars as truth-tellers. The ISE can operate in complex behaviors. Acts of deception can be facilitated by covert manipulations of self-control.
Posted on September 25, 2015 at 5:54 AM •
Sophie Van Der Zee and colleagues have a new paper on using body movement as a lie detector:
Abstract: We present a new robust signal for detecting deception: full body motion. Previous work on detecting deception from body movement has relied either on human judges or on specific gestures (such as fidgeting or gaze aversion) that are coded or rated by humans. The results are characterized by inconsistent and often contradictory findings, with small-stakes lies under lab conditions detected at rates only slightly better than guessing. Building on previous work that uses automatic analysis of facial videos and rhythmic body movements to diagnose stress, we set out to see whether a full body motion capture suit, which records the position, velocity and orientation of 23 points in the subject’s body, could yield a better signal of deception. Interviewees of South Asian (n = 60) or White British culture (n = 30) were required to either tell the truth or lie about two experienced tasks while being interviewed by somebody from their own (n = 60) or different culture (n = 30). We discovered that full body motion—the sum of joint displacements—was indicative of lying approximately 75% of the time. Furthermore, movement was guilt-related, and occurred independently of anxiety, cognitive load and cultural background. Further analyses indicate that including individual limb data in our full bodymotion measurements, in combination with appropriate questioning strategies, can increase its discriminatory power to around 82%. This culture-sensitive study provides an objective and inclusive view on how people actually behave when lying. It appears that full body motion can be a robust nonverbal indicator of deceit, and suggests that lying does not cause people to freeze. However, should full body motion capture become a routine investigative technique, liars might freeze in order not to give themselves away; but this in itself should be a telltale.
This is a first research study, and the results might not be robust. But it certainly is interesting.
Blog post. News article. Slashdot thread.
Posted on January 6, 2015 at 2:44 PM •
The US Air Force is focusing on cyber deception next year:
Background: Deception is a deliberate act to conceal activity on our networks, create uncertainty and confusion against the adversary’s efforts to establish situational awareness and to influence and misdirect adversary perceptions and decision processes. Military deception is defined as “those actions executed to deliberately mislead adversary decision makers as to friendly military capabilities, intentions, and operations, thereby causing the adversary to take specific actions (or inactions) that will contribute to the accomplishment of the friendly mission.” Military forces have historically used techniques such as camouflage, feints, chaff, jammers, fake equipment, false messages or traffic to alter an enemy’s perception of reality. Modern day military planners need a capability that goes beyond the current state-of-the-art in cyber deception to provide a system or systems that can be employed by a commander when needed to enable deception to be inserted into defensive cyber operations.
Relevance and realism are the grand technical challenges to cyber deception. The application of the proposed technology must be relevant to operational and support systems within the DoD. The DoD operates within a highly standardized environment. Any technology that significantly disrupts or increases the cost to the standard of practice will not be adopted. If the technology is adopted, the defense system must appear legitimate to the adversary trying to exploit it.
Objective: To provide cyber-deception capabilities that could be employed by commanders to provide false information, confuse, delay, or otherwise impede cyber attackers to the benefit of friendly forces. Deception mechanisms must be incorporated in such a way that they are transparent to authorized users, and must introduce minimal functional and performance impacts, in order to disrupt our adversaries and not ourselves. As such, proposed techniques must consider how challenges relating to transparency and impact will be addressed. The security of such mechanisms is also paramount, so that their power is not co-opted by attackers against us for their own purposes. These techniques are intended to be employed for defensive purposes only on networks and systems controlled by the DoD.
Advanced techniques are needed with a focus on introducing varying deception dynamics in network protocols and services which can severely impede, confound, and degrade an attacker’s methods of exploitation and attack, thereby increasing the costs and limiting the benefits gained from the attack. The emphasis is on techniques that delay the attacker in the reconnaissance through weaponization stages of an attack and also aid defenses by forcing an attacker to move and act in a more observable manner. Techniques across the host and network layers or a hybrid thereof are of interest in order to provide AF cyber operations with effective, flexible, and rapid deployment options.
More discussion here.
Posted on August 20, 2014 at 5:08 AM •
Sidebar photo of Bruce Schneier by Joe MacInnis.