Disinformation Cascades, Espionage & Counter-Intelligence

ABSTRACT

Information cascades occur when decision-makers beyond some point in a decision-making sequence find that it is optimal to follow the decision that others have taken even if their own private information indicates a contrary course of action. Information cascades are studied because they can help us to explain decision-making conformity and behavior convergence, including fads and fashions. Cascades can be correct or incorrect, strong or fragile. The intelligence community has often been concerned with deception and disinformation. We call an incorrect information cascade precipitated by deception a ‘disinformation cascade.’ How easily can a disinformation cascade be started? How many intelligence officers must choose a certain way before a disinformation cascade starts? These are the questions we address in this paper.

KEYWORDS:

• Disinformation
• Information Cascades
• Intelligence Agencies
• Decision Theory

Notes

1. See Whaley (1969), Hendricks and McAfee (2006) and Clements (2014).

2. Though it has been around for longer. See Maiolo (1999), Hulnick (2002), Reiss (2019) and Zamir (2019).

3. Doubly so when one considers that during the 1950s Kim Philby was one of the foreign correspondents working out of Beirut for The Observer and The Economist.

4. It is possible that the East German security service (‘the Stasi’) were involved (Geissler & Sprinkle, 2013). This would not be surprising given the widely circulated quote attributed to Colonel Rolf Wagenbreth, director of Department X (disinformation) of East German foreign intelligence: “Our friends in Moscow call it ‘dezinformatsiya.’ Our enemies in America call it ‘active measures’, and I, dear friends, call it ‘my favourite pastime’” (Boghardt, 2009, p. 1).

5. This can seem confusing at first. Let ‘completely consistent story’ = indication # 1. Let ‘contradicts story’ = indication # 2. A guilty suspect always contradicts a key part of his story. Hence, the interrogation procedure always produces indication # 2 when the suspect is guilty. However, sometimes, an innocent suspect contradicts a key part of his story. Hence the procedure can produce indication # 2 when the suspect is innocent (30% of the time in our example).

6. $P\left(E|H\right)$ is the probability of the evidence being observed given that the hypothesis is true. For the interrogation process, evidence in the form ‘indication # 2’ is always observed when the hypothesis is true. As such, $P\left(E|H\right)=1$. However, such evidence is also revealed when the hypothesis is false. This is taken into consideration by $P\left(E|\mathrm{\neg }H\right)$ in the denominator. That is, the probability of evidence in the form of ‘indication # 2’ when the hypothesis is false. This is expected to occur 30% of the time. $P\left(E|\mathrm{\neg }H\right)=\mathrm{0.30.}$

7. The symbol ‘$\mathrm{\neg }$’ is the negate symbol. $\mathrm{\neg }H$ means not $H$.

8. See Gale (1996), Smith and Sørensen (2000) and Çelen and Kariv (2004).

9. Bikhchandani et al. (1992) explain why just two decision-makers in a row are needed to start a cascade. We explain the process in this section.

10. While in this case, the example involves sneakers being adopted in a cascade, they could also be rejected in a cascade. The posterior probability goes straight to 1.00 in the sense that the hypothesis (e.g. the new sneakers are the epitome of style) is completely confirmed or straight to 0.00 in the sense that the hypothesis is rejected. Furthermore, either direction could be correct under different circumstances.

11. While this can be changed to any number, if there is always the possibility of being targeted by disinformation, that this possibility is 50–50 on any given day is a plausible place to start.

12. If $s=0.60$, then a signal ‘G’ will be correct 60% of the time and incorrect 40% of the time. This is analogous to the indications given in our interrogation example (above).

13. Given decision-maker 1 has, by his choice, signaled that he has private information ‘G’ and since all decision-makers believe before any evidence in the form of their private signals or observed choices of others that there is a 50–50 chance that the information is genuine, if decision-maker 2 has private information ‘D’, he believes there to be a 50–50 chance that the information is genuine. If he chooses to accept, then decision-maker 3 will also choose to accept because he can observe two accepts, leading him to believe that the chance that the information is genuine is better than 50–50.

14. The model can be generalized easily to a case where different decision-makers have different degrees of signal quality. Bikhchandani et al. (1992, p. 1002) assume that in such a case, decision-makers with higher signal quality choose first. These might be the more senior decision-makers in the organization. Their signal quality can then be inferred by subsequent decision-makers. Of course, if the signals very in accuracy but are still not perfect, the chance of a disinformation cascade being led by senior decision-makers remains a strong possibility.

15. Bikhchandani et al. (1992) also provide a generalized n-person case.

16. Remember, we are saying 0.00 or 1.00 because it depends on which way we look at it. Genuine intelligence can be correctly accepted or incorrectly rejected (and, conversely, disinformation can be incorrectly accepted or correctly rejected).

17. While the direction is incorrect, Bayes’ law has not been broken. As such, the decision-makers cannot be accused of being irrational.

Additional information

Notes on contributors

Peter J Phillips

Peter J Phillips is an Associate Professor (Finance) at the University of Southern Queensland in Australia.

Gabriela Pohl

Gabriela Pohl is a Lecturer in Social Science at the University of Southern Queensland in Australia. The authors specialize in the use of decision theory to help people make better decisions. They stress the importance of 'thinking about thinking' and strive to show how decision-makers can use behavioral economics to simplify decision tasks and identify the systematic patterns of behavior displayed by friend and foe alike. Their work applies decision theory to terrorism, counter-terrorism, national security, intelligence and espionage.