Looking at social media

Some thoughts on simulating social media from Tim Price, international man of mystery.

---

I recently wrote a review of Hostage Negotiator by A.J. Porfirio from Ran Ryder Games and speculated that it might be a useful point to start a discussion on a possible model for social media influence.

There is an understandable amount of interest in government and the military about the effect of social media, and a number of large multi-national defence companies that offer expensive products that propose to use “AI and machine learning” to generate synthetic social medial feeds to allow training to take place.

The only demo that I have seen involved chat-bots degenerating into the inevitable repetitious trolling from a bank of comments scraped from Russian fake social media accounts. It was utterly unconvincing and, when I put the question to Paul Rimmer, the Deputy Director of the UK’s Defence Intelligence organisation at a recent talk he gave in Oxford, he chuckled and said that if “I had £1 for every company that said AI and machine learning was going to solve my problems, I’d be very rich and wouldn’t be talking to you now…”.

My personal view is that if we can’t get Alexa, Siri or Cortana to lose their temper, then attempting to mimic social media through AI is a waste of time – after all, it is the manipulation of emotion and deeply held beliefs that create the effects people are trying to achieve.

Current training consists of scripted “social media” injects into events that are either trivial “box-ticking” exercises or at best short-term interventions that are deliberately limited in their effects so as to avoid upsetting things too much. It is necessary to have a proper discussion about the subject and avoid the rush to buy “shiny toys” simply because they have the words “social media, AI and machine learning” in the advertising brochure. There is a real danger that “social media” is the new “cyber” in the short attention span of some senior officers…

The quote from F.W. Lanchester is appropriate here ” “Simple models that provide useful insights are often to be preferred to models that get so close to the real world that the mysteries they intend to unravel are repeated in the model and remain mysteries.” (The Lanchester Legacy, Volume III, Chapter 9). Even if AI and machine learning were able to replicate social media feeds, their inability to explain why actions succeed or fail mean that they are still unlikely to be useful for training.

We should construct a simple manual model for social media, that can provide a relevant degree of help or harm to a “normal” military training game.  Preferably something stand-alone and scenario agnostic, that can be adapted to the circumstances.

There are some fundamental questions that we need to answer:

• What measurable effects does social media have? (Brexit? Trump? ALS Research?)
• What actions cause effects in social media? (Russian Bots? Policy speeches? Cat Memes?)
• What actions can the Players take to affect social media? (Hire Cambridge Analytica? Bribe the Russians? Both?)

Hostage Negotiator, at it’s core, is a game about your interactions with a hostage taker. If things go badly, he gets more angry until he starts killing hostages, and if things go well he becomes more relaxed until he starts releasing hostages unharmed. As a player you have different gambits you can try, each with their own risk/reward outcomes and probabilities. You seek to calm the hostage taker down and build trust, enabling you to get access to more advanced gambits giving you a wider choice of actions. Each turn there are cards representing the passage of time and the effect this has on the hostage taker (usually making him more frustrated and angrier).

Assuming we take the track from Hostage Negotiator to represent the aggregated sentiment analysis directed to the players organisation and actions, we can look at a “sentiment tracker” below.

The assumptions are (for the moment) that sentiment in the green areas represent general support for the player (providing some advantage such as intelligence, warning of attack, donations to the cause, etc); while sentiment in the red areas represent the opposite (providing disadvantages for the player such as false reports, no warnings, and support for opposition groups). When the marker in in the “S” zone this represents significant Support for the player and in the “T” zone represents support for protests, violent clashes and Terrorism against the player, both of which would translate into concrete independent actions in the real world.

In addition, it is assumed that as the situation reaches the extremes of the tracker, it become correspondingly easier or harder to influence sentiment (represented by additional dice in the green zone and fewer dice in the red zone).

The “Political Freedom Tracker” represents the degree of political support within the player’s organisation, so higher levels will provide access to more extreme (and possibly riskier) options expressed in the Social Media Action Cards.

The game starts with some negative event that the player is seeking to respond to. If they do nothing, the situation is likely to get worse (the equivalent of the “Terror” cards in the Hostage Negotiator game). The simplest way of doing this is by a “Sentiment Roll”, where they roll 1D6 and if the score is equal to or higher than the current sentiment, it gets worse by a point.

The player response can consist of one of several options, from the usual platitudes (“thoughts and prayers”) through stronger alternatives, such as “heartfelt denouncement by the Head of State”, to concrete action such as “legislative changes” or “direct action” by the police and military. Each option has a risk/reward matrix with high scores being beneficial and low scores representing failure, public anger, and a cost in “political freedom” to carry out.

The dice mechanism from Hostage Negotiator is to have a score of 5 or 6 on each dice as a success, a 3 or less being nothing, and a score of 4 requiring an investment of additional resources (reducing the players options in later turns). So, a roll of a 5 and 6 on 2D6 represent two successes; a roll of a 3 and a 6, one success and a roll or a 1 and a 3 represents failure.

An example card is: “Thoughts and Prayers”, where two successes gains the player +2 Political Freedom points (perhaps the speech was heartfelt and actual tears were caught on camera), a single success +1, and a failure triggers an additional Sentiment Roll because of the backlash. The card itself costs 0 Political Freedom Points.

Other example of 0 cost cards would be “internal enquiry”, “policy speech” or “divert attention elsewhere”. A “public enquiry” or “appoint a Task Force”, might have a cost of 1 or 2 Political Freedom points, but something such as a “Constitutional Amendment” would need at least 8 points.

Of course, the actions have to be appropriate to the level in which the game is set, be it nation state, military operation or local village council elections.

The player would start with a hand of 0 Political Freedom cost Social Media Action Cards and would attempt to defuse the situation with the usual political actions and, at the same time, build up support for riskier actions with a much greater direct effect if they succeed.

So – what next? This topic needs a wider discussion in a much broader audience than I currently have access to. Even a short examination of the subject reveals that this is difficult to model (which comes as no surprise whatsoever), so it will require a determined effort and imagination. Any suggestions would be very gratefully received.

Tim Price