Networks of fear and silence

Mastering the psychology of network behavior is the secret of success (and money) for the digigiants like Meta, Google, Amazon, Twitter, TikTok and others. They all have developed services and algorithms that invite and inspire people to join and remain as their audience. The success is indirectly measured in terms of site visits, clicks, and transactions that take place within their network.

The theme of this blog is the silent network and even the network of fear that many of us have but which is seldom discussed. However, I start with a roundabout approach and discuss some of my earlier network psychological observations.

Active network

In our study on network behavior in 2002 we used the term active network to refer to the connections people activate, maintain and use in their daily life and work. (Marttiin et al., Proceedings of the 35th HICSS). We had followed network collaboration in software production work at an international communication technology firm and learned that project managers had problems in dealing with passive members of their network. This is easy to understand and occurs in today’s remote work arrangements as well. It simply means that some members remain silent or inactive when they see that activity would mean more work, new tasks or other forms of engagement they do not want to have. They can have different strategies to hide the passive attitude, like being active in communication but not in doing or taking responsibility.

One reason why this phenomenon has been overlooked is that it is difficult to measure and point out. A person’s overall network activity can be monitored over time, but is difficult to monitor its content in every interaction. In f-f meetings the organizational values, culture and “multiple eyes” guide our behavior. Passivity is easily seen then but in remote work and networks it can remain an invisible behavior. We did not use the term passive network then.

Early experiences from email traffic

I had early interest in the psychology of human communication, especially email behaviors. In 1988 (!), with my colleagues Kirsi Ahola and Markku Silen, we studied the email activity at a major international ict firm in Finland. The reason was that, in the firm  – where about 500 people worked –  the middle management had started to suffer from email overload. 

Quantitative measures of email traffic and activities between the communicating partners did reveal the objective situation, the scale of the problem. However, I soon realized that we need to understand what assumptions about the consequences of their communication behavior people have when they send emails in the way they do. It’s about the psychology of active and passive networks. 

So, in addition to the recorded objective email data, we asked the personnel to tell how many emails in a day they thought they could send to their managers –  when they felt it is necessary –  but without overloading them. Then we summarized this frequency data from each person in the group reporting to their manager by adding up the potential email frequencies to him or her. The data was not surprising since we could expect that people are not aware of the communication situation of their managers and colleagues.  They did not imagine that the total outcome could be a disaster. This was the early time of emails and messaging culture was just emerging. The problem lives with us today although alternative and better tools are continuously emerging.

What indeed surprised us was the scale of underestimation of how much mail the managers could receive per day. They underestimated it by a factor of 4 to 5, if I still remember the data right! There was an interesting psychology underlying this network dynamics. People had important topics on their minds and sent the emails thinking it did not cause any trouble. As a solution to this, we suggested a soft approach, that is, to build a new, reasonable communication psychology and culture. However, many seemed to think that automation and AI could take care of this. It did not. We even had the idea that the sender should receive a note “You are number N on the list of received emails”. This did not happen, though. At that time, the hierarchic position of managers could act as a hinder or even block communication channels, but when email was new and the organization tried to avoid hierarchies, crowding occurred.

Silent and passive networks of today

We have dozens, hundreds and even thousands of connections in our everyday social media and the mobile. There are methods to map these networks and activities and there is abundant knowledge about the negative effects of the social media. The problems relate typically to our active networks, while the passive networks have remained less well known. Of course, this is interesting to the strategists of the network giants who want to invite everyone to join their net, including the silent ones.

Systematic algorithms are used for amplifying the impact of negative comments and posts so that they receive more attention and spread more effectively than the positive ones. There are many, e.g., cultural, political, personal, situational, and spiritual reasons why so much negative communication emerge in network discussions, globally.  As far as I know there are no national exceptions to this unhealthy pattern. It is a global, human-social communication pathology.

Harmful communication in the net is known to cause bullying, eating disorders, and it can feed even wars, driven by these algorithmic pathologies, which  for example, Frances Haugen bravely took up at the hearing in October 2021 when she presented her case about Facebook. An excerpt of her statement: 

“The result has been a system that amplifies division, extremism, and polarization — and undermining societies around the world. In some cases, this dangerous online talk has led to actual violence that harms and even kills people.”

Haugen referred to the well-known, terrible risks, but they are only the tip of the negative network-iceberg. People have been forced to deal with less dramatic but still painful psychological experiences in the net. For some this has continued for years. Running away from it, is often not a real option since the giants of the cyberspace have their ways to lure as many of us as possible to react to their messages, to study the sites which their customers push for us to see and experience and to stay as their willing audience and users. Then there are the social pressures.

Activity maps show the connections between network nodes and reveal the activity levels between different nodes, which can be people, content providers, firms and any other organizations. Quantitatively this can be expressed as the frequency of contacts, various forms of messaging, transactions, and so on. What remains practically invisible in these maps are the connections that exist, but which the person managing his or her personal net has voluntarily kept silent.  In other words, the net hides a psychology of silence. Figure 1 shows an imaginary example of a Personal net which has been divided into two: Active and passive network. An example a real twitter network demonstrates its potential density.

Figure 1. Active and passive net. Psychological management of the personal net generates an active and a silent or passive net.

It is possible to follow the temporal developments in the connections between individuals or communities and observe any change there, like abrupt silence, which would inform of something significant happening there. Keyword analysis in the maps can even reveal the content, tone and style of communications like in the popular twitter maps, but in the long run the silent net can remain a mystery unless its psychological drivers of individual reactions are known. What makes this problematic is that silence has many different drivers and people are often proactive in their behaviors and especially when they want to avoid something unpleasant. 

What drives disconnect and silence?

The list of negative psychological symptoms in net behavior is exhausting: anxiety, depression, envy, low self-esteem, eating disorders, sleep problems, bullying, isolation, and stress. Some are able to avoid contacts and communication that generates these pathological experiences and symptoms, but many if not most of the young generation are unable to do so. I have not found studies about detailed network management practices that people use to avoid these problems or remain intact. The network giants do their best to keep their audiences.  Guidelines for healthy social media use are abundant, though, for example from NIH and other relevant organizations and professionals. They are especially targeted to children and their parents.

The tools that some social media platforms and apps suggest for preventing the worst negative consequences are curation, censorship, regulation, AI based detection, fact checking, blocking, limitation and so on. My educated guess is, however, that people start with more subtle strategies and social behavior-scrips of their own, already before something catastrophic has happened, just like we do in traditional face-to-face social situations. We try to avoid negative company, trouble makers, aggressive people, and anyone with questionable habits or a tendency to threaten our well-being and privacy. 

As a result, our personal nets include people/firms/orgs who are connected with us but whom we avoid since they have the psychological power to disturb and even hurt us. This prevents us from posting something or participating a discussion or a group or a circle because we anticipate negative, bullying, or otherwise disturbing and harmful comments and reactions to our messaging. If such negative experiences have occurred once or more often, they have made us careful in how to communicate with these sources of trouble, directly or indirectly. Some are strong and disciplined enough to act as gate keepers at their own sites and discussions and in this way help others. 

My hypothesis is that people use similar and even the same behavior scripts that they have adopted in “flesh life”. In the network environment it can mean direct behavior, like having low communication frequency with the potential problem sources, avoid reading their comments (the skill of not-reading) but it can also be an indirect behavior, to avoid and suppress any intention or plan to participate in communication threads where these problems will probably occur. These reactions are complex and probabilistic in nature and triggered by various conditional probabilities as well.

Social geographics of fear

One could ask that so what, is this somehow important and why don’t you just block them? On many platforms it is possible to manage our active networks and block or constrain the problem cases but it can be difficult, especially when involved in strong emotional interactions, social circles and when it is about people we know personally. Furthermore, the behaviors of the problem sources in the net and in flesh life can differ significantly and there are many other social complications like work and customer relations. Furthermore, many of us have learned to respect diversity and it is natural to give space for different opinions and world views. The borderline between diversity and pathology has become fuzzy in the net, however. I’m not sure if there are tools to do the blocking and at the same time shortly describe, from personal viewpoint, why that was necessary. Such data could give some indication on the psychology of these problematic communication phenomena and lead to better designs of these tools.

When we decide with whom to communicate and where to participate, we span and mold a personal social network. We can think of it as an open virtual terrain where there are potential, interesting trails we would choose to enjoy, but have learned or expect that there are danger zones – and we avoid them. My Facebook friend Laura Tiilikainen commented in one of our discussions that this reminds of the Geographies of Fear, the concept which describes how especially certain urban spaces evoke fear of crime and which was applied for cities and especially from female perspective (e.g. H. Pain, Social Geographies of Women’s Fear of Crime. Transactions of the Institute of British Geographers , Vol. 22, No. 2 (1997), pp. 231-244. A natural follower of this is the concept of psychological safety and the notion of safe places. Their psychological meaning for individual well-being is now well recognized and similar requirements should concern network environments as well. There is an acute design challenge to produce and maintain such spaces in the net. When negative consequences occur, it is already too late.

Silencing of the net 

We can look at the “geography of fear” of the net from the negative psychological perspective. An unpleasant net, for an individual, can emerge as a result of negative or threatening communication or a network cluster of people that dominates the content and style and uses aggressive or unpleasant style to talk to the discussants outside this cluster. It can consist of firms, other groups or communities, activists, even political parties and their supporters, having a specific “brand” but from which the person assumes she/he receives unwanted, harmful or uncomfortable communication as a response of her/his communication. Over time a network geography of fear develops and it becomes the net of silence.

Above I wanted to point out that behind the active network we and our network friends can have a silent net, which can be characterized as the social network of fear. It consists of connections that have not been blocked, for one reason or another, and which by their mere existence make the silent network possible. It is possible to collect relevant information about these psychologies by looking at data about blocking behaviors. However, when blocking has not happened, the psychological impact remains powerful.

When physically meeting people, we can experience these negative psychological effects every now and then but we have learned to avoid and cope with them. Our ways of living and socializing aims at a personal balance and as part of personal growth we learn ways to manage it. The net is different: it has given permission for almost anyone in our net to have access to our minds, souls and life situations, at any time and always. It opens the gates to dangerous forces, which require a treatment of its own. It has become a serious design and business model challenge.

On well-behaving AI, now with the GPT

This is again very speculative, and I’m by no means a pro on Large Language Models (LLM) but I wanted to take up a peculiar possibility that LLM/GPT/Dall-E -type, learning-based systems could be designed to generate specific forms and styles of behavior, just like GPT and Dall-E do today with texts and images.   They could produce any kind of behaviors for video, animations and robots, especially good behaviors. Hence, I believe this is worth considering.

On the 5th of February, 2018 I published a blog Teaching manners to AI: Internet of good Behavior. GPT was still being developed and GPT-3 had not been released. Perhaps there was something in the air since I wrote “…we would let it use all the relevant data related to that behavior and to learn from it”. Here is the excerpt: 

“What if there was a systematic way to offer models of good behavior for AI to follow, to teach it behaviors we know and define as good behavior? In many cases it would be easy to define the criteria and to use such behaviors as models for AI to follow and learn. With the Internet of good Behaviors (IogB) approach we could offer AI access to behaviors (and companionships) we think are good for its development just like we do to our children. By allowing this we would let it use all the relevant data related to that behavior and to learn from it. It is quite possible we could learn from that too, but that’s another matter.”

Of course, there was no explicit method to accomplish this then, but now we almost have it. Having seen what LLM:s can and will do I wanted to return to this topic, but now knowing the potential of the GPT, Dall-E  and others. I have shortly touched this topic in my book Internet of Behaviors (IoB) – With a human touch.

Coding behaviors is nothing new

IoB is a system for coding mental and physical behaviors and their detailed or larger components. It can be physical, artistic or of any abstract form. Incidentally, when I have introduced the IoB, its “mental side” has often been ignored, although it has significant and novel value in IoB.

Codes for certain classes of behaviors can be defined in code books, just like it has been accomplished for musical notations. They have evolved for centuries and are now standard and in global use. When a musician plays according to the notes, we don’t usually consider it as a behavior which obeys a code that has been invented, stored and shared. Originally, it is the composer who has coded her or his “playing behavior” (or imagined a player with certain instrument) and written the behavior codes down for others to imitate it. Lento, Adagio, ppp, fp etc. are, ideed,  behavior codes.

A violinist playing a composition by Sibelius, for example, repeats the behaviors of Sibelius or what he imagined for a violinist. In front of a symphony orchestra, the conductor makes sure the orchestra follows the codes, perhaps with slightly modified interpretations. In other words, by systematic coding of musical behaviors, extremely complex, even creative behaviors can be coded, stored and then generated, by humans or machines.

It is possible to build behavior code systems for any behavior. Historically, we know that at around 1677, Louis XIV commissioned Pierre Beauchamp to create a notation for baroque dance so that it can be “put on paper”. The emperor wanted to preserve and perhaps share these dances outside France and for the generations to come. (Karl-Johnson, G. (2017). Signs and Society, vol. 5, no. 2.)

Tracking behaviors

Coding of visual behavior material like it is seen videos and movies is not easy. Manual video coding has required massive work, but now AI-based, deep-learning systems for this have been evolving fast. Clever multiple object tracking (MOT) methods have emerged for video and their overall performance is improving. Some have considered the potential of the general Transformer. It aims at modelling spatio-temporal interactions of objects and can be considered as a candidate for modelling behavior sequences and interactions among humans in video materials.

I cannot estimate the future potential of these rather complex tracking methods. Nevertheless, they seem like a promising possibility to extract both visual and textual data from movies and videos and to use this as training material in machine learning. Interestingly, in countries like Finland, where subtitles are used, movies and other filmed or televised materials carry synchronous visual and textual information about the presented scenes.  Then there are the manuscripts, and spoken language which could also be used as sources for texts that are directly related to the visual material and to use these as training material. There are many complexities in this and it is a demanding challenge to use such materials, but it would be possible to demonstrate.

Some speculation

I hope to present the following idea clearly enough and why not, for improvement and perhaps even testing. My educated guess is that something like that is already happening, but I have no knowledge of that. Below is my speculative outline for such an approach.

The idea is to use a pretrained generative network which has been trained with extensive textual and visual materials (movies, videos and other) and so that its “attention” mechanism is focused on the visual and textual material in parallel, synchrony and with suitable temporal windows. The aim is to teach the system so that it connects the verbal expressions with related, visually shown behaviors. It learns to predicts the scenes and texts following a prompt scene and/or its texts. With a GPT-like arrangements we could then expect the system to react to textual, visual or textual+visual prompts by generating textual, visual or textual+visual expressions of behaviors.

I have presented a very general outline for the system in Figure 1 as I think about it. 

Figure 1. Outline for a visual-textual transformer system

For curiosity, I tested how ChatGPT generates instructions for good behaviors. If we had a code book for good behaviors in a specific context, and a way to generate these behaviors, then the texts by ChatGPT that produce guidance for “good behavior” could be fed to a visual behavior generator and observe the outcome as a video, or perhaps as a static image. Different versions can be imagined how to accomplish this. 

ChatGPT suggesting good behavior

Here I have demonstrated how the ChatGPT performs when asked to generate a manuscript for a good behavior in a simple human scene. It is easy to imagine that if the system had been trained with associated visual and textual data, its response could be in visual form as well.  It would rely on a “score book” for certain behaviors.

EXAMPLE. I presented a simple prompt to ChatGPT and received a good manuscript for a simple behavior episode. I did not (yet) test it, but Dall-E could be used for generating the imagery for this.

GN: 

Write a short manuscript of a scene where an American man, of age 50, comes to a Japanese home for the first time. He does not know how to behave. Please describe what would be good behavior in this situation?

GPT-3: 

John, a 50-year-old American man, was feeling nervous as he walked up to the front door of his Japanese friend’s home. He had never been to Japan before and was worried about making a mistake and offending his hosts.

As he entered the home, he noticed that everyone had taken their shoes off and placed them neatly by the door. John quickly followed suit and took off his shoes, placing them next to the others.

He was then greeted by his Japanese hosts with a bow. John wasn’t sure what to do, but he tried to imitate their actions as best he could. He lowered his head and gave a slight bow in return.

The hosts then offered John a seat on a cushion on the floor, which he accepted gratefully. John tried to sit with his legs crossed, as he had seen in movies, but found it uncomfortable. He shifted his position several times, worried that he was being rude or disrespectful.

As they began to serve tea and snacks, John watched carefully to see how his hosts behaved. He noticed that they always used both hands when passing objects, and that they never poured their own tea.

Taking these cues, John tried to be as respectful and polite as possible. He always used both hands when passing things, and made sure to pour tea for his hosts before pouring his own.

Despite his initial nervousness, John began to relax as …

Is there any sense in coding behaviors?

The idea behind IoB is to code behaviors with relevant resolution, depending on the situation and context, and not necessarily for every detail. No explicit IoB systems exists as far as I know although in computer game environments and in animation, coding of behaviors has been used in various forms for quite some time.

One possibility for testing the generation of behaviors with the IoB approach is to start from top down. That would mean the use of behavior codes that cover high-level, perhaps complex behavioral components (e.g. a video of a man approaching a Japanese house) in a specific situation/context.  In other words, codes for the walking style of a man would not be used unless it has information value and it is something to be controlled. The output behavior codes could then be fed to a software that generates virtual characters with that behavior. Dozens of such tools are already on the market, but evidently and so far, having a compatibility issue.

What use could this be? If the ambition level is as high as it is now with recent AI developments one can imagine various meaningful uses where certain type of behaviors can be generated, for any purpose from education, entertainment, work and art. And then, returning to the question of well-behaving AI, this could offer one way towards generated, positive or ethically sustainable AI behaviors.

Creative teachings of the nightly jet lag

I have never really studied the psychology of jet lag, so this is a highly personal view and experience.

A week and half ago, I made a one-week trip from Helsinki, Finland, to San Francisco and Palo Alto. Coming back home I’m experiencing an unusual jet lag, probably because of my working schedule, which has been very open here in Helsinki and I’ve even enjoyed a nap every now and then, at noon and in the afternoon. But his is not about the problems of jet lag but about its creative benefits I have (again) met.

What is peculiar about this kind of jet lag, where I stay awake from about 2AM to 5 AM? I have gone to bed at around my usual time, 10-11 PM, and wake up after a 3-4 hours’ sleep. This means that I have then had a reasonable, although short rest. In this sense being awake due to jet lag is different from insomnia that is caused by other, perhaps disturbing experiences and stress.

In my case I stay awake, trying to sleep (without success) and not to disturb my wife with restless moving. This means that I don’t have any strong sensory experiences since it is dark and quiet where we live. This comes close to sensory deprivation situations where all sensory stimuli are minimized.

After trying to sleep, and failing in it, for about half an hour I’m fully awake, but I still don’t move too much, see or hear anything that would lure my attention. This state of “trying to sleep” has its pauses and it somehow opens the gates to peculiar mind processes to find their chance to activate.

Then my mind starts to work vigorously. Some might say that it is “the brain” but my take is that it is mind that matters now and how we understand and conceptualize it.

Experiences, ideas and problems enter my consciousness in a rather haphazard order and it is not possible to predict what will pop up next. Most of these thoughts and images are not of the famous idea genre where in the morning – if you have written them down – make no sense. Quite the contrary, they make a lot of sense and interesting pieces of a puzzle can sometimes find their places. For example, this is the first time for me to think about jet lag like this and it came to my mind at 3.30 AM, about. It took some time for me to decide to get to my Mac and write this.

Examples

The impulse to write this blog about the creative jet lag ideas – it’s 5.11 AM now – was one of many. Another one was about a series of experiences I have had since mid 1970’s when I learned about artificial neural networks at Professor Teuvo Kohonen’s lab and the Finnish Pattern Recognition Society. My first insights came from the work on associative memory with artificial neurons that Kohonen developed together with Professor Juhani Hyvärinen, a neurophysiologist, and Mr. Pekka Lehtiö, specialized in psychology of learning. I worked at Hyvärinen’s lab and together with Pekka we developed microprocessors and digital systems for our vision lab and elsewhere.  These connections allowed me to follow these the development closely and sometimes participating. It is a creative-integrative process to start building this picture on my mind at 3AM. This is the first time I formulate it like that.

What I learned then and during a decade, have prepared me so that Deep Learning (DL) models and GPT:s have not  surprised me at all instead I have seen them as natural follow-ups of something that I saw happening over the years. I will probably collect these experiences and write the story. It will start from Perceptron and perhaps the visual system architecture. The point of insight  I realized during my jet lag moment is, how good and clever basic research can inform and prepare us for the future, which we cannot predict but which we will understand when we face it.

What is so special about jet lag?

It is preceded by rest and occurs during a bodily state where we would usually be awake. Normally, without jet lag, if we decide to wake up at 2 AM, for example, we are tired, perhaps exhausted, nervous, and our sleep mechanisms try to take over. The creative bursts of the mind can still be activated but they will soon be damped and defeated by sleep. Mental organizations in these two situations are totally different. 

Can we arrange for creative bursts to occur?

This time I imagined that perhaps, in our everyday life, we could have or allow the same creative moments and bursts of mind if we can arrange a similar psychological situation as we experience during jet lag.  I will not go to meditation techniques here, which, of course, have a significant role in this. The critical conditions for the creative bursts to have their chances, are the following: 

• You have a good mental balance.
• You are not exhausted or tired.
• There is no excessive sensory stimulation, like lights, sounds, touching or motion.
• You spend in this quiet state long enough (!) so that it starts feeling slightly disturbing.
• You don’t fight against this forced “passivity”.
• You let it happen.

When these conditions are fulfilled, the mind finds its opportunity and fills the consciousness with whatever it sees possible, not necessary the best content, but we can control that and stop the session if it feels dangerous or uncomfortable in a way we don’t accept.

Then there are the usual everyday obstacles to this: interruptions, ongoing activities, active mind solving problems and preparing for something, and lots of environmental stimulation. The heavy responsibility. Finishing this text now, I started thinking that perhaps this feeling of responsibility (to do my work and stuff) while still maintaining the quiet posture and state of mind, could boost the creative mind just like the feeling of forced rest does in the state of jet lag. I don’t know.

It’s 6.06 AM now and I can perhaps get a quick sleep before waking up at 7 AM.