Sander van Splunter: Trusting to find Sense

Sander van Splunter is a researcher at the Intelligent Interactive Distributed Systems (IIDS) (www.iids.org) at VU University Amsterdam, and at D-CIS lab in Delft.

Sander van Splunter has been investigating how complex software can adapt by itself. One of the issues he focussed on was whether a local small change has an effect on the whole system. His research has been applied to crisis management and is currently used in research for the energy market.

Having worked for many years behind the computer, van Splunter notices that he is still trying to make a balance between his digital working environment and other ways of communication. Most adaptation to the digital world happens through your own working environment, which makes it even more important to design this well: not too intrusive and efficient in using the right medium for the right task. Sometimes paper and pen are more handy, or a conversation, to structure thinking. Nevertheless, for van Splunter the first impulse is often to turn to the screen.

Time is an issue because of a-synchronicity. In distributing programs time is an issue because the same time cannot be tracked on different locations. For computers, of course, time is not as influential as for humans, as in day- and night rhythm.

Computers are (almost) always available, whereas people are not. For humans there is a discrepancy between concept and perception of time, also when people use computers. A few seconds waiting for opening up a program can be perceived as very long. Computers do not have this discrepancy in perception. Time functions in computers on much smaller scale as they are much faster at what they do then human beings can be. Programs and networks of computer do have time response issues though. Response in time is one of the things on which you assume to be connected or assume another program is ‘alive’.

There are time-based protocols to regulate interactions between computers. These protocols can be considered as rhythm based for regular transactions and if something disrupts the rhythm then either something new happened or something went wrong.

If you want to know whether another computers still lives, you want to receive a message every once and a while to say ‘I’m alive, I’m alive’, as long as it functions correctly, you’re in that rhythm. When you are out that rhythm and you don’t get a message back in time then problems arise. Though this rhythm functions for interactions between machines, the question is whether the rhythm of the machine connects to the rhythm of an individual human being, van Splunter suggests. When you look at larger scale rhythms of use on the Internet of news-sites for example, one can clearly see a rhythm in this, but such rhythms connect more to groups and less to an individual user. In current research van Splunter is investigating how personal patterns of use influence these larger scale rhythms, and how this can be influenced to make better use of local energy sources. So if a whole street turns on the coffee machine at 8 o’clock, van Splunter develops scenario’s where the machines make coffee between 7.30 and 8 o’clock and cut the peak load of energy use for example.

When discussing the example of the tabla player, who could change any delay into a different beat, van Splunter emphasizes the fact that the tabla player can hear the rhythm, where computers do not have such sensorial capabilities unless they are told to have it.

If you have an experienced tabla player, who is used to improvising, he listens to the rhythm and adapts his interactions. If you have a group, also as a group they have their interactions. You need some kind of roles in such improvisation or adaptation processes. If everybody starts adapting at the same time, maybe something new is created, but it also might be too much change at the same time ruining the music, argues van Splunter.

The problem of synchronizing rhythms between human beings and machines is even deeper, according to van Splunter. Human beings can relate to a tabla player, a human tabla player, by assuming that his experience of the world is similar to theirs. “We can hear his music and even though we might not understand the rhythm as good as he does, we have a sense of it. To create this effect with computers we have to create their sensory environment and which is really difficult. Talking to the tabla player, you can explain at least to some level, how he experiences, how he feels. You can say, he has an aggressive rhythm. For a computer, something like ‘aggressive’ is hard to relate to.” Van Splunter argues that the “body” of a computer consists of hardware like a CPU and a hard disk. But, argues Nevejan, the tabla player consists of 70% water so maybe we could argue that because of their capabilities computers have gone into a different realm of beings. Van Splunter agrees this is a nice philosophical statement, but argues that for human beings it is hard to know what that entity is for a computer. It is hard to know how much that entity is formed by our own projections and attributions, compared to what this entity actually is. This makes it hard for humans and computers to relate.

The place of a computer could be defined by the locality of its inputs (keyboard, mouse and monitor). However, in a network environment the place of a computer is less evident. Computers on the other end of the world and even onwards can be accessed although one does get experiences of delay.

So place is still a factor of distinction for computers. On the other hand, the availability of computers affect many places. Another problem, van Splunter points out, is the local availability of data. Google has these large containers with a lot of computers in it, which need to be located together because there is a lot of parallel processing going on at the same time and they have to go through a whole lot of data, which is present at that location. If you want to have the same interactions somewhere else, you have to send all this data over the network, which is inefficient. So the availability of data is also an aspect of location. With privacy and sensitive information location is even more important.

In distributed environments the issue of trust is significant. Therefore a distinction is made between trusted and non-trusted environments. In these environments the kind of interactions differ based on the level of trust of the hosts.

A simple approach is a list of trusted locations for servers and hosts. Next to this there are trusted third parties, like McAfee, who flag other sites as ‘evil site’ or not. In peer-to-peer environments, where computers communicate without a regular structure, an experience-based trust is used. For humans this works quite well. “If I have a lot of good interactions with you, it is an investment of your time. If you do that just to frame me, at least it will cost a significant investment of work for you. For computers, experience-based trust is more challenging. A number of good interactions is not necessarily a significant investment for a computer. You can send a few files and then turn evil.”
A different aspect of trust according to van Splunter is that a basic level of trust is just needed. We are forced to trust the few top domain name servers, for example, most of which are American hands. We need to trust that the power to redirect all sorts of router information to other parties is not misused. Computer environments remain dependent on humans providing the infrastructure of communication.

Values within the design of distributed systems are a push and pull between transparency, traceability, security and privacy. Human values are very much bound to location oriented, says van Splunter. If they build something, it is on a specific location. If you design something on a computer and you do it in the right format, you can send it off to others. So values and trust in a human environment are different than in a computer environment.

Human cultural values are bound to locations and easily in the global network environment, overgrowing national boundaries, clashes happen on issues of privacy, data retention, control and monitoring. From a law perspective it is a challenging and interesting environment. Van Splunter argues that values of systems are very much defined by the political and economical cultures within which they function. On the other hand, individuals have access to the whole network and can get around some parts of these political and economic structures.

When designing complex systems, keeping the system understandable is a value as well. Otherwise you cannot use or maintain it. From an adaptation perspective, simplicity may be a value, according to van Splunter. You want to understand the relation of a program or small subset to its environment, to understand the effect of an adaptation, which is difficult to grab. So simplicity might help to abstract the main points for which that part is designed and to identify the basic aspects that need to be retained. An issue of adaptation is to mapping between human values (what behaviour would they like) and technical values (what are the technical constraints).

Human beings can only witness the system by the grace of the designers: the error-logs and pop-up screens saying ‘something went wrong, sorry’. Observations of the system need to be transformed into something the human can understand in either files or, as in our case, messages on a screen.

The change of the capabilities of computers is another factor. Now it is easy to get email via a mobile phone and therefore interact with and observe machines. Before this was only possible at work or at home. And, asks van Splunter, how acceptable is it to turn off your mobile phone or not checking your email? In some environments it might not be acceptable. There can be a need to witness.

On the other hand witnessing might be fun. When a system keeps adapting in a dynamic way, this might be interesting to witness even though looking at it might not give new or useful information. It is just the observation of the dynamics, which is fun. If there is an interesting rhythm to observe, then it becomes fun to observe it, even more so when you can also have some interaction affecting this rhythm.

Using language in a computer environment is one of the ways to communicate with the computer itself. Instructions are a natural way to communicate with the computer. However, there is not that much interaction; a human user expects the instructions to be carried out.

The computer is assumed to be not too autonomous. Words are assumed to be deeds for the computer. It is a command-based level of interaction.

In agent-based system, there might be more distinction between words and deeds. Such more autonomous programs can have beliefs, desires and intentions, which can be stated for a program; what view it would have and what it thinks the world is like. When writing software programs, one defines what kind of actions are related to what kind of words. It would be nice to be able to make the distinction between small deeds, big deeds and large deeds, as Nevejan suggested, but it is not regular.

If you try to create a concept in research, you try to capture all your ideas in words on paper to give it to others. The same can be said for programming. You want your ideas on how things interact in a program for others to further interact with them. The root of creating software is engineering. But, argues van Splunter, if you are constructing it is not just the mechanical act of construction. It is also the interaction with the material and to try to create some ideas with it. You always have a creative layer on top of it. And that might come close to linguistic aspects of communicating ideas.

When asked about the story of the potter who gets centred by the wheel, who creates a centred pot, which centres the woman and creates a centred community, van Splunter points out that this is something we are still trying to reach.

If you write emergent computer programs with emergent behaviour, which create a rhythm by itself and you also have the human society with a rhythm, you would like to make this centring happen because then those two environments can come together. The centre in this case would not be the point where it comes together, but the point where coming together makes sense, as van Splunter formulates it. Basically it requires a bit of understanding of both sides. On the one hand one needs to know what the needs of society are. On the other hand one needs to know how the interaction of systems and previous systems interacts with the needs of society. Another challenge is to create some kind of rhythm in there. This is something that can be difficult to model. One aspect is a control problem on the systems side.

Humans have relations on many levels and the same goes for computers. But if systems want to reason about relations, the system has to have the right specifications.

We can think about relations for a system, but the system needs to make them explicit. It is difficult for a system to think of relations themselves, as this is meta-reasoning. If we can define some kind of function for a relation, a system is able to assess a relation. For example if you send a message and it is send it back directly, you have a response time. Because of the response time you can have an assumed distance, which is related to location. You can say this machine is close to me and this machine is far from me. Computers require functions to assess relations.

In a computer and system environment you can smash everything down to relations. Indeed, you can program statements, have some kind of syntactic structure, so basically you have a relation of terms. And instructions are coupled to specific hardware instructions. From that perspective, it would be all relation based. Indeed. But where does that bring us, asks van Splunter and he explains that we can understand and reason about the level of syntax in relationships. But if you want to abstract from that and have it perform tasks by itself, it also needs to understand these tasks, to add meaning to the syntax. To be able to understand a task it needs to understand how the relations that are needed to perform a task, relate to itself and/or its own syntax.

Performance is always based on and limited to a specific application of systems as it is for humans. If you are searching for a certain book than the result is more important than the response time.

There is no inner quality to performance for computers, as there is no inner value of performance for humans beings as well, argues van Splunter. When looking at a piece of art, a human can say ‘I have a feeling with it’, but still it is bound to a person and taste differs for example. How does a computer environment get a sense for something is good or worse? Basically an assessment of performance is always bound to a specific instance. You can make local assessment of performance. But if you look at it from a different perspective (e.g. in relation to a group) then other attributes might be more important. And how do you abstract a general idea from that, asks van Splunter, and what would be the opinion of the overall system? Nevejan suggests that children perform for fun and witnessing can be fun van Splunter argued, so maybe performance may also be an inner drive? Van Splunter replies that if you go to emergent systems, this kind of drive could be key for modelling it.

View full transcript including film fragments here

Hereunder the transcript in text.