This site is operated by a business or businesses owned by Informa PLC and all copyright resides with them. Informa PLC's registered office is 5 Howick Place, London SW1P 1WG. Registered in England and Wales. Number 8860726.
starts with a theoretical case for what we call activity-orientated structure,
arguing that activities are an important part of human life, and one that
will become increasingly important in future simulations including human-like
agents in believable contexts. Second, since there are so many different
activities, it is important they can be added easily to the system and
realized in the game world as actual (albeit non-physical) objects, rather
than (say) abstractions in the minds of agents.
Implementational issues are also explored, focusing on the way activities should be added to the system. We argue that activities should be defined in a high-level language, which is compiled into C++. An example piece of activity-content, defined in this high-level language, is presented.
The Case for Activity-Orientated Structure
Activities are Important in the Simulation of Agents
Since this article is concerned almost exclusively with the notion of 'activities', it seems prudent to begin by trying to explain why we believe this concept merits such attention. In particular, we shall seek to establish successively that:
activities are part of human life - agents who do not understand them
can appear dumb. This minimal claim is most easily illustrated by
considering a series of examples where an agent's lack of 'understanding'
of activities and their relations can be cited as a failing of that agent.
Activities add colour to our lives, lives which are otherwise coarse and
materialist. From a certain perspective, our lives can seem empty.
If all that there is in the world is other objects and agents, what is
there for us to do except manipulate those objects and agents? All that
we can want is to acquire as many objects as possible, and have as much
influence over other people as possible.
This coarse and materialistic view of human nature is based on the assumption that all that exists is other objects and agents. Getting away from this dark picture involves admitting the existence of a variety of social activities. These activities elevate us because they give us new things to want. The lives of the Black and White creatures and the Sims characters are unquestionably materialistic because their desires are materialistic.
Maximally strong claim: participation in sophisticated social activities is the critical property which distinguishes people from animals. Heidegger distinguished between merely existing, as an animal exists, and full-fledged human existence. (Heidegger called this "Dasein" or "being-there", and characterised it as participation in the social flux of the world). This insight has filtered through into popular culture: a well-known mobile phone company has mooted that "life is made of one to ones". Even Swedish popsters Abba echo the sentiment - "Without a song or a dance what are we?" - in other words, it is our participation in forms of life that makes us uniquely human.
The sophisticated skills we prize about ourselves, our ability to reflect, communicate, care about others, are dependent on our ability to participate in various social activities. Social activities, rather than just being one of the many things the mind thinks about, are actually the things which make sophisticated thought possible. It is because we can participate in sophisticated social activities that we can have the sophisticated thoughts that we prize. Philosophers have given many examples of aspects of sophisticated personhood which are conditional on participating in social activities.
Are Varied and Must be Easily Addable
Here are a few paradigmatic examples of activities.
of these activities, like a meal, are short-lived, but others are extremely
long-term: a family lasts as long as there is anyone in it. Some activities
can only exist within the context of a parent activity - a Goodbye activity,
for example, cannot exist on its own. These activities which must have
parents are called sub-activities.
These activities group people together, and these groupings can overlap:
Activity groupings can overlap..
In this dramatically charged example of activity-overlap, Arthur and Belinda are married. Arthur and Charlie are friends, but Belinda and Charlie are flirting.
Since there are so many different sorts of activity, we shouldn't think in terms of "adding activities to our simulation" being a one-off operation. Rather, once our simulation has become activity-enabled, activities have become a kind of content. When building a simulation of a society of agents, there are many different levels at which we should easily be able to add new content - there is no magic formula or technique to generate plausible agents, there is just lots of different behaviour to simulate. Each bit of behaviour needs to be easily-addable; behaviour needs to be addable at lots of different levels (e.g. objects, animations, actions, and goals). If our simulations are to address activities, as we have argued they must, then activities become one of these levels. The sophistication of a simulation is a function of the level at which we can easily add content.
Activities Exist Inside or Outside the Minds of the Agents?
We have presented arguments aiming to illustrate that 'activities', in a broad sense, are something which must be present in realistic simulations of people. Further, we have argued that new activities should be easily addable. We have been deliberately vague, thus far, in precisely how this importance is to be addressed, and in what sense they should be 'included'. It is our claim that the best way to include them is for there to be actual 'activity-things', non-physical entities with their own state and internal logic to them, existing in the game world. Based on their state, they influence the behaviour of the agents within them, and based on what happens, they change their state. Vitally importantly, an 'activity-thing' does not command its participants to obey - it just requests that they perform an action, and tells them the consequences of ignoring this request. The agents themselves decide which requests to listen-to, based on their evaluation of the consequences of performing the various options.
There is one superficially impressive argument against this model, viz the fact that, in the real world, it seems that activities do not exist, at least in this sense. If they exist at all, they exist in our minds alone. "Ought we not, therefore, simply build the agent minds in an appropriate way, and hope that they 'discover' or 'learn' activities?"
The activities-really-exist approach has significant advantages over the all-in-agents'-minds one. Firstly, it is vastly simpler. Second, it is much less wasteful of memory - the state of activities would, in a simulation based on the second approach, have to be stored multiple times, once for each agent. Third, it is much clearer how newly created types of activity (the creation of which, we have argued, must be as simple as possible) will fit into the activities-really-exist approach - they just create new kinds of activity objects in the world. Under both approaches, the agents minds ultimately make the choices over which actions they want to take - activities are only a guide.
Having dismissed the in-agents-minds approach to activities, no other natural contenders, apart from the one we have suggested, present themselves. It has the virtue of straightforwardness, and of being (to some extent) tried-and tested: in Black and White,. towns, reactions and dances were all implemented as 'group minds'.
Now that we have explained why we think there should be 'activity-things' in the game-world, let us drop the shudder quotes and silly name, and call them simply group activities, or (where no confusion with the naive meaning of the word arises) simply activities.
"Are You Not Really a Behaviorist in Disguise?"
There is a worry that by focussing on the social activity, and allowing activities to influence how agents behave, we are taking something away from the individual agents - where is agent autonomy in this world of overlapping activities which jointly determine behaviour? Are we not really behaviourists in disguise, who are taking the cleverness out of the agent, and putting it into the social activities instead?
Remember that an agent does not blindly follow the commands of an activity. Instead, an activity requests that an agent does something (and communicates to him the consequences of failing to follow this request). An agent who is in many overlapping activities at once will have a number of requests pending at any time, and must choose between them. He decides which to follow by looking at the consequences of accepting or rejecting each request, and chooses the option which best satisfies his current goals.
"OK", says the objector, "your agents have some autonomy because they can choose between the various requests they receive. But they are not properly autonomous because the range of choices is dictated from outside (by the activities), they do not themselves decide what the range of choices is".
This objection misses the point made earlier, by Wittgenstein and others, that participation in an activity enables us to make choices we couldn't otherwise make. Unless we are participating in the game of chess, we cannot choose between castling kingside and queenside. This choice is only available to us once we have entered into the chess-playing activity. Social-activities, in other words, are not constrictive but enabling. As another example, a couple cannot get married except in a societal and cultural context. Without culture, while they can visit a civic building, swap rings, say appropriate words, sign in a book and take lots of photos, they have not got married despite the fact that (in the particular culture and society that the authors come from) these are the physical actions that a couple do perform if they get married. Participation opens up new possibilities.