General, or low fidelity modeling relies on generalizations or assumptions about a large group of clients without accounting for all the variables that impact an individual’s ability to make sound decisions. Low-fidelity models do not account for all the variables or circumstances that can impact a homeless person. The unique risk factors and uncertainties that face the homeless can cause a chain of events that impact their ability to benefit from many of the social services that are available to them. A high fidelity model will capture all the factors that may impact the client’s progress. It will enable the social worker to choose a scenario that considers the unique combination of factors and obstacles that stand in the way of progress for clients like Jane.
What kind of knowledge do you need in order to construct your simulator?
We’re currently reviewing several studies that look at how and when the homeless use certain services, and what influences their decisions. Due to the variability of participants in these studies, it’s difficult to reach conclusions that are specific enough to predict the impact of a study on any given sub-group or homeless profile. In order to mimic human decision-making, we need to understand how the homeless make decisions given their unique attributes and circumstances. We need to understand their motivations. We also need to understand the physical and psychological barriers that prevent them from making what we would consider “rational” decisions.
For example, consider a 40-yr-old chronically homeless man who’s been panhandling all summer in Toronto. As winter approaches he decides to move into a shelter and get off the streets. The social worker can’t find space at any nearby shelter. She suggests that he look into low-income housing. The social worker informs him that he is eligible for a rent subsidy. After two years on the streets, the client is only able to think in short terms. The client is suddenly faced with alternatives that do not fit with his short-term goal to ind temporary shelter. The client is wary of the government. He may not want to live where permanent housing is available because it’s not in a familiar neighborhood. But most of all, the introduction of unexpected alternatives is confusing and does not match his goal. So he leaves the social agency and returns to the streets.
What research challenges or obstables lie ahead for building a homeless person emulator?
Human beings are neither completely rational nor completely irrational, we lie somewhere in the middle. People are much more complicated than computers. It’s easy to model explicit knowledge, but people often reason with implicit knowledge. So how we represent and model implicit knowledge is critical to the success of our project.
On the engineering side, emulating how homeless clients create and rank goals using existing engineering techniques has been challenging. For example, the Mars rover needs to perform experiments. That’s its goal. To perform those experiments there will be a set of plans, or algorithms, that enable the rover to find good spots to drill, with some level of adaptability to handle unexpected obstacles or uncertainty. Humans are different. We have the same basic needs, but what we do to satisfy those needs depends upon the uniqueness of who we are, where we came from and where we want to go.
Much of the research in emulating human decision-making, up to now, has assumed that we are rational, and that we use all the information that is available to us. This isn’t the case. We have irrational biases and beliefs that influence our decisions, many that are based upon emotions. Mapping emotions to beliefs and the resulting actions we choose to perform is still an open problem.
Accessing the right data is another challenge. It’s difficult to gain information from social service practitioners and government agencies due to privacy laws. At this point we are relying on existing studies. Capturing small but impactful differences among homeless groups and subgroups is key to our research. We want to ensure that we're capturing the most accurate set of characteristics that match specific groups of homeless people as they move through stages of a treatment or intervention programme.
At what stage are you in the development of your simulator?
I'm building a simulator that we call BRAMA which means “gate” or “passage” in Polish. The underlying components are complete. I’ve chosen theories that have been used by practitioners in the field. The psychological theories that relate to how we choose goals have been identified. Maslow’s Hierarchy provides the framework for dictating how clients choose and rank goals. I’ve modeled the way people progress through stages of a treatment plan.
From the engineering side, a reasoning infrastructure is completed. It uses engineering techniques to understand how social workers think about their clients, how they evaluate them and come up with a treatment plan. I’m also building the environment that represents a homeless client’s world. This includes client characteristics, how they reason and interpret their world, and how they achieve the short-term and long-term goals.
I’m currently collecting and comparing data on the progress a simulated client makes, by graphically representing their progress and any obstacles they encounter to reach a goal.