Digital Humanitarians and Disaster response: A Discussion with Rescue Global

I recently attended a workshop with Rescue Global (RG) and thiRescue-Global-Logos blog post captures some of the discussion and points that are interesting and useful to digital humanitarians like myself. The better we understand how disaster response works (or doesn’t work) the better we can build our tools and get them used in anger.

Rescue Global

Overall RG goal is to be a trusted reliable broker of good, up to date information in disaster response scenarios.

  • Show up, provide good information, build trust, get tasked to do stuff
  • Work with the local country, NGOs, and authorities. Build up local resilience so they don’t always rely on the big guys to fly in and run the show.
    • Small, low level partnerships to prove it works and get small wins, build trust and local experience
  • Have some state of the art equipment and technology to trial and test

Crowd mapping

Humans are great at understanding but can’t process lots of data, can get bored and are expensive. Computers are cheap, dumb but keep going. Need them both in disaster response situations. There is just too much data in modern disaster response events for humans to process. We need computers in the loop. You can look at raw tweets (distribution / rate). However does not tell you what is actually happening on the ground. Need to understand the data more, need to get the computers to actually interpret the data/images. Bootstrap this by relying on a human subsample who classify tweets (or other data) in event classes: hospital operating normally, food deficit, temporary shelter, missing person, …

  • Challenges:
    • Sparse information
    • Untrustworthy reports
    • High submission frequency
    • Constraints and coordination (agent roles, deadlines, …)
  • RG explaining how they understand what is going on, building up intuition and an operational plan
    • Fascinating to hear the thought process: hypothesis generation, inferences about communications, infrastructure, political influences, type of disaster, etc.
    • Very nice task division between team members. Areas being looked at include terrain, major access routes and airports, political environment, other organisations who will appear, what access complications that will cause, etc.
  • Once basic picture is understood there is the need to start cross referencing with other sources to go to the next level

Remarks

  • First priority is to deploy to gather data and prepare a response, not deliver aid
  • Key: don’t get dragged down to low level tactical, stick to strategic level
    • Important to maintain consistent comms link between strategic in London and operational on the ground
  • Big problems are in data integration, coordination, joining up.
    • Needed: powerful, flexible, collaborative, UIs, built on open standards and open, shared data
    • Many tools exist: micromappers, ushahidi, … problem is not all brought together, not in one place
    • Importance of sharing data in templates people are used to and expect
    • Politics between organizations play a large role as well
      • Not unusual for good command structures and plans to be overridden by politics
  • Advantages of algorithms
    • beware of biases in human expertise. For example, classification of event type will be biased towards area the person is comfortable with (e.g., health), algorithms reduce that bias
    • earthquake hits: that not the big problem, big problem is spread of infectious disease, so the response should focus around this. Computers are much better a picking up this kind of thing.
  • Disadvantages of algorithms
    • beware of biases in data, e.g., tipping point when smart phone batteries run out, report characteristics change though situation on the ground may not
    • importance of capturing full history of data, review regularly how things work, did we do the right things, …
  • Important difficult question: how to figure out, once the response has started, how much impact the response is having. Is it having the expected response, are we making progress, …

Human Agent Collectives

Multi-agent explorative path planning of UAVs in a risk aware framework. How to coordinate UAV path planning with humans setting tasks?

  • Driving two UAVs probably max, even then, having another person help out makes a big difference, more resilience & automation needed
  • Danger when controlling individual UAVs to get dragged into the tactical level and lose sight of the strategic level
  • Battery life still a problem and forces prioritization
  • Good that an algorithm can make suggestions what to look at, may be things a human may not think of. But need to be able to override it. Can avoid the human passion/bias

Remarks

  • Generally more demand and push for CRIPs: Commonly Recognised Information Picture
    • allows: “here is what is happening, here is the resources we have, lets take action”
    • Need to be shared and dynamic, need to be easily exportable in predetermined exportable templates (e.g., for WHO: population centres, main routes, …) with attached provenance. Builds trust.
    • Problem: standards and politics
  • Ideally, there is good top town – bottom up collaboration between strategic and tactical levels. In reality most happens at the bottom and then updating top/strategic at home later (especially given timezone differences)

Planning

How to most efficiently do distributed resource allocation under spatial and temporal constraints. Approach combines humans and agents.

  • crucial: need to understand why an algorithm makes certain decisions, what is the rationale? Need to build trust and understand. People lives at risk
  • You don’t want to give people on the ground strategic information/directives. You want actions not a discussion. Need different information for different levels (strategic, operational, tactical)
  • Computational support useful
    • In training: useful across all levels, allows to communicate a holistic view of how RG works
    • In operation: different levels (silver, bronze, gold) need different information. Sector commander does not need weather information for example. Need for different UIs.
    • In briefing: to communicate with other organisations and to allow for field information gathering and sending back. Avoiding too much planning low down.
    • To let teams know how they fit in, where they make an impact, how they relate to others, .. Shows impact, very uplifting to morale.
    • In general: helps with division of labour

Remaks

  • Nobody wants a 400 page report or a 10hr lecture. Whiteboards, sketches, … can be empowered more by computers/data/viz.
  • You can (and people do) give numerical values/ordering to people, environment, structures, … to drive priorities
  • Utopia: agile teaming algorithm that puts together kick ass team from different agencies. Problem is UN structure, way they operate, tasks are siloed, politics / mechanisms don’t work that way, and liability (don’t know everybody, their experience, their insurance, …)
    • Only works in an informal way

Provenance

  • Challenge: how to track flow of data in the system
  • Store eg: tweets, data derived from tweets (e.g., location), data derived from that (people, …) -> provenance graph
    • do inference and pattern mining
  • Cleveland incidents website: allow reports through social media
    • Gets them more reports
    • Also allows them to inform people more efficiently
    • <-> London does not take into account social media, still requires 999 call

Remarks

  • Big issue: unreliable network, default should be: no connection or very unreliable
  • Need to be able to assign trust values to sources
  • Articulate provenance data standard before an event and get acceptance and buy in from other organizations.
  • The question always is: is it information or is it intelligence?
    • Answer depends on the numbers. Volume is correlated with accuracy. If there is little volume, need to look at the source of info more closely
    • Also depends on the plan you are going to execute. If you are putting lives at risk you are going to audit the hell out of it.
  • Disaster response not currently intelligence led (unlike fire brigade, military, …), the world just rocks up. Often with the wrong stuff/goals.
  • Overall very much relationship and behavioural driven. People can easily get into arguments. Data & trustworthy provenance can play a very important role.

Crowdsourced Data

  • RG find crowd and micromapper data useful, feeds into their models, but not trusted by the UN, and others …
    • Generally the level of trust people give crowd data is proportional to their liability
    • Micromappers in London: sheer distrust of the data as the people are so far removed from the place where its happening
    • However, virtual workforces around the world to collect data and crosscheck: apolitical, would assume less bias
    • Need to work on this, need to prove how data was collected, checks in place, confidence/uncertainty bounds,
  • RG: “We give risk data to others to allow them to respond”. We need enough transparency and confidence in that crowd sourced data.
  • Have been cases with unreliable information, RG physically mapping it out themselves and flying there and work against the tide to say hey you are wrong, you missed this

Overall remarks

  • Computational support is good as it can avoid the human bias/passion and prompt a human to think about other things, avoid getting sucked in to specific issue
  • Data quality & understanding paramount
  • this is a force multiplier, take in more data than you can otherwise do, help you make decisions
  • Information/uncertainty can change very quickly need to be very agile in planning. Re-planning very important
  • The CNN effect: everybody turns up and bandwidth drops significantly
  • Biggest possible win, massive: multi agent command support tools, common platform, interoperability

UAS

What do RG really use UAS for & what are the real pain points? Endurance? Autonomy? …

  • Good use cases:
    • Reconnaissance: Quicker to send UAV 5km to look at something than send people. Use UAV to work out where to send resources
    • Situational awareness when team already deployed
    • Triage of jobs too DDD for a human (e.g., follow a radiation cloud)
    • Live feeds: political advantage (televise, ensure it gets attention), strategic advantage: aerial platform forces you take a strategic view
  • More UAVs useful, all depends on the area you are dealing with
    • the more UAVs you have the more you can sectorize, leverage division of labour
      • requires more autonomy though

— Dirk

One thought on “Digital Humanitarians and Disaster response: A Discussion with Rescue Global

  1. Pingback: CDAC Workshop: What does the future of digital humanitarian response look like? | Machine Doing

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s