Problem 1 - Improve dispatching to send more appropriate community responders (avoid escalation)

[juliuscesarsalad]

Brief description of your idea: There is a growing demand and transition toward police alternatives. What if there was a way for 911 dispatch to send callers more appropriate help, which often is not a police officer. How can we build a system leveraging machine learning (predictive technology) and NLP to help 911 dispatch get callers on the phone with more trained and specialized professionals, and or send callers trained, unarmed professionals to respond to nonviolent matters (i.e. mental health, school discipline, etc.)....

SCENARIO: Caller calls 911 and is connected to 911 dispatch. Both the 911 dispatch and the system listens to and analyzes the caller's purpose for calling. Based on the information the system gathers from the caller (key words, sense of urgency, location, etc.), the system makes a recommendation on what department/service will best handle the situation to the Dispatcher 1) Send the police 2) Send an alternative. The Dispatcher can decide to accept or reject the systems recommendation. If the system and the Dispatcher agree on the Alternative option, the system can automatically connect the caller to the appropriate department/resource and facilitate the required action steps.

What would be the impact of your idea if implemented?: This would support the idea that police officers are not always the best option to respond to non-violent matters. It would free up police officers for more critical matters.

[Lisaboneta]

I love this idea, I wonder if it's possible if we can make an app or perhaps a "chat bot" for less urgent situations that can do similar to what you said such as

system gathers from the caller (key words, sense of urgency, location, etc.), the system makes a recommendation on what department/service will best handle the situation to the Dispatcher 1) Send the police 2) Send an alternative

This could potentially allow users to add information to report situations such as photos, videos, etc.

[Lisaboneta]

Going off of what Boz just mentioned in our office hour, perhaps this could also record or track "abusers" of the system who call police or dispatch: i.e how many calls were actually substantial vs. discriminatory. This does not need to be tied to specific users necessarily but could be important data for analysis of certain neighborhoods and relationships that exist among those who live there. This could also then be used as a way to help improve the way calls are answered to in the future.

[KallieFerguson]

Going off of what Boz just mentioned in our office hour, perhaps this could also record or track "abusers" of the system who call police or dispatch: i.e how many calls were actually substantial vs. discriminatory. This does not need to be tied to specific users necessarily but could be important data for analysis of certain neighborhoods and relationships that exist among those who live there. This could also then be used as a way to help improve the way calls are answered to in the future.

My word of caution is that the majority of what we might classify as "abusers" of the system are likely people navigating mental health issues and we want to be careful not to increase discrimination against those individuals while trying to identify calls that might be racially motivated. On the face, it's probably very difficult to tell from the just 911 call whether there's a racial motivation.

I do really love the idea of being able to screen to send out police alternatives. However, we'd need a massive overhaul of the system first!

[Lisaboneta]

On the face, it's probably very difficult to tell from the just 911 call whether there's a racial motivation.

I think the idea here that I didn't mention would be there would have to be some sort of follow up reporting system from the responder. Of course this is difficult to navigate because if the responder has a racist response the the follow up could be also misreported. This is a challenge I guess we have to figure a more neutral solution to and leaves little to no room for any type of bias.

If the system has a backend ML algorithm and can learn which situations seem to repeatedly happen then end up being followed up as being perhaps racially motivated (i.e if someone is reporting a younger person just walking on the street who they deem is being "suspicious" or uses vocabulary that could be racist/classis/albeist) it can better detect which responder can be sent out.

So I believe for this to be successful there needs to be strong and significant follow up that can help the program classify events based on statistics and analysis of outcomes from the reporting.

[daveshack-ibm]

This is an example of an area where the systems are going to be different in different communities.

In my city, 911 operators have a decision tree online (with a paper backup) that guides them through the information that is spontaneously provided along with targeted follow up questions to arrive at the (hopefully right) decision about how to dispatch the call. A technology add on which analyzes the same information in real time can reduce mistakes but it might need to be customized to the decision tree, or require a standardized tree to be useful.

Tech might really help in a situation where there are simultaneous calls for the same incident. Depending on workload and how the calls come in, two or more calls for the same incident could be handled by dispatchers who are unaware of the other call. Tech that can recognize that and better organize it might be helpful.