More workAlonso Holmes
SynapseDesigning a control center for AI-powered security checkpoints.In 2018, Synapse began adding AI to security checkpoints in airports around the world. Their mission is to make checkpoints safer and faster, by pairing human screeners with clever software.Synapse reached out to me in mid-2018 to design (and build) a product for security checkpoint supervisors. The goal was to give them control over the way the Synapse AI worked, and the information and statistics to understand how well it was working.
For competitive and security reasons, details on research and process have been omitted, and mockups have been changed.
My roleI was solely responsible for UX and UI, and produced all of the sketches, wireframes, styleguides, and UI specs for the Supervisor product. I built a collaborative relationship with the folks at Synapse - often posting in-progress work in Slack, and regularly holding more formal feedback sessions.
I was lucky to have good access (through the client and their translator) to a Japanese security team at Narita International Airport in Tokyo. This meant that I was able to do a lot of quality research early in the project, and was also able to iterate with screeners "on the ground" as the UX and UI developed.
I also helped build the product, architecting and coding the frontend (react + graphql + mobx) in partnership with the Synapse development team.
I was lucky to have good access (through the client and their translator) to a Japanese security team at Narita International Airport in Tokyo. This meant that I was able to do a lot of quality research early in the project, and was also able to iterate with screeners "on the ground" as the UX and UI developed.
I also helped build the product, architecting and coding the frontend (react + graphql + mobx) in partnership with the Synapse development team.
BackgroundStaring at bags is tough
A screener at work. Image from SRI & video from SynapseScreening baggage is a very demanding job. Screeners spend hours staring at chaotic x-ray images of suitcases, keeping their eyes peeled for anything that looks like a threat.
Focusing this hard on complicated x-ray images can really wear out your brain and eyes, so screeners have to take regular breaks - every 15 minutes is typical.An extra set of eyes
Video credit: SynapseSynapse adds an AI to each x-ray machine, which gives screeners an extra set of eyes. The AI looks at the same x-ray images that screeners do, and labels anything that it thinks may be a threat.
The screeners still make the judgment call about whether to pull and search the bag - Synapse simply calls their attention to threats they might have missed.Solving for supervisorsEvery checkpoint has a “supervisor”, who is responsible for the overall safety and efficiency of the checkpoint. Each supervisor manages multiple screeners and x-ray machines.
Synapse was working great for individual screeners, but supervisors had begun asking for ways to understand and control the AI across the entire checkpoint.
Focusing this hard on complicated x-ray images can really wear out your brain and eyes, so screeners have to take regular breaks - every 15 minutes is typical.An extra set of eyes
The screeners still make the judgment call about whether to pull and search the bag - Synapse simply calls their attention to threats they might have missed.Solving for supervisorsEvery checkpoint has a “supervisor”, who is responsible for the overall safety and efficiency of the checkpoint. Each supervisor manages multiple screeners and x-ray machines.
Synapse was working great for individual screeners, but supervisors had begun asking for ways to understand and control the AI across the entire checkpoint.
The solutionAfter significant research with the Synapse team and the screeners on the ground, I designed three main interfaces:
StatsUnderstand the AI's performance through statistics.
SensitivityControl the AI’s sensitivity to each type of threat.
Bag HistoryFind any bag that was previously scanned.
Stats
The stats screen is focused on answering questions that supervisors regularly have.Questions like…"How many threats have been detected today?"
Every day/threat combination gets a card, which makes it easy to see how many of a certain type of threat has been flagged, as well as how much that threat made up of the whole."What impact did my change have?"
Changing the AI’s sensitivity will lead to marked increase (or decrease) in detections, so these changes are clearly marked and have explainer tooltips."How sensitive is the AI right now?"
Visual sensitivity is much faster to parse at-a-glance, and meshes well with the discrete sensitivity settings on the “edit” screen.Sensitivity
The AI’s sensitivity can be controlled for each different type of threat. The Sensitivity screen allows supervisors to set the sensitivity, and to preview the effect of the change.Common-sense sensitivity settings
The sensitivity slider simplifies a set of domain-specific concepts (thresholds, confidence scores, etc) into a single setting that is intuitive for supervisors.Simulated impact
It’s important that supervisors understand the impact of the changes they make to the AI’s sensitivity. Every time they move the sensitivity slider, they can see estimated statistics and example bag images for the proposed setting.Bag History
It's often useful for supervisors to "go back in time" and inspect a specific bag. The Bag History screen is designed to make it fast and easy for them to recall a particular bag, or browse through a set of bags.Matching mental models
Supervisors most commonly recall bags by bag style, threat, and time frame. The Bag History screen allows supervisors to filter on all of these attributes, giving them multiple ways to find the bag they're looking for.