Transport Canada has piloted the use of Artificial Intelligence (AI) to perform risk-based oversight by scanning Pre-loading Advance Cargo Information (PLACI) to identify potential air cargo threats (i.e. “bomb-in-a-box” scenario). By using AI to perform this function, there is potential for Transport Canada to save 1,000’s of hours and conduct risk assessments on all air cargo shipments coming from abroad -- in real-time.
Innovation Summary
Innovation Overview
To keep pace with the rapid changes in the transportation sector, Transport Canada must explore using disruptive technologies like artificial intelligence (AI). In “thinking big,” Transport Canada set its sights on AI augmenting every process and procedure within the department, freeing up humans to work on the ‘sticky’ problems. However, Transport Canada recognizes the importance of “starting small” with an ideal use-case that allows for applicable testing. Then, if that case is successful, “scale fast” to other areas of the department and possibly, the whole of the federal government.
In this context, Transport Canada identified the perfect use-case for AI with its Pre-load Air Cargo Targeting (PACT) team that receives approximately 1 million air cargo messages per year. For example, if one employee, working at an unrealistic rate of one message per minute, spent all their time reviewing messages, they wouldn’t even have enough time to review 10% of all messages received in a year.
To date, there are very few governments that have the dedicated resources to scan air cargo messages for risk before cargo is loaded and of the ones that do, none use AI. When considering the resources responsible for customs and border protection, no country is using AI in that capacity either.
This use-case represented a true collaborative effort by senior executives from different areas of the department. The PACT team resides within the Aviation Security area of the department and is being directly supported by the Digital Services (IT) area of the department. Transport Canada is investing effort to renew the way it conducts business, so strong support also comes from the Transformation area of the department. Additionally, Transport Canada hired one of Canada’s Free Agents to lead the project. To top it off, the department did not have enough AI capability in-house, so it partnered with an outside firm that subcontracted with an award-winning IT firm with expertise in AI.
Through this use-case, Transport Canada attempted to answer two questions:
1) Can we use AI to improve our ability to conduct risk-based oversight?
2) How can we improve our effectiveness and efficiency when assessing risk in air cargo shipments?
To help answer these questions, we proceeded in two steps. In the first step we used machine learning (ML) on a subset of the data and in the second step, we used natural language processing on a different subset of the data.
In the first step, we embedded a Transport Canada employee onsite with the IT firm. We took two different ML approaches in working with the subset of data: unsupervised learning and supervised learning. In unsupervised learning, we blinded ourselves to the ‘outcome’ of the risk assessment applied to the cargo message to try and learn about the relationships between all cargo messages. Then, we used supervised learning, where we tried to learn the relationship between the inputs (cargo messages) and the outcome (i.e. did this cargo message warrant a greater level of risk). In the second step, we used natural language processing on a different subset of data with the goal of being able to automatically tag a cargo message with a risk indicator based on ‘free text’ fields.
The results from steps one and two were very promising – AI will increase safety and security 15-fold, as every single message will be risk-assessed. Additionally, through better use of resources, PACT can use AI to increase capacity, while minimizing the number of people required to do the work.
PACT’s capabilities before introducing AI:
*Very burdensome to conduct data analysis
*Duplication of effort in data handling tasks (200+ hours to import, clean, and archive data)
*Single dedicated resource for scanning cargo messages through batch processing
PACT’s capabilities after introducing AI:
*Save 1,000’s of hours by allowing for timely and proactive data analysis of cargo data
*Conduct risk assessment on all cargo in real-time
*AI will support PACT in meeting its security outcomes, while allowing PACT to scan cargo messages from more air carriers
For Transport Canada to stand up a team of targeters similar to their counterparts around the world, PACT would have to hire 24-36 new employees. Not only does the AI save on all the hiring costs, it realizes the productivity of an ‘employee’ that can work 24 hours a day, 7 days a week – without needing to take a break.
Beyond the gains for PACT, this model could be adapted to aid in targeting other modes of transportation (e.g. marine, rail, road, etc.) or even expanded to support the mandate for Canada’s agency responsible for customs and the border. Ideally, all government departments with an interest in the safety and security of Canada – including intelligence, border, and police agencies – would have a single database with information that could be used to optimize the process for providing risk-based oversight to cargo. Thinking bigger, maybe even governments worldwide.
Innovation Description
What Makes Your Project Innovative?
This project is innovative because, to the best of our knowledge:
1) Within Transport Canada, no teams use AI for risk-based oversight.
2) Within the Government of Canada, no other departments use AI for risk-based oversight.
3) Of the countries that have dedicated resources for scanning air cargo messages, none use AI for risk-based oversight.
4) Of the countries that have dedicated resources for border protection, none use AI for risk-based oversight.
What is the current status of your innovation?
As of the date of submission (September 2018), this project is moving into implementation and because of the nature of AI, the implementation phase will be iterative (i.e. there will be ongoing evaluation to understand whether the AI is delivering as intended). In April 2018, we completed the first step of the 'proof-of-concept' (machine learning). The second step (natural language processing) was completed in May 2018. Since then, Transport Canada has worked on creating a procurement vehicle to bring in a firm with expertise in AI to develop a 'minimum viable product' based on the success of the proof-of-concept.
Innovation Development
Collaborations & Partnerships
This project was a true collaborative effort.
The partners within Transport Canada:
*Aviation Security: where the PACT team resides
*Digital Services: IT function of the department
*Transformation: mid-stream on renewing business processes of the department
The partners outside the department:
*GCStrategies: represents network of firms with expertise in AI, cloud, and blockchain
*Lixar IT: the award-winning IT firm with expertise in AI
*Canada's Free Agents: hired Free Agent to lead project
Users, Stakeholders & Beneficiaries
Transport Canada benefits by increasing its capacity to perform risk-based oversight, improving its effectiveness in identifying trends in air cargo, and developing in-house expertise in artificial intelligence.
Canada – writ-large – benefits because cargo in the aviation industry is safer and more secure. By extension, this translates to the aviation industry across the globe, as Transport Canada will share best practices with other governments who perform similar operations.
Innovation Reflections
Results, Outcomes & Impacts
Early results indicate PACT’s capabilities with AI:
*Save 1,000’s of hours by allowing for timely and proactive data analysis of cargo data
*Conduct risk assessment on all cargo in real-time
*AI will support PACT in meeting its security outcomes, while allowing PACT to scan cargo messages from more air carriers
Upon completing the first two steps of the proof-of-concept, it is apparent that PACT’s capabilities after introducing AI are significant. For Transport Canada to stand up a full team of targeters to resemble their counterparts in the United States or Europe, PACT would have to hire 24-36 new employees. By using AI, not only does Transport Canada save the cost of hiring 36 new employees (not to mention the costs it would take to train all the new employees in the very specialized skill of targeting), it realizes the productivity of an ‘employee’ that can work 24 hours a day, 7 days a week – without ever needing to take a break.
Challenges and Failures
As with any project that employs AI, the key ingredient is data. Luckily, PACT has plenty of data. Unfortunately, the data wasn’t in a format that easily facilitated the use of AI. Before the “AI” portion of the project could begin, we had to ensure that the data were in a format that allowed for the use of AI. In the next phase of the project (i.e. minimum viable product), the first order of business will be to address this challenge head-on by creating a pipeline, so that all cargo messages received by Transport Canada will feed into a single database.
For a machine learning algorithm to be successful, it needs to be trained on “positive cases.” In this scenario, that would mean the “bomb-in-a-box” scenario. Fortunately, there aren’t many “positive cases” of bombs on airliners. However, this complicates the process for training the algorithm. To mitigate this moving forward, PACT will construct other key performance indicators to ensure that the AI maximizes its usefulness.
Conditions for Success
For any AI project to be successful, one needs copious amounts of data. Similarly, the necessary IT environment with the right tools and capabilities is a must. Given the risk aversion around disruptive technologies in general, it was also essential for this project to have support from senior management. Not only is there support from the Deputy Minister of the department (highest ranked non-political person in the department) , so are the other senior executives who have oversight of the project.
Replication
This solution could be replicated by any other team (or country) that performs a similar targeting function. Within Transport Canada, there have already been very preliminary discussions around incorporating the lessons learned from this project to replicate a risk-based oversight model that uses AI within other transportation modes (e.g. marine, rail, road, etc.). Additionally, this model could be tweaked to include the expanded mandate for Canada’s agency responsible for customs and the border.
Ideally, all Canadian government departments with an interest in the safety and security of Canada – including its intelligence, border, and police agencies – would have a single database with information that could be used to optimize the process for providing risk-based oversight of cargo. Similarly, there could be a simple way for sharing information between countries, as it relates to the safety and security of air cargo.
Lessons Learned
The “Artificial Intelligence Revolution” is coming sooner than we think. The technology is growing at an exponential rate, even in the last 3-4 years. Technology firms have developed AI tools that can read text that appears within images! That would have sounded impossible 15 years ago and even 5 years ago, that would have sounded like something that would have taken decades before we would have figured it out. According to one study, two years ago, an average of 80 percent of the work translated by machines needs to be fine-tuned by human translators. Today, it’s 10 percent!
Many firms in the private sector are already realizing the benefits of AI and given the potential gains to be had, it’s categorical – the public sector must use AI if it has any hope of keeping pace.
Additionally, in some countries with aging populations (i.e. more people retired than ever before and fewer people working in the public sector), it will be all the more important for governments to figure out how to maximize its resources. No longer can the public sector afford to dedicate teams to the same kinds of issues and problems it did in the past. It will need to allocate its resources judiciously. And as was demonstrated with PACT, AI can go a long way to helping augment processes to freeing up humans to work on the sticky problems.
Status:
- Implementation - making the innovation happen
- Evaluation - understanding whether the innovative initiative has delivered what was needed
Innovation provided by:
Date Published:
25 October 2018