Using Artificial Intelligence to Automate Tasks
Artificial intelligence (AI) tools present new opportunities for public servants to find ways of working smarter, starting with automating common, low-risk tasks they perform on a regular basis. Working more efficiently can free up mental capacity for higher-level thinking and problem solving, leading to more effective stewardship of public resources and excellence in meeting the needs of those served by the Government of Canada.
Start small: choosing the right tasks
Not every task can and should be automated. Tasks that require judgment, critical thinking, empathy, and creativity are best performed by humans who can bring to these types of tasks their experience, situational awareness, and understanding of relationships and context.
However, AI, and particularly generative AI, is expanding the range of tasks that can be automated. Tasks that involve synthesis, writing and editing, classifying, analysis, and data or information extraction may be good candidates for AI tools. An task can be entirely or in part performed by AI.
As a starting point for identifying tasks where AI might automate or augment your work, consider the following aspects:
- Task complexity – Does the task involve following a consistent process with multiple steps with a clear outcome, such as analyzing data or finding patterns from several sources to create a report, or summarizing meeting notes and sending them out with action items?
- Frequency – Is this a task that you or your team perform on a regular basis, such as generating a monthly report, or when a condition is met like whenever a foreign leader gives a speech or once a complaint is registered?
- Potential for human error – Does performing this task manually introduce mistakes due to improperly following the steps or other types of errors, on a regular basis?
- Time consuming – Does this task take up a lot of time compared to other tasks?
If you answered "yes" to most of these questions, then this task may be a good candidate for automation. It’s also worth thinking about the strategic alignment and what impacts automation might have on business goals, such as better compliance, faster turnaround times, or freeing up employees for higher value work. This will help to gain buy-in from managers and other stakeholders for automating the work, which is important because setting up automated processes takes time and effort.
Consider the strengths and abilities of AI and public servants
In general, generative AI (gen AI) is good at several tasks, such as:
- repetitive and rule-based tasks, like data entry, information retrieval, content formatting, and basic code generation
- time-consuming and resource-intensive tasks like research, data analysis, content summarization, and initial draft creation
AI must not be acting alone; humans must provide instructions and oversight, which include:
- data selection and preparation to ensure that gen AI are provided with high-quality, relevant data, either as input, or in its training, that is representative of the intended use cases
- prompt design and refinement to craft prompts to the AI tools that elicit accurate and useful responses
- output evaluation and refinement to review and edit AI-generated content to ensure accuracy, relevance, and alignment with relevant content guidelines, such as the plain language standards from Accessibility Standards Canada, the Canada.ca Content Style Guide or other relevant standards
- continuous monitoring and feedback, since AI tools, especially generative AI tools, can make mistakes; humans must provide feedback on AI performance and identify areas for improvement
Find the right tool for automation
Gen AI can be very useful, but it's not always the right tool for the job. Consider which tasks your team would like to automate and which systems could help: gen AI, non-gen AI based systems, and even traditional software tools. The tools in this list are available to most employees and can be used to automate tasks.
- Microsoft Power Automate: Create simple workflows to move files, send notifications, and process data between apps.
- Excel macros: Automate the generation of formulas for calculations, and finding trends and patterns in spreadsheet data. Copilot for Excel has made creating macros easier and more intuitive.
- Robotic process automation: Tools like UiPath and Blue Prism mimic routine computer tasks, such as copying data or generating reports.
- Generative AI and AI assistants: Microsoft Copilot, Google Gemini, and other similar tools can summarize documents, draft emails, extract information, and write code that enables automation through Python or JavaScript. Tools deployed on GC systems can be used for non-sensitive internal documents and emails; external web tools can be used for brainstorming and analyzing publicly available documents and information.
- No-code platforms: Power Apps and AppSheet help you build simple apps and automate tasks without needing programming skills.
Make the most of the tools already available in your department first. Ensure you have the required permissions for the software, and take any training on the systems offered by your department or the Canada School of the Public Service so you understand how to use it appropriately, ethically, and safely within the Government of Canada.
Protect privacy and data security
As you are working to improve efficiency, you also need to protect privacy and data.
- Know your data: Only automate tasks with non-sensitive information unless you're sure the tool meets government privacy standards.
- Control access: Make sure only authorized people can see or change automated workflows and their results.
- Secure storage: Use tools that encrypt data and keep it within approved government cloud environments.
Some tools may require you, or your department's IT team, to take additional steps.
- Review third-party tools: Check privacy policies and security certifications before using outside platforms.
- Monitor and audit: Set up logs to track automated actions and regularly review them for unusual activity and expected performance.
- Human oversight: Assign someone to check automated outputs and respond to any issues.
Check with your IT or privacy officer before automating tasks involving sensitive data. For further guidance, consult and the documentation on responsible use of artificial intelligence in government.
Use case: data extraction and clean up
A team was tasked with analyzing the last five fiscal years of a program's reporting on services funded across all provinces and territories. The analyst used ChatGPT to create Visual Basic for Applications (VBA) macros and complex Excel formulas, but relying solely on these tools still required a significant amount of manual work. Inspired by reports of successful Python code generation with ChatGPT, he chose to introduce Python to automate processes, despite having no prior coding experience. The system was first designed in writing, followed by the creation of prompts for ChatGPT to generate VBA macros and Python scripts to automate all ETL (extract, transform, load) processes.
This approach involved:
- extracting data from hundreds of PDF reports
- transforming raw extracted values into appropriate formats
- loading them correctly into pre-established spreadsheet templates that fed the program's Power BI dashboard; Excel files assisted in transforming data formats with complex formulas and mapping the destination for each piece of data in the spreadsheet templates
Python scripts coordinated these Excel tools using VBA macros for data extraction, formulas for data transformation, and maps for data loading, managing data across four different Excel files until a final spreadsheet was created with data from all reports. The automated system was built in two weeks and performs ETL processes twenty times faster than previous, error-prone manual work. The solution is accessible to users without coding or Excel experience and is easily scalable.
Currently, the team is expanding the system to other programs, which requires new data value standards and data label maps, and is working to remove all VBA and Excel-dependent processes by incorporating them into Python functions for greater reliability and speed. ChatGPT was selected for its reputation in Python code generation, with specialized coding GPTs and the canvas tool providing valuable support. The essential human element involved designing the system, determining process types, creating data value standards and data label maps, and reflecting these in the prompts. This experience highlighted how AI can augment time-consuming processes, emphasizing the importance of creativity, testing, piloting, and reviewing results. AI, at this stage, serves to augment rather than replace jobs.
Conclusion
It’s clear that artificial intelligence offers great opportunities for making operations more efficient, if we apply it to the right tasks, respect data privacy and security, and ensure adequate human oversight. It’s important to remember that humans are ultimately responsible for AI outputs, and so with this potentially transformative technology, public servants bear great responsibility in implementing it responsibly. There are many considerations to keep in mind, which means it’s always better to work out in the open, and to collaborate widely to make sure different perspectives contribute to shaping decisions around automation.
Resources
