How can the public sector use AI effectively? Part 2
We take a pragmatic look at AI in the public sector and discuss the ethics surrounding it. Read part one.
Following on from last week's blog, we will discuss the latter half of the four factors that often have a significant impact on the success of an AI project:
Data availability and quality
People with the right skillsets
Choosing the right AI solution
AI ethics including fairness and explainability
Selecting the right tools
The market for AI applications is still maturing, with new offerings emerging frequently. Navigating this ever evolving landscape can be hard, especially if you're not familiar with AI terminology.
We've compiled some commonly used machine learning techniques below to demystify the topic and demonstrate the value certain tools can offer.
Classifies unknown data points into existing categories based on the characteristics of a given category
- Deciding if an email is spam or not
- Deciding if a package is high-risk
Predicts the value of an unknown data point based on existing data
- Predicting market value of a property
- Forecasting traffic conditions
Groups similar data points in a dataset into categories
- Grouping citizens to find subgroups requiring more support
- Clustering smart-meter data to identify groups of appliances
Structured prediction (supervised)
Generalises supervised ML techniques by consider the structure of the dataset, e.g. sentences, objects, audio
- Parsing a sentence to understand intention
- Convert handwriting into text
Dimensionality reduction (unsupervised)
Reduces the complexity of the dataset by keeping the most relevant variables to improve accuracy of models
- Used as the first step when evaluating & developing algorithms
AI business solutions
There are certain business problems for which AI is commonly used, and there are usually commercial applications that can be customised for an organisation's particular use case. For context, we've outlined some specific examples of AI applications that are used to improve efficiency and effectiveness.
Natural language processing (NLP)
Processes and analyses natural language to extract words, meaning and context
- Virtual assistant
- Machine translation
Emulates human vision for machines, e.g. detecting objects and features
- Camera tracking for broadcast
- Automated passport controls
Picks out anomalous data points within a dataset
- Spotting tax-evasion patterns
Understands how data varies with time
- Forecasting budget
- Finding economic trends
Predicts relevant items for a user
- Suggesting relevant webpages based on viewing history
- Recommending forms based on information provided
Sometimes, simpler solutions using mature technology can be more effective and less expensive. For example, using computer vision can extract information out of scanned documents, but a digital form requiring manual input can be more accurate, quicker to build and easier to support.
Non-cognitive technology such as robotic process automation (RPA) can also offer excellent near-term opportunities when paired with repetitive, predictable processes such as forms input, scraping data and logging.
In an ideal world, the most efficient solution applies the right approach for the job, which sometimes may be a combination of technologies. New solutions such as AutoML from Google are looking to allow users with limited knowledge to build custom models tailored to business problems, but it is still important to understand the differences and how and when to best apply each for the most effective results.
Ethics: Getting the right balance
While AI solutions enable governments to be more efficient with improved outcomes at lower costs, there is a need to minimise the risks while ensuring that taxpayer money is well-spent. GDS worked with the Alan Turing Institute, the UK’s national institute for artificial intelligence, to produce guidance on AI ethics and safety in a public sector context.
The guidance suggests that organisations prioritise the following four goals, which will function as the ethical building blocks for organisations to innovate responsibly. They should be incorporated in relevant governance framework.
Ethically permissible – always consider the impacts of the solution on affected stakeholders and communities;
Fair and non-discriminatory – bias-free decision making or, when protected classes of individuals are involved, avoiding disparate impact to legally protected classes;
Worthy of public trust – guarantee as much as possible accuracy, stability, security and robustness of the solution;
Justifiable – easily explained algorithms, ensuring that stakeholders understand the design and rationale behind the algorithm’s recommendations.
Another aspect to keep in mind when building AI solutions is data privacy, integrity, and vulnerability, especially in light of GDPR. For example, can the solution be reverse-engineered to show specific confidential data? Is the data secure and protected from external and internal threats? These questions are extremely important to maintain citizens' trust in the use of AI in the public sector.
In a nutshell
AI technologies will fundamentally change how governments work, and the changes will come much sooner than many think . Sometimes, in the rush and the hype of AI, it is easy to forget the foundations required to deliver the technology successfully.
The good news is that organisations can prepare for those adequately and capitalise on AI to improve lives.
At Credera, we collaborate with clients to develop their technology and data strategy required to enable the successful adoption of AI solution. Read our case study below: