This is the third in a series of three articles that provide practical considerations for organisations submitting an application to the AI Challenge. This article focuses on how to determine the feasibility of AI uses cases.

Related articles in this series: 

Part one: becoming familiar with that AI can and can’t do 

Part two: Identifying use cases

Data access and quality

When it comes to using data, it is useful to consider access, quality, and types of data.

Access

What kinds of data will your organization use for an AI use case? How will you get this data? Do you have permission? What data could you easily access for a Proof of concept? For the initial proof of concept, testing with fake or anonymised data is fine.

Types

Do you have an understanding of the expected volumes of data and types of data involved? We would only need a small data sample for an initial proof of concept.

Quality

Data quality is central to all AI applications. Do you have an understanding of the likely data quality of your suggested data?

Automation suitability

Not all tasks can be automated. So, it’s important to consider whether the use case really is right for AI. Some starting points for when AI is not appropriate are listed below.

  • If it involves decision making for high stakes areas, then it should not be undertaken by AI. AI could instead help triage information to support decision making. For example, in medical, policy or legal decision making.
  • If it involves high context variability and where human judgement is important then humans are likely to be heavily involved. For example, where case workers may need to consider a highly nuanced and contextualised circumstances in order to process information. AI can still play a role in helping triage information or reduce admin burden, If it involves high social aspects or physical components then AI is not suitable. For example, social work and teaching are examples where human interaction is highly important.

AI ethics

Ensure AI is used lawfully, ethically, and responsibly to all stakeholders affected and ensure you are able to comply with your own company policies as well as following responsible AI principles which are outlined in Scotland in a National AI Strategy.

Data protection and security

Public sector data can contain sensitive and personal information. It must be processed lawfully, securely, and fairly at all times. For example, if your idea involves processing personal data, you need to consider how to protect it. You must also ensure that you comply with General Data Protection Regulation (the GDPR) and the Data Protection Act.

Effort

What is the likely effort involved in implementing the use case? How does this relate to the value it will bring to the organisation and its citizens. Some aspects to consider may be:

Models – Do you have any broad assumptions around the AI models which may be utilised, such as pre trained models available via APIs or custom models?

Domain specific AI – If your tasks require highly specialised knowledge or domain-specific responses that general-purpose models cannot adequately address, training on your own data may be necessary.

Systems integration – Does your AI solution require integration into other systems and how feasible will this be?

Skills and experience – Do you have the necessary skills in house or do you need to partner with suppliers to develop and manage an AI solution?

You may wish to assess of your use case ideas, which could deliver the highest ROI compared to the projected effort involved in delivering it.

Sustainability

It is also worth considering how you may track the environmental impact of AI solutions. This includes data from your cloud provider.

In summary, there are a number of factors you may wish to start to consider at an early stage when looking at AI use cases. However, developing a small-scale proof of concept (PoC) is also a good way to test your assumptions around feasibility quickly.

Visit https://www.aichallenge.scot to learn more about the AI Challenge and how to apply.