Artificial Intelligence (AI) is everywhere at the moment, from the Cambridge Analytica scandal to the UK’s announcement of nearly £1 billion for AI sector development. Here at Methods we think AI is an exciting new technology which, if we think creatively enough, can be used to transform government services. We also believe that AI is for everyone, and the more people who get involved in AI the better it is for society overall.
To get AI to work, you need as much data as you can lay your hands on. It is the lifeblood of AI. There are a number of open source datasets online that can get you started, or in the case of applying Machine Learning to an existing service (i.e. automating an onboarding process) all you need is the data you currently have collected through this service.
A word of warning – in the current era of GDPR and Cambridge Analytica, the way that you collect and use data to power your AI system needs to be carefully considered. There are many ethical and governance issues that need to be thought through before you just throw a bunch of data at an AI system.
These are only a few questions to start with and hopefully they’ve made it clear that it’s not enough to select a single dataset and go forward without interrogating it and making sure it’s appropriate to use.
There’s a misconception that implementing AI solutions means that any previous issues around people and process will be resolved by the software. Unfortunately it is not that simple. AI solutions may reduce manual processing work, or speed up work that would be time-intensive if conducted solely by humans. At the same time, for the AI solutions to work and to deliver any value, they require ongoing oversight by humans (to make sure they’re producing the right results) as well as maintenance and software support (to make sure they’re not broken). Putting in place a multi-disciplinary team that is resourced to provide these functions will set you up for success!
AI systems are generally implemented into existing networks or series of processes. The AI will need to form a productive part of this business process network, and so the design and specification for the AI system will have to understand and incorporate the requirements of the network. This includes having a process for managing the AI system if it gets something wrong – which could span a customer complaints process or a risk and audit function.
The best next step for any organisation thinking of implementing AI is to just get started! Each organisation is at its own stage of maturity when it comes to digital transformation and so understanding your organisation’s own position is key. For some organisations it will be worth investing in an onsite team that can build and test your own AI solutions, and for other organisations it will make more sense to procure an existing product or service on the market.
If you still have onsite servers and the data collected across your organisation sits in isolation of itself (i.e. in its own spreadsheet), then looking at cloud-based solutions to house all of your data in a single location is a good first step. From here, it becomes easy to automate a simple process – enabling you to then build in more sophisticated techniques like Machine Learning to improve the process, or automate more and more of the business processes until it’s automated end-to-end. These are the building blocks of then implementing more complex AI techniques or apply the techniques to more complex problems or processes.
A final tip – when you do get started, involve everyone in your organisation. Customer service people, policy people, operations people – they will all be impacted and, more importantly, their perspectives will help make sure the AI system that you implement will work for the whole organisation.
This article was originally published here.