Implementing an AI Strategy
“A vision without a strategy remains an illusion.”
— Lee Bolman
THE GOAL OF AI should be to take human ingenuity and to attach a rocket to it, blending technology with ethics, accountability and inclusive design to empower as many people as possible. AI should benefit society, not dehumanize it. That’s why it can help to think with a “humans-first” approach. If it’s not adding value to humans, you have to ask yourself why the AI exists in the first place.
To get started, I recommend following this simple five-step process:
- DEFINE THE STRATEGY: Identify what you hope to get from AI and how you plan to implement it, as well as any resources that will be required along the way.
- RECRUIT AND TRAIN: This stage involves hiring the staff you need to make your vision a reality whilst simultaneously training existing staff to take advantage of new systems.
- OPERATIONALIZE: Prepare to launch your new AI-based systems by making them a part of your operations. Ensure that all of the systems, tools and processes are in place and understood by all employees before they go live.
- DEPLOYMENT: This is the stage at which your new AI-based systems go live and begin supporting physicians and serving patients. This is where all of the hard work starts to pay off and to deliver results.
- OPTIMIZE: One of the main benefits of using digital technologies like artificial intelligence is that they can provide metrics and analytics that are designed to help you to further improve your systems.
Define the Strategy
With this structure in place, you’re ready to start developing your AI strategy. Remember that the strategy that you work on now will essentially form the foundations of the company that you’ll build. Developing your strategy is one of the most important things that you can do and so don’t be afraid to invest your time into it.
By now, you should have a good understanding of the opportunities that AI has to offer. The next step is to identify your focus, which is the technical way of saying that you need to find solutions that marry human ingenuity with the smart use of technology. Perhaps you’re learning from imperfect data through the smart use of natural language processing. Perhaps you’re using AI to interpret text, voice and images. The strategy that you develop will ultimately be informed by what you’re trying to achieve.
Recruit and Train
When you’re hiring staff and developing an artificial intelligence team at your healthcare company, you need to consider what kind of talent you need. Tapping into existing APIs and using them to develop new healthcare-specific use cases is one thing. Developing complex custom modelling software is something else entirely. That’s why you’ll need to have a good idea of what’s needed before you start inviting people for interviews. There’s no point building a team if its expertise doesn’t tie in with what you’re trying to achieve.
Custom modelling often calls for machine learning solutions, which despite being a subsection of AI still requires a very specific set of skills. No, not like Liam Neeson in the Taken movies. We’re talking about the ability to create machine learning algorithms and to understand data collection and analysis, as well as the rules and regulations that they’re required to follow. Healthcare industry experience is always a plus because, like the finance industry, it’s heavily regulated and bound to a different set of criteria than many other industries. On top of that, industry expertise tends to lead to a better, more bespoke piece of software.
For most companies, the first step towards operationalizing their approach to AI comes in the form of bringing older, legacy applications up-to-date. That’s because it’s often easier to focus on one tool or one process at a time than it is to try to change too much too quickly. The problem with this is that the ultimate goal is to transform your business processes and to reach full maturity as an AI-based healthcare company.
Still, there are plenty of opportunities for AI in the healthcare industry, as is evidenced by Tractica’s 2017 Artificial Intelligence Market Forecasts report, which listed patient data processing and medical image analysis amongst its top ten use cases. The key to actually deploying these new AI-based systems is to make them a part of your operations and to provide training as appropriate so that everyone’s ready to use them before they ever go live.
To understand how AI models are deployed, you first need to understand how they’re created. The goal of a model is typically to identify the relationships between input data and historical outcomes in such a way that it can start to draw conclusions.
For this to happen, the data must be gathered from its sources and prepared for the model. This may require a data specialist who can clean it up and remove any duplicates, a task which can take up to 80% of the time required for the entire development and deployment process. With this complete, the model can be built and a subset of the data can be fed into it. At this point, the model starts to evolve as it makes predictions based upon the input and then evaluates how it performed. It can then “learn” to do more of what worked and less of what didn’t.
Once the model is ready to be rolled out, we enter the deployment stage. Here, the focus shifts from data scientists to developers, whose role it is to take the model file and to build it into a user interface that’s easily accessible for end users.
No model is complete once it’s been deployed. Further testing is always an option, and you can take what you learn from the tests to further optimize your software. On top of that, software can be upgraded or kitted out with new and improved functionality.
One example of this is Microsoft’s Speech API, which uses artificial intelligence to convert recordings of human speech into text. The problem is that while it might work well for conversational, everyday speech, it’s going to struggle converting a doctor’s notes into electronic health records (EHRs). Using the API helps to keep costs low and to cut corners, but it’s not necessarily the best solution in the long run. Better the API than nothing, though.
Another good example of optimization is moving your data to the cloud. When your data is hosted in the cloud, it can be accessed by any validated user on any compatible device, allowing staff to access real-time data in a much more convenient format. The cloud works particularly well with AI because you can upgrade your account and add resources as needed.
Want to learn more?
I talk more about artificial intelligence and its impact on the healthcare industry in my book, The Future of Healthcare: Humans and Machines Partnering for Better Outcomes. Click here to buy yourself a copy.