Today’s real-world use cases for AI

AI is going to replace us all, or save us. Or both. It’s hard to tell reading the conversations in today’s media. Away from the headlines, more practical AI conversations are taking place. They’re happening in boardrooms, in lines of business and in dev teams as people try to figure out the hype from the reality.  

We decided to address some of the board-level questions we often hear. Moreover, we wanted to address one question in particular: what are some of the real-world use cases for AI? We also wanted to throw in a few case studies to liven things up — like AI helping Sweden produce award-winning malt whisky. We kid you not.

The 101 Ways podcast seemed like the perfect forum for a stimulating conversation on AI, so Oliver Happy (Product Principal) sat down with Jon Parish (Engineering Principal) to discuss the ins and outs of AI and its real-world use cases.

Here are the highlights from the conversation.

What is AI and where did it come from?

AI encompasses anything related to machine intelligence. Your recommendations on Spotify would be a basic example of AI. It’s been around a long time, with the first artificial neural network invented in 1943. 

AI was confined to academia until the 1970s, when it then dropped off the map. No one gave it much thought until the 2010s when processor power and data storage became cheaper and widely available. Then, Silicon Valley picked it up, and boom! The pace of development has exploded in the last few years, making it feel so new. 

What are some AI basics?

There is generalist and there is specialist AI. You can choose one or the other or both. Generalist AI, as Jon made clear, is provided by services like ChatGPT. It can give staff AI assistants to summarise documents, write content, translate content, and analyse data. 

Specialist AI is more tailored to a business’s specific needs and uses a company’s data. An example of a specialist AI application could be your typical recommendation list for website visitors. 

Companies must ask themselves a few questions before deciding what kind of AI to pursue. These include: what data do we have? In what quantities? And covering what period? For AI to work well, it must process a lot of data that stretches back years. 

Furthermore, data for specialist AI should be “cleaned”, meaning labelled and correctly categorised for AI to work its magic. The cost of cleaning data post-storage is hefty; therefore, businesses should ensure data is AI-ready from the moment of ingestion.

How long before there’s any ROI?

Accumulating years of AI-ready data requires investment and time. Hence, it seems reasonable to assume AI is a long-term investment. This is true, but there are short-term gains to be had, said Jon. Companies can quickly extract insights from their growing data using data analytics systems. Plus, they can use the time before AI comes on stream to integrate data into their decision-making processes and become data-driven.

What are some AI use cases?

Returning to the theme of generalist AI, we all know of people using services like ChatGPT to successfully summarise reports, write emails, and support human-like online chats. Many companies use generalist AI services to support sales and marketing, finance and accounting, and customer care. 

Businesses active in specialist AI are using the technology to support product development. Jon used the example of photo apps for smartphones. The technology allows users to make significant changes to their photos – like removing a person – without leaving any trace to the naked eye. What amazed Oliver (unsurprisingly) is that a Swedish company created a specialist AI model to figure out how to create an award-winning malt whisky. And it worked!

Does AI come with small print?

There are some critical issues to consider with AI. A major one is “failing silently”. With AI, you never get an error message. It will always have a go; everything may look like it’s working. But it may not be. Think about it this way: your AI summarises content, and the result is accurate 90% of the time. That means a 1 in 10 error rate. Isn’t that high? The errors have to be identified and their frequency assessed.

Also, generalist AI is only suitable in some fields. As Jon and Oliver agreed, the language used in law and healthcare, for example, is specialised and requires expertise. Hence, organisations in these areas would want to use specialist AI rather than generalist AI to underpin their AI assistant applications.

What approach should I take to AI?

Start by looking at generalist AI services that you can sign up for and use. It may not transform how your business is run overnight, but it could provide employees with an AI assistant to help them work more productively. 

Jon confirmed that a serious approach to AI starts by building a data foundation from which you can unlock value. That foundation includes an effective way to store and manage data so that it can power AI applications. Discussions around training and data governance also need to happen. 

What’s most important is that companies begin to familiarise themselves with AI technology and terminology so they can start having productive conversations about the best way forward.

Looking for insights on real-world AI use cases? We can help. To hear more from industry experts like Jon and Oliver, check out our Team Takeover podcast series.