Going back to basics to make AI a success

Going back to basics to make AI a success

By Chris Coates, HPC Systems Engineer at OCF

As someone who works with Artificial Intelligence (AI), I regularly hear that: "AI is the future" or

"AI can do anything". But, in reality, customers, vendors, and even engineers struggle to get an AI project off the ground.

I’d like to suggest an analogy, in order to help in the thought process to aid in the likelihood of success.

In my opinion, AI is like building a fire. To make a fire you need three items - fuel, heat, and oxygen. The removal of any one of these things means the fire goes out, or cannot start. Let's take this thinking in order to understand how a good AI project can work.

Data is fuel

The data is the fuel for any AI project. If you don't have enough of it then your project won't last long.

If you have the wrong data or poor quality data then your project may work, but poorly.

For example, you might have some source data that needs cleansing because there are lots of erroneous results. This is often the equivalent of having ‘damp wood’ in that it will work, but not nearly as well as if you had dry wood. The process of cleansing data is the same as drying out wood in that respect. However, no matter how much you dry it, it may just be the wrong fuel and that is something to be very aware of. Get the right fuel for the fire and you get the right data for the project. And it'll work well.

Using the right tools

Finding the right tool for the job is essential. Using the right algorithms and training model is critically important for an accurate AI project. Without these tools, just as a fire without oxygen, it won't work.

You need to make sure you're using the right tools for the job.

The bright spark

The correct tools and good data are nothing without a spark to begin the project with. Just as a fire needs that heat source. A good AI project needs an idea in order to solve the problem. Lots of people expect AI to just be able to ‘solve it’ when they don't necessarily know what they want to solve.

You might find that you have lots of data, but not the right data to help solve the problem you have.

Bringing the elements together

The process of putting this together can sometimes just be a case of asking three questions:

  • What problems do you have in your day-to-day?
  • What data do you have that supports that you're monitoring those problems?
  • What tool is best to help solve this problem?

You might find that during this process, you find that in fact you don't actually collect the data you need to operate on or that you aren't aware that you have the data. Really it all boils down to if you collect it or have the tooling and storage to collate it.

Let’s put the theory into practice

Take a university’s archaeology department for example. They have thousands of physical slides relating to artefacts that have predominantly been hand-written over the course of the last 100 plus years. To categorise, document and organise these manually would take an extortionate amount of time but is needed if the University wants to send any artefacts to museums with the appropriate text or research in order to easily exhibit them. How could this be done?

The questions you would need to ask would be:

  • What do you currently do to categorise items?
  • What number(s) of categories do you wish to cover?
  • Do you have the means of taking high-resolution images of the artefacts?
  • Do you have experts who can aid categorisation in a point-and-click manner on a PC?
  • Do you have the document/research items stored in an electronic format, fully searchable?

If you have answered yes to most of these questions then you could overcome your categorisation challenge. You could simply use a camera to convert images into high-resolution data, so that an expert can verify the categories to feed into an AI model. The same AI tool can then run a character recognition that finds the text in the record inside the document or research and link them into a database.

This model could then be applied to all of the artefacts and linked correctly to the research data and information stored into a database, and any time an artefact is sent to a museum, a copy of the research referenced against it could be sent with it.

This is just one example of a use for AI in a real-world setting, but there are thousands of use-cases. If you're unsure where to start then engage with a specialist such as OCF that can help define a simple approach and support you to deliver a successful AI project. For more information please contact info@ocf.co.uk