Artificial Intelligence (AI) is very much a reality today. No longer confined to the realms of science fiction, hi-tech companies, AI is being adopted by companies across various sectors. In my line of work, I deal with AI every day, as I bring this technology to work for our clients.
One of the biggest advantages of AI is that it better enables and extends the capabilities of a host of other technologies in use today. For instance, Big Data has been around for a long time now and yet, in my opinion, it’s benefits have been somewhat difficult to access. Theoretically, Big Data technologies are supposed to allow companies to make sense of vast quantities of unstructured data. Such data can be in the form of documents, images, videos, audio files and more. Yet, how many organizations are truly using the information contained in such data? For example, I have had the ability for a long time to analyze images but rarely have I implemented a solution to do this.
AI helps to solve these problems and they are much more accessible. AI services come pre-packaged with your Cloud subscription (I assume that every company is using cloud services now). All the main Cloud providers provide AI solutions to analyze images, video and sound (and more). Most, if not all, even provide pre-trained services which you can access immediately. If you need to train your own model, they will even provide the services to do that too.
For one of our Insurance clients recently, we built a Proof of Concept (PoC) to identify a car from a picture of the license plate. Once we identified the car, we could find the value. We then passed the information to a model we developed and trained to test for an economic write-off. Another client, in the transportation sector, is interested in the travel of passengers on their transport services, so we are developing a proof of concept to count people using a camera, against bus stop GPS co-ordinates.
To develop these solutions, you require lots of existing data. For instance, training a visual recognition model required lots of annotated images. If you are looking to find car write-offs in images, you need lots of images of written off crashed cars and lots of images of cars which are not written off. If you are looking to identify the car from the registration plate, you can simply access a pre-trained AI Text Detection solution from a Cloud provider. Accessing the services from your app is achieved through a simple API, using your favorite programming language. (For added value, why not analyze lots of images using a scene detection service, log the results into a database and then add a search utility for Natural Language Processing? This can then be connected to a chatbot service to allow users to ask for images of objects – “Show me images of damaged motorbikes”).
So how do you get started?
• Not with the technology! Talk to your business and decide a business prototype (Sogeti uses the Thinkubator approach for this).
• Access your Cloud subscription to see what pre-trained services are available. Avoid having to train your own model if you have too.
• It is also worth testing different Cloud services. In one service, my 10-year old son was identified as a 40-year old man.
• Also, the Cloud service providers obviously have to train these services for you. Remember, continued training and retraining can change results. Another Cloud service I used for Invoice Text Detection changed their model and stopped recognizing the testing data I had used for some time.