2020 - The Year of Industry AI Solutions

The various analysts talk about the size of the AI Market being in 10's of Billions for 2020 and beyond. My frustration with this is that I don't get clarity around what that really means. 

For the last couple of years or longer the AI market has mainly been consumer driven (home smart speakers and other similar devices), chatbots or demos and Proofs of Concept. In my view 2020 has to be the year of AI solutions, delivering real value to organisations and I can see that starting to happen across the board which is important for many reasons, one of which is the danger of AI fatigue.

AI fatigue is my way of describing the dip that you see in all the Gartner curves (Trough of Disillusionment) but specifically for AI. Why is this different from any other area? Mainly because at the moment many organisations see Datascience as a "dark art" and difficult; which impacts on their view of AI. Also when an industry starts to consider AI, such as healthcare, (quite rightly) there are trials and regulations that have to be met - this takes a long time. So why do I believe that 2020 is the year of AI solutions? Two main reasons:

1) We are now at a point after a few years where some of the trials have been done and where AI models are being used to augment skills - even in healthcare - such as radiologists, to speed up diagnoses. If you look at manufacturing and retail, they are now starting to implement some of the usecases they have been investigating over the last year or two. In short, industry has gained confidence in AI.

2) We are now able to do much more with the tools available as subject matter experts. Tools such as IBM POWER AI Vision allow the business users, the real experts, to create their own models fairly simply. This does not negate the requirement for developers who need to turn these models into applications, or datascientists who will still have to work with data and understand it to create further models that will enhance the "auto" AI models. We are now also seeing the emergence of edge applications that again are user friendly and quick to deploy; an example of this being Visual Inspector .

I am enthusiastic about what 2020 will bring and that we will see more products on the market that bring AI to the edge where it is all happening.

Rather than just type about it I thought I would post a video to illustrate my points above. I used the tools we have as part of our IBM Systems AI portfolio to work with the data, create models and deploy on an edge device.

You can only really understand a product if you use it. So I decided to use Visual Inspector (IBM'S iphone App for running #AI models at the edge) to create a simple demo, simulating "widgets" (actually bulldog clips) on a production line. To get to that point I had to first use POWER AI Vision to take the data (collected using the VI app on the Apple iPhone), label the objects, train a model, create more data in PAIV using data augmentation, run the model to auto label the objects (so much easier!) and then connect up my iPhone. This sounds like a big job but actually I did it all in less than 3 hours and most of that was the labelling! 

You can create a CoreML model to run locally on the iPhone with no connection to any server, however, for my demo I connected to an IBM AC922 server sitting in one of our lab datacentre rooms - the one I had done the training on.

The ease by which the model could be deployed and the confidence of the results was impressive. You will notice that the model not only identifies good and bad, but the colours, so if I wanted to check that I had the right number of Yellows, Blues, etc. I could do this also. The video leads on from after the model has been built and deployed.....

#ml #dl #aisolutions