Digital Transformation for SMEs — How easy is into venture in to AI & Deep Learning?

I really hate this — Being carried away by the trends! But there is no alternative in today’s digital world. While we have seen a clear path that a Small and Medium Enterprise or Business can take with the cloud technology, it does not seem to be an easy go when it comes to AI & Deep Learning yet. It seems to be little far away from practicality.

I am making this statement after attending the online conference by Amazon –AWS Innovate Online Conference Special Edition — Machine Learning. And here are top 3 reasons why I am making these conclusions.

Reason 1: Quality and volume of data is core to AI

The success of AI depends on quality and quantity of data. For two reasons one lack of proper business data management practices and lack of proper volumes of data, AI might not be your cup of tea.

So, if an SME does not have a structured approach to managing its data, this might not really help as it might be really difficult to have a quality dataset.

Training data sets are the core to training any AI. And AI generally needs large amounts of similar data to really bring in the human way of reasoning, planning, and execution. Further, it is still unclear on what would be the sample sizes required to train any AI algorithm to let us say it to provide intelligent answers to you.

As the above slide shows, its all about data and AI is all about answering business questions with data. Image Credits: Amazon AWS

Reason 2: AI software stack is not easy to plug and play

You need to have Machine Learning practitioners and experts to handle AI & Deep Learning projects. It is not still a readily available service, like a cloud storage service, or data analytics service where you can do a plug and play work.

And AI offerings are not still so matured, where you throw them any folder or any database from your business data and expect magic to happen. The technology stack available in the market is not fully autonomous.

The broader classification of machine learning stack currently easily available in the market is shown below. Each of the stacks shown below needs AI experts and developers to build AI models, Train the model, Test the model and Deploy the model in a business environment.

Broader classification of Machine Learning Stack available in the market. Image Credits: Amazon AWS

The frameworks and interfaces aspect is the core and is aimed at the core AI & Deep learning practitioners. The number of such experts available in the market seems to be a challenge. This is an area, where we will see more experts coming in in the future, not immediately though. So, for an SEM to afford such highly skilled resources at this time would be a real bottleneck.

Even if you want to have your existing set of IT experts and developers to think of entering this area, it is not that easy. A set of very popular AI & Deep Learning Frameworks is shown below. How many of these do your everyday experts working on say- .NET, JAVA or DBMS really know these?

Most Popular AI & Deep Learning Frameworks — How many of these, your existing developers and IT experts know? Image Credits: Amazon AWS

And again every framework at this time has a specialty, some might be good at processing images and some for textual data, and expecting your developers to gain be experts in all these is a too audacious expectation.

And most importantly, the AI application services readily available for developers to plugin and play are also really limited. The currently available services are broadly classified at the vision level and language level. And the function of these is basically to recognise images and videos; and comprehend, transcribe or translate textual content. And most of these service offerings are at early stages.

Reason 3: You really need interdisciplinary teams with different set of skills

Because at this point of time Machine Learning is heavily interdisciplinary. It seems that having just an IT Architect in your team does not really suffice towards having a successful AI implementation.

As machine learning sometimes needs to work with sensor data, a core mathematician may even not come to your rescue as I understand. So you will really need to have the following four set of people in your AI Team — a Physical Scientist, an Analyst, a Computer Scientist and an Architect.

It takes a lot of cross coordination among various high-level skill sets to lead a data science team that will drive AI projects. Image Credits: Amazon AWS

Can an SME or SMB even think along these lines? Do they have the right set of vision to scout for the right set of talent — that neither the organizational heads or HR teams barely know about. How do we even work towards composing a right set of team, based on our business problems? Again, it is too scientific and technical when it comes to such skill sets that will scare away any SME type organization.

Looking at these three reasons, I think it is too early for any SME to even think of venturing into the space of data science to drive AI or Deep Learning projects.

Let me know your perspective on this!

P.S: While the images have been taken from Amazon AWS presentations, the views here are purely mine.

Source: Deep Learning on Medium