“TensorFlow is by far the most popular tool among our respondents, with Keras in second place, and PyTorch in third. Other frameworks like MXNet, CNTK, and BigDL have growing audiences as well”
As if businesses today didn’t already have enough to worry about, then along comes a new wave of game-changing technologies that they must master quickly if they are not to fall behind their competitors, with pressure mounting to start using AI. .
Artificial Intelligence is the most visible of these technologies – and arguably the most important. Open a newspaper, and it might seem as if every business is making great strides towards developing and using AI applications that will transform their operations and enable them to deliver new products and services to their customers.
It’s easy for businesses yet to achieve success by using AI – or even to get started on their journey – to get despondent about the lead they perceive their competitors to have. They often see AI in absolutist terms: you either have cross-organisational, fully automated systems, or none at all.
But AI isn’t a binary – it’s a spectrum: one where successful applications are built on a platform of smaller, successful projects, which themselves were the result of trial and error. Rather than betting the farm on rolling out AI across the enterprise as quickly as possible, it’s much more effective to experiment with initiatives that deliver real benefits on a smaller scale.
That doesn’t change the fact that there are several obstacles standing in the way of successful AI projects. None of these is insurmountable; nevertheless, organisations must understand what difficulties they need to overcome in order to develop and deliver projects that solve real business challenges.
Addressing Obstacles to AI
Earlier this year O’Reilly asked over 3,000 business respondents about their preparedness for AI and deep learning, including their adoption of the necessary tools, techniques and skills.
Of particular note is an AI skills gap revealed in the survey. A paucity of talent is seen as by far the biggest bottleneck for successful AI projects, identified by a fifth of respondents. This is an especially big issue in AI projects, since building such applications from scratch relies on end-to-end data pipelines (comprising data ingestion, preparation, exploratory analysis, and model building, deployment, and analysis).
It’s not just technical talent that enterprises need, though. They also require people with the business acumen to make strategic decisions based on the data and insights that AI provides.
Deep learning remains a relatively new technique, one that hasn’t been part of the typical suite of algorithms employed by industrial data scientists. Who will do this work? AI talent is scarce, and the increase in AI projects means the talent pool will likely get smaller in the near future. Businesses need to address the skills gap urgently if they are serious about developing successful AI initiatives. This will likely involve a mixture of employing outside consultants and developing the necessary skills in-house – for example, by using online learning platforms.
To be fair, most businesses in our survey (75 per cent) said that their company is using some form of in-house or external training program. Almost half (49 per cent) said their company offered “in-house on-the-job training”, while a third (35 per cent) indicated their company used either formal training from a third party or from individual training consultants or contractors.
The other side of the coin – the business rationale for AI – requires management to identify use cases and find a sponsor for each specific project, ensuring there is a clear business case that is served by the technology.
Using AI: Don’t Forget the Data
Another key challenge to successful projects is ensuring that the data used is completely accurate and up-to-date. Machine learning and AI technologies can be used to automate – in full or in part – many enterprise workflows and tasks. Since these technologies depend on pulling information from an array of new external sources, as well as from existing data sets held by different internal business units, it’s obviously essential that this data is properly labelled.
The first step in this process is to establish which tasks should be prioritised for automation. Questions to ask include whether the task is data-driven, whether you have enough data to support the task, and if there is a business case for the project you plan to deliver.
Enterprises must remember that while AI and ML technologies can work “off-the-shelf”, to get the most out of them requires them to be tuned to specific domains and use cases, perhaps involving techniques such as computer vision (image search and object detection) or text mining. Tuning these technologies often – essential to delivering accurate insights – demands having accurately labelled large data sets.
Designating a Chief Data Officer is key to solving the challenge of accurate data. A CDO is responsible for thinking about the end-to-end process of obtaining data, data governance, and transforming that data for a useful purpose. Having a skilled CDO can help ensure that AI initiatives deliver their full capability.
Harness the Right Deep Learning Tools
Returning to our research, three quarters of respondents (73 per cent) said they’ve begun playing with deep learning software.
TensorFlow is by far the most popular tool among our respondents, with Keras in second place, and PyTorch in third. Other frameworks like MXNet, CNTK, and BigDL have growing audiences as well. We expect all of these frameworks—including those that are less popular now—to continue to add users and use cases.
These deep learning libraries are all free and open source, and are used by most AI researchers and developers. Once you start exploring these tools, you’ll quickly find that AI is a very open and collaborative community where people share papers and code; it makes sense to use these frameworks when assembling your internal AI teams.
If businesses set the right priorities, invest in training, make the best use of external expertise, and concentrate on the quality of their data, then there’s every reason to be confident of making a success of their first forays into the world of AI.