Avoiding failure of your AI project. Key take-aways from the Internet of Things World Forum, May 22-24, 2017 – London

Trudy Mockford, CEO, CloudMaitre

What a remarkable occasion it was at Tobacco Dock, London, to be able to hear what Industry Leaders have to say on current and forthcoming use of technology in the Internet of Things space. Key learnings from the exemplary session entitled “Automation to AI. From Devices to Insights to Action” presented by Kevin Bandy, Chief Digital officer, Cisco along with Michael Demshki, Director, Strategic Business Development, Intel.

This discussion style review session on the topic threw up plenty to consider and summarise. Comparing Classic Machine Learning in simple terms, to up and coming Deep Learning. Classic Machine Learning relies on a data scientist or a team of such, to identify data features, apply algorithms and then train the system – possible with tabular and structured data. Progressing to Deep Learning, where the more data you give the system, the more it learns – and applicable where your data is made up of speech, video or images.

The Shape of Your Data

According to Michael Demshki, Director, Strategic Business Development, Intel

“80% of Project Time is allocated to the preparation of the data and data mis-structured prior to the project is a key cause of failure”

Fundamentals of Planning

AI failures have often happened due to over engineering of business outcomes and of business data. At the very outset of scoping this type of project, suggests Michael Demshki, there are 2 essential points for consideration here,

“what problem do you want to solve? and What specific questions in the business do you want to answer?”

The learned approach is to provide an incremental solution to the client. To be later refined.

Specialist Roles and Lack of Talent

The backbone supporting the potential success of any Device Automation and AI project both the Data Scientist and the Business Transition Specialists. Prioritise on-boarding or developing the best of this talent. Today a talent pool that often falls short both in numbers and standards,

Case Studies

Top-line examples of application of these new technologies.

Smart Health

IoT to provide personalised patient care. There has been a notable slow adoption in the healthcare sector – a large part of this has been establishing with health care professionals that this new wave of technology isn’t about telling a Doctor what to do with a patient; but enabling him (or her), to assess and decide more effectively what care to provide to the patient. There are many examples of Smart Healthcare applications and only few of successful implementations.

The case study that we heard about in this session was a great example of a successful project. Taking something that we do well in healthcare: Early Diabetic Retinopathy Detection and then applying IoT allowing the commissioning healthcare provider to take camera devices to remote communities and extend the preventative programme. The execution side of the diagnostics can then be re-deployed to other regions.

Agricultural Robotics

This particular case study from Blue River Technology took the standard scoping statement for a technical project to another level:

 “if we can take a repetitive manual task and explain and map it, we can automate it.”

The task was around automating crop care. The training data was applied to create a deep learning algorithm to identify where a crop stem meets the earth.  The data was in the form of images – and to train the system tens of thousands of accurately labelled images were needed. In fact, organisations providing the labelled images as a service have entered into the eco-system surrounding AI.