machine learning – Robin Riley

What it was:

Talks by subject matter experts on AI hosted by the Civil Service Alumni, at the Royal College of Defence Studies, London on 18 December 2017

What I learned:

AI is permeating business:

No single agreed definition of “AI”
‘Narrow’ AI does specific tasks, as opposed to ‘General’ AI which can solve new problems
AI is already in use in hundreds of Google products
AI is increasingly in the plans of the forward-looking global industrials
Huge landscape of firms involved in different sectors

Deep Neural Networks:

The workhorse of modern AI is the deep neural network, where each layer of neurons is connected to the prior layer
Supervised versus unsupervised learning
AI performance is better than humans in many narrow case, using supervised data and deep neural networks

Conditions for success:

Many AI applications can be characterised and understood as input X leading to output Y to solve problem Z. For example, medical scan data leading to disease symptom recognition to solve diagnosis.
A precondition for the successful deployment of AI is a clear definition of the system or process, and an understanding of what AI can do
A social technical approach, including human impact, is typically desirable
AI typically requires widespread digitisation of the system
Data requirements: Common definitions, high quality, accessible, sufficient size
Domain expertise is key

Impact on Government?

A viable, shared business model for both public and private sector is needed
Key challenges for government in terms of capability, ethics, regulation, getting on front foot in terms of impact, and in nurturing the wider AI economy
There is a significant discussion happening now around ethics, bias, regulation

What I will aim to do differently as a result:

Robin Riley