The widespread application of Machine Learning and Artificial Intelligence marks a secular economic change akin to electrification or the containerisation of shipping, according to Richard Sargeant of ASI Data Science. Organisations that set clear objectives to invest in developing skills and creating the right conditions stand to leverage the greatest benefits.
In the fourth and final of Freshminds’ 2018 Digital Transformation breakfast seminars, Richard offered examples from his own career in AI to emphasise the significance of this change. These included how machine learning was deployed at the Home Office to differentiate extremist material from pop videos on social media, and how it was used to improve passenger flow and disembarkation times at the world’s busiest airport. The programmatic ability to differentiate Taylor Swift from the Taliban, or to accurately predict when an aircraft is likely to be ready to re-stock are very different challenges, but share the characteristics of needing to be done at speed, at scale, accurately and with consideration to freedom, safety and comfort. This is where AI comes in and presents an emphasis on how machines can help to solve very human problems.
Before delving into the detail, Richard sought to put some clear definition around what is (and is not) meant by “Artificial Intelligence”. AI stands distinct from statistics and analysis due to complexity, the algorithmic ability to learn, and the creation of outputs that are typically consumed by other data services. AI most commonly exists as an analogue to human intelligence, and as a set of tools to make complex predictions, prescriptions and decisions effectively.
To deliver a successful AI project, Sargeant argued, you need the right raw materials and the right conditions.
The right raw materials
Raw materials include data, the skills within your team, and the core tools to create useable outputs.
The data should be identified through a data audit, establishing a catalogue to itemise what data exists, how it can be accessed and who controls it. For projects with limited data, reinforcement learning algorithms can assist by iterating around a stated goal (for example in running dynamic pricing experiments).
Skills should be built across teams, and not within “heroic individuals”. As a starting point, teams should include strong mathematical and statistical capabilities, data engineering, the ability to code and the stakeholder management skills to communicate effectively and gain buy-in from stakeholder communities (see our earlier article on the rise of the data translator). Teams should probably include a data scientist (or two), a business analyst, a data engineer, and a designer.
The tools required are distinct from the tools required of analysts or developers and should include a programmatic access mechanism to get at the data, and a platform on which to model data effectively.
The right conditions
With these raw materials in place, your AI project now requires the right conditions to thrive; the optimal Petri dish environment to grow and take hold. First is an identifiable crisis or burning platform? What is the crisis that this project will address? If there isn’t one, and you can’t create one, then this might not be the right project. Secondly, and closely linked, is the need to galvanise senior leadership behind the project. You will need them to support with resource, focus, sponsorship and access. And lastly is a realistic scope and control. Projects should be kept focused, and small.
Conversely, projects that are likely not to succeed can be characterised by having poor data (or access to data) and unrealistic expectations at the outset, and wide-reaching briefs lacking in corporate ownership.
Looking to the future Sargeant pointed to a commonly held view that increasingly powerful and complex algorithms will become more general in their application, evolving from the highly focused functions most common currently. This development would, he predicted, require an ever greater ability for humans to communicate the value, capability and limitations of AI and to build and maintain the trust of end-users actively.
And for those businesses that don’t build capability, invest in skills and look for the angle for greater automation and augmentation? Well, they are probably not businesses you want to invest too heavily in for the longer term.