Laurent - stock.adobe.com
Improving talent attraction and retention and adopting better approaches to data are core aspects of making artificial intelligence (AI) strategies work, delegates at a Computer Weekly leadership roundtable heard.
Jonathan Midgley, director of engineering at Trainline, and Daniel Hulme, chief executive at artificial intelligence consultancy firm Satalia and director of the business analytics master’s degree at University College London (UCL), discussed the impact of AI in technology decision-making and shared practical advice at the CW500 Club, Computer Weekly’s regular networking event for IT leaders, held in partnership with Mortimer Spinks in London.
Discussing how AI will transform technology leadership, Hulme kicked off with his views on the “weak” and “strong” definitions of the concept. The most popular way to describe the technology is “weak AI”, he says, whereby computers can emulate narrow human tasks – but the most interesting definition is related to the emulation of increasingly more human cognitive functions.
“Humans are very good at finding patterns in about four dimensions and solving problems up to about seven, while computers can find patterns in thousands of dimensions and solve problems with thousands of different things moving apart, so benchmarking machine intelligence against human intelligence is the wrong thing to do,” he says.
“A better definition of AI is around goal directed adaptive behaviour, where you try to achieve an objective, such as selling as many ice creams as possible, and move in a quick and frictionless way towards that goal – when you have those two things, you have automation, which can provide a lot of value,” says Hulme.
He says the word decision-makers need to pay attention to when it comes to AI is “adaptive” – whereby a system learns whether a decision is good or bad, adapts its own internal model, then makes a better decision next time. However, the expert notes that such systems are still hard to find.
“I haven’t seen a single successful enterprise system in production that does that yet,” says Hulme. “No IT manager ever wants to hear that a system will behave one way and tomorrow in a different way. I would argue the true paradigm of AI systems [lies in their ability to] learn and adapt in production. But these systems are very hard to build.”
Solving optimisation problems
AI subsets such as machine learning can drive a lot of value in organisations, but companies need to know what they want to use it for. Hulme says that while machine learning is useful to find patterns in data, it is not ideal for solving optimisation problems that involve a decision element.
“I would argue that companies don’t have machine learning problems, they have optimisation problems, so they need to ensure they are harmonising machine learning with optimisation, which is about making decisions using the insights that machine learning provides,” he says.
Attracting talent with interesting AI plans
There are a number of considerations around systems architecture and ethics associated with building systems able to adapt and make decisions. However, with increasing commoditisation, the technology will not be the main point to worry about.
Daniel Hulme, Satalia
“Over the next decade, [AI] technologies will become commoditised,” he says. “Someone will do the optimisation, someone else will do the self-learning aspect, and we will have free access to computing power. We already have a lot of tools required to find patterns, which are all open source.”
The big battle ground and the future for IT leaders is not technology, but how they attract, motivate, retain talent to innovate – if an organisation is not innovating and adapting to the world quicker than its competitors, the competition will win.
Hulme says the companies often try to hire his students at UCL but invariably fail to do so as the problems they want to solve with AI are “not sexy enough”. He says that having a clear plan with motivating projects is key to attracting talent and “making it stick”, as is adopting some key organisational principles.
“Lots of investment is going into using machine learning and AI to sell more products and services, as well as optimisation to reduce costs. If you’re not making effective use of these technologies to change how your organisation operates, you are losing talent and won’t be able to recruit – you are probably dying from the inside,” he says.
“Once you’ve figured out how to attract talent, whether it’s by building your own AI teams or working with third-party suppliers, you then need to motivate that talent,” says Hulme.
Autonomy, mastery and purpose
Autonomy, mastery and purpose are the three core elements to effective motivation and successful AI projects, the expert notes – so giving people the freedom to do what they want, the ability to become really good at what they want to do, as well as providing a goal they can align themselves with.
Considering the impact of AI on jobs is another point that leaders will need to unpack further – both in terms of what happens to their IT functions and in the entire organisation – if they are to use the technology to their advantage, says Hulme.
“We are freeing people up from tasks, but at some point, we will be freeing people from their whole jobs as we’re driven to reduce costs and provide shareholder returns, so one way of doing that is by removing labour,” he says.
“But companies won’t be able to retrain fast enough for the new jobs. The economy is not ready for job loss at this scale and IT leaders need to be thinking about addressing some of these challenges.”
Drawing on the technological singularity point of super-intelligence, Hulme expects that in 30 or 40 years, an artificial brain that is “smarter than humans in every single possible way” is likely to be a reality – and the technology profession needs to get ready for it.
“[Super-intelligence] will be the last invention that humanity creates – and some people think it’s the most glorious thing that will happen to us, while others think it’s our biggest existential threat,” he says. “If we are not cooperating as a species over the next three decades, this thing will most likely see us as a threat and remove us from the equation.
“As technologists, we have to be quite diligent in terms of the types of technologies we build when it comes to artificial intelligence. And we have a huge responsibility, too.”
How Trainline is applying AI to the business
Jonathan Midgley has been adopting a diligent approach to artificial intelligence at Trainline as the business pushes ahead with plans to become a global business, and “ultimately to do for trains what Uber has done for taxis”.
Though it may come across as a startup, the firm was created 20 years ago as a spin-off of Virgin Trains. During that time, it has collected vast amounts of data from a variety of sources, including SQL Server and Oracle databases, and all manner of data repositories. That challenge had to be handled so the business could evolve around AI.
“Two decades worth of disparate data is a real problem, especially considering that historically, we have captured that in a very dirty, disparate way,” he says.
Trainline currently captures 3TB (terabytes) data every day, which corresponds to two billion records, and has been focusing on how to better utilise that resource.
“We have realised there is a lot of value in the huge amounts of data we capture and have been on a journey to leverage that asset – and AI is at the heart of that,” says Midgley.
“We have spent a lot of time focusing on how we can make that data clean and usable, and ultimately put it into a place where it is accessible and agnostic. We have also brought data into the heart of all our conversations and everything we do.”
From a technology perspective, he sought to bring data oversight under engineering as a first step to enable the evolution into AI – and has seen instances where data is owned by finance, or even marketing – and that is far from ideal.
“At Trainline, bringing data into my organisation means I can then wrap it in great engineering practices and processes,” he says. “I can build my data into microservices, I can make it agnostic, and then I can generate really magic outputs from it.”
The company started by building a data lake on Amazon Web Services’ Simple Storage Service (S3) cloud, which, according to Midgley, has made all the difference in making effective use of data.
“[Having the data in the cloud and appropriate tooling] means I can focus on solving my customers’ problems,” he says.
“AI is genuinely unintelligent, without high-quality data. You can build all the models you want – that is the easy bit. Having usable data is the hardest part. We now have a large supply of clean, meaningful data that our teams started to leverage on very quickly.”
Jonathan Midgley, Trainline
The chief technology officer notes that “AI is like sex talk in a secondary school playground: everybody’s talking about it, but nobody’s doing it”. But Trainline claims to be walking the AI talk, through offerings powered by the technology such as a price prediction tool that, according to the executive, has delivered savings to consumers that exceed £9m since January 2019.
Additional Trainline products that are underpinned by AI capability, such as Busybot, which helps customers get a seat on the train and personalised travel disruption alerts. All of the company’s per-click advertising spend is also driven by AI capability, which Midgley says will continue to evolve, though it can already deliver some “super nuanced” insights.
“When you arrive on my website, I know what you want to search for with scary predictability and that’s based on a large volume of very clean data,” he says. “I’m very proud of that, because we’re solving real customer problems and the ultimate customer problem is saving money.”
Building an innovative IT organisation
Alluding to Hulme’s points on culture, Midgley stresses that at the heart of Trainline’s AI success so far is building an innovative IT organisation. Some of that has involved building small, agile teams that work independently.
“We’ve given teams autonomy to focus on customer goals and needs, using data and pretty much whatever tooling they decide to use to generate value,” he says, emphasising another point made by Hulme that a very early lesson learned is that technology is the simplest part of making effective use of AI.