'There is no practical reason why trust in AI cannot be created now' - Rolls-Royce's bold ethical initiative takes shape

The debate around trust and ethics relating to Artificial Intelligence has been an ongoing theme for the past few years - and in all likelihood will continue to be so for many more to come. Attempts to devise solutions to the issue range from international efforts to come up de jure standards as well as de facto offerings from real world users. But the great thing about standards is, of course, there are so many to choose from. 

Last week as part of London Tech Week, engineering giant Rolls-Royce came up with what sounds like a highly promising initiative to, as it puts it, “help gain society’s trust” of AI and in the process “accelerate the next generation of industrialisation”.

The announcement is based around two main elements. The first is an AI ethics framework that any organization across multiple sectors can use to make ethical decisions around the deployment of critical and non-critical applications. The second aspect is a five-layer checking system focused on the output of algorithms, with the intention of preventing biases developing undetected and building faith that they are trustworthy.  

In a significant move, both the framework and the checking system will be published in full under Creative Commons licence later this year on the Rolls-Royce.com website. Warren East, CEO of Rolls-Royce, says that the firm believes its work has a wider applicability that will be beneficial across sectors: 

I believe the impact of our work has the potential to go far beyond just our own field of expertise into broader industrial AI applications and well beyond industrial application…During our peer review process with experts from big tech companies, academia, automotive and pharmaceutical industries, we were genuinely surprised to learn that our work – which began as a way to address an internal challenge – stands at the leading edge of the practical and ethical application of AI in a critical industrial context. And as we went on, we came to recognise that its implications could extend way beyond that.

It’s work that builds on more than three decades experience in applying advanced analytics within Rolls-Royce, he explains: 

We are probably best known for our advanced manufacturing and jet engines rather than our developmental work in Artificial Intelligence, but we have more than 30 years of experience in advanced data analytics and have been developing our AI capabilities for many years…Our industrial journey in Machine Learning, Artificial Intelligence and Advanced Analytics started in 1980 with engine test bed measurement and analytics. It really came to life as a disruptive technology for us in the late 1990s with our Engine Health Monitoring system. That has been a key enabler for the delivery of many billions of pounds worth of aircraft maintenance contracts.

At present - or until COVID-19 devastated the airline industry at any rate - Rolls-Royce has monitored up to 8000 flights a day worldwide, with 3000 of its engines being in the sky at any one time, adds East. Analytics and AI play a crucial role: 

Each engine has multiple sensors on board that relay information back to us during the flight. In total, we analyse more than 5 million data parameters from our engines every day and use AI to provide insights to our engineers. If we spot anything out of the ordinary, we can dispatch a maintenance crew to be ready on the ground when the aircraft lands. [The AI] learns the normal operational state of our engines and scans for relative system changes and looks for unknown anomalies rather than relying only on a preconfigured set of fault signatures. This means our engineers can make decisions quicker than ever, reducing disruption for our customers. 

Another tool to use

But Rolls-Royce's use of AI tech goes beyond aviation and the behavior of engines: 

Our current work includes applying AI, through our in-house data innovation team, R2 Data Labs, to improve the management of risk in our supply chains, to predict market demand signals and to examine how to improve the efficiency of our operations. We’ve deployed this, for example, with the micro-grids created by our Power Systems business, making our industrial power technology more reliable and sustainable. In the future, our AI work will involve increasing our use of cloud-based services which will be governed by a data ethics framework that covers more than 200 of our digital projects. And we will continue to rely on AI to efficiently design and operate our products. 

For East, AI is regarded as another design tool that can be used by creative people to produce new solutions and, as such, future applications are as yet untapped: 

At the moment, a lot of our business is around commercial aerospace, so there's some very straightforward applications around how can we make our engines last longer and how can we trust that they last longer so that airlines can not  be repairing engines as often as they would otherwise do? Well, that's a very boring application, but it's really enhancing the business that we do today. We look at the business of tomorrow and it's all about making things like aviation zero carbon. That's going to engage a whole lot of different technologies and if AI helps in the rollout of that, then that's goodness as far as I'm concerned. Don't forget this is all about data. We extract a lot of data from the products that we have or that our customers are using and if we can explore that data and analyse that data and do clever things with that data to enhance the efficiency, to enhance the way in which stored energy is turned into useful power and make that carbon free, then that's a good thing.

He cites other tech developments as a useful comparison point, such as the rise of connected ‘smart’ devices’: 

People say, you know, ‘What’s the use of putting connectivity into everyday products?’ and at first you think, “Well, I can't really think of why'. Then, actually, before we know where we are, there's all sorts of products. COVID at the moment, for instance, is causing people to be nervous about touching things and we're seeing a rise in the use of things like contactless door handles that you can open with your smartphone. That's come about because somebody once upon a time just trusted in adding connectivity to something which wasn't connected before. And so it is with AI in a business like Rolls Royce - our mission is to help our customers get to zero carbon, and we're all in the business here of taking stored energy, which today is mainly fossil fuel, and turning it into useful power and AI is going to help us, just like connectivity.

Safety first

The nature of Rolls-Royce’s business means that the importance of trust is hard-wired into the corporate DNA, says East: 

We are a company where having a safety mindset is fundamental to our products and services, and so it must be fundamental to our data innovation. I believe that is why we’re the first to go beyond theory in this space, so that we can actually demonstrate that the AI systems we use are trustworthy and safe. Our AI is deeply embedded into other people’s products and services and it’s BA or Lufthansa that fronts up to the consumer. Rolls-Royce’s AI doesn't feature in a consumers' understanding of how the digital world is changing their lives. We need people to trust it, as they trust the power of the engine to propel the plane into the air, but they don’t even know the AI is there. 

This recognition of the importance of trust is informing Rolls-Royce attitude to AI: 

In wider society AI can scare people, and there are scaremongers out there. Like most of the other tools we use, if misused of course there is a dark side. But I believe we can do better than that. Dismissing a new technology or tool because of the scary bits won’t help, and isn't justified…I quite often liken using AI and trusting AI to how we we trust each other as people. In a business, one person starts off a business, and they do everything and then they have to trust in another person, and eventually they do and then they find that they can do a lot more and so the business scales up. We're going to learn to delegate more to AI and this framework is helping us build confidence in that delegation.

As for the decision to open up the results of Rolls-Royce’s work to the open market, this is a decision that East sees as essential to driving the wider adoption of AI and being able to tap into its potential benefits: 

We are not seeking a commercial return from this work beyond that which its use will achieve for our business and customers. We have come to realise that it’s potentially so useful that everyone needs free access to it and as a CEO I’m a strong believer in open source sharing, open architectures, integration and networking. I realise that some things you just need to give away and trust in the business benefit materialising through overall growth. 

As a business we are open to collaborative innovation and we will continue to talk to key stakeholders, customers, counterparts and technology leaders to share our work in detail, to see how we can help each other progress for the greater growth, wealth and health of our world…There is much more to do and by publishing our findings we want to move the AI ethics conversation forwards from discussing concepts and guidelines to accelerating the process of applying it ethically. There is no practical reason why trust in AI cannot be created now.

My take

Someone once characterised efforts to come up with de jure standards in technology as following their own standard model. At first everyone plays nicely, says all the right things and works together to create a vision of a greater good. Then as soon as the results of the standards work reaches a point where it can be productized and there’s money to be made from being first to market, everyone reverts to fighting like a bag of cats. I’m impressed by the sentiments expressed by East and the Rolls-Royce team about driving a wider change in attitudes to AI built on practical experience. Leadership from such a trusted industrial brand takes the debate onto another level alongside the more prevalent one which focuses on consumer sector impacts. I look forward with interest to seeing what the published work looks like in the coming months.