By Nick Hartley, Co-Head of Active Equities at LGIM
In recent investor presentations and earnings calls, we have increasingly come across the terms 'artificial intelligence' (AI) and its terminology bedfellows, 'machine learning' (ML) and 'Big Data'. While companies have addressed these concepts for a few years now, they have taken centre stage in the last 12 months.
While we see a lot of hype in the near term, there are enough real life examples underway to indicate AI will be a significant investment theme in the next decade.
In its purest form, AI is defined as 'computer systems able to perform tasks normally requiring human intelligence' 1. In practice, it is self-learning technology driven by machine-learning algorithms that try and fail, learning from experience and improving millions of times over. The algorithms require access to vast amounts of data to allow them to fail, learn and discover ways to improve.
Advances in computing power, data processing and connectivity have triggered a sharp rise in potential AI applications. Real world examples of computer systems able to perform tasks previously requiring 'human intelligence' – including visual perception, speech recognition and language translation – are already achieving superior accuracy.
VISUAL PERCEPTION: CATS TO CRIMINALS
In 2012, computer scientists at Google X built a neural network of 16,000 computer processors with a billion connections. They then put it in front of millions of randomly selected YouTube videos. Remarkably, it taught itself to recognise cats. While this may not sound like a dramatic breakthrough in practical terms, the fact it achieved 'unsupervised learning' was a big step forward.
More helpfully, the system also trained itself to recognise facial features, achieving 81.7% accuracy in detecting human faces at the time. Today, AI and ML have enabled computers to zero in on the features that will most reliably identify a person, improving the accuracy rate and potential applications.
On the back of recent such improvements, companies across all sectors are now ploughing investment into visual perception and speech recognition. Forecasts suggest spend on AI-focused technology will reach $58bn by 20212, making it one of the fastest growing technology segments.
It is important to stress that AI is not a product offering per se but an algorithm-based model that discovers patterns and logic in large amounts of data. As it transitions out of the lab and corporates look to adopt the techniques more widely, we consider three components to be crucial for financial success:
1. The technical capability for AI and ML
2. Access to a large volume of data from which it can learn
3. The ability to deploy and implement AI at scale, and then ensure it keeps learning and improving
The final step may be the hardest to achieve and is potentially a major barrier to fully harnessing AI, as demonstrated by the following two contrasting examples.
WHEN DO WE SEE MORE THAN A 'SECOND OPINION'?
The case for applying AI in healthcare is compelling. In theory, diagnoses informed by a machine that has been 'taught' by some of the best medical professionals and is able to draw upon data from millions of medical records ought to be more reliable than any one single medical professional.
IBM certainly believes so, having spent billions of dollars employing their machine learning tool Watson to the healthcare vertical. However, investors are starting to question when and whether they will see the financial fruits of these investments as the adoption of this service has been disappointing so far.
FROM LOOK TO BOOK – BEYOND THE HUMBLE TRAVEL AGENT
AI in travel has been more successful, deployed in part to address some of the challenges thrown up by the dramatic shift of travel booking in the digital age. As a consumer, the booking process may feel simpler; from a data perspective, it has become infinitely more complex.
Amadeus, the market-leading provider of IT solutions for the travel vertical, has used AI and machine learning to improve the user experience, using the predictive capability of its algorithms. It has both expertise and a significant volume of data from which to work (Amadeus processed more than 595 million travel agency bookings and boarded over 1.3 billion passengers in 2016).
Crucially, though, it has also been able to deploy its learning at scale: as a marketplace model, its use of AI increases its relevance to travel providers by improving the conversion rate and supports their core business.
AI FOR HIRE
Outside of the technology sector, building the capability for AI and ML is more difficult. However, due to the evolving business models of some large-cap technology companies that have pioneered the development of AI, it is not as difficult to access as one might imagine.
Amazon and Netflix both use an AI framework to power their respective recommendation systems, which can automatically predict user preferences based on past behaviour. Amazon, Apple, Google and Microsoft are leaders in the field of voice and personal assistants. And Google incorporated RankBrain – a deep-learning technology that helps refine queries and rank web pages – into its search engine from 2015.
It is estimated the FANGs (Facebook, Amazon, Netflix and Google) each invests more than $5 billion annually in AI. The motivation is not only driven by a desire to innovate their core businesses, but also in the cases of Amazon, Microsoft and Google, to provide cloud technology services to businesses across all sectors.
The technology giants are able to commercialise their AI capabilities by renting them out to their cloud customer base, opening up AI to all sectors and companies of every size.
THE ULTIMATE TEST – AUTONOMOUS VEHICLES
As a top use case for AI, autonomous driving is one of the most complex forms of AI underway for two reasons: first, it involves combining a broad and constantly changing set of data inputs; and second, if it goes wrong – as in the recent case of Uber's fatal self-driving crash – it results in human casualty.
Progress towards full autonomy has been gradual, given a requirement for instantaneous decision making. Once again, the barrier is not the technical capability, nor the data; various programmes have demonstrated this capability already. The challenge is being able to deploy the capability at scale and benefit from the virtuous circle of more driving, providing more data, enabling more learning and an improved capability.
COST-SAVING OR LABOUR-SAVING?
Incorporating AI into business processes can help improve productivity by automating tasks that previously required human labour. While broad fears of labour displacement persist, the reality is more nuanced. AI is often deployed to support – rather than replace – labour. AI can therefore be seen as a critical component in raising economic productivity from a macro perspective and in raising corporate profitability from a micro perspective.
The possibilities yielded by AI apply across every sector. In our view, the real beneficiaries of the growing capability of AI will be the companies that can combine technical capability
and large datasets with the crucial ability to deploy AI at scale in their core businesses. However, the scale providers of AI capability – the large-cap technology players – are, once again, front and centre of this profound long-term investment theme.
- Arm Holdings, Global Artificial Intelligence Survey, 2017
- International Data Corporation, Worldwide Semi-annual Cognitive Artificial Intelligence Systems Spending Guide, 2017
The value of any investment and any income taken from it is not guaranteed and can go down as well as up, and investors may get back less than the amount originally invested. Legal & General Investment Management Ltd, One Coleman Street, London, EC2R 5AA www.lgim.com. Authorised and regulated by the Financial Conduct Authority.