Market Perspectives September 2023
Read our latest round-up of the global themes, trends and events currently influencing investors.
Artificial intelligence
04 September 2023
Luke Mayberry, London UK, Investment Analyst; Lukas Gehrig, Quantitative Strategist, Zurich, Switzerland
Over the past year, Artificial Intelligence (AI) has captured widespread attention due to its potential to alter radically how we live and work. While some form of the technology has been around for seventy years, the latest examples of so-called generative AI have finally made it real, and somewhat scary, for many.
In the run-up to this year, there was a surge in the number of AI research publications (see chart). The pace of implementing the technology, meanwhile, is accelerating too: the famous natural language bot ChatGPT, that can trawl the web to create human-like answers to queries and has prompted many universities to rethink how they assess students, has already been succeeded by GPT-4, a more versatile and powerful tool, only four months after GPT-3 was launched1.
Number of publications on AI and increase in computational power as measured by number of transistors (semiconductor “switches”) per microprocessor (core of the central processing unit) on a logarithmic scale
With companies increasingly promoting their AI capabilities and investors driving up the share prices of some sector darlings, this article looks at a few of the core principles of the concept, what it means for the economy and how it might reshape investors’ portfolios.
AI describes a machine's ability to perform human-like cognitive processes such as learning and interacting with others. It can be seen in everyday applications, such as chatbots, search engines and voice-recognition systems (for instance, Apple’s Siri and Amazon’s Alexa).
The term was initially coined in 1956, when John McCarthy and Marvin Minsky invited scholars to the Dartmouth Summer Research Project on Artificial Intelligence. But AI could not thrive on ideas alone, it needed raw computing power, and lots of it.
As such, the history of AI is related to the availability of computational power. In 1957, Newell and Simon’s General Problem Solver managed to solve the Towers of Hanoi game, which involves moving a stack of discs, via three wooden rods, from one end to the other.
It then took until 1997 for IBM’s chess computer programme to beat reigning world chess champion Gary Kasparov4 – the computational power, as measured by transistors per microprocessor, increased by over 1500-fold during that time.
AI can be thought of as being ‘narrow’ or ‘general’. Narrow AI (or weak AI) is designed to solve a specific task efficiently, for example face recognition or speech detection. The technology cannot complete a task that it has not been programmed to do. General AI (or strong AI), more of a theoretical concept than reality so far, allows machines to apply knowledge and transfer skills to complete new tasks. The latter variant arguably experiences consciousness, like humans.
Machine learning is a subfield of AI, focused on developing algorithms that can draw from large sets of data to make predictions or decisions. It can be split into supervised learning (using labelled data), unsupervised learning (finding patterns in unlabelled data) and reinforcement learning (learning through interaction with an environment and feedback).
Deep learning is a subset of machine learning, in which artificial neural networks, that mimic the human brain, are used to complete more sophisticated reasoning tasks. And all this is done without our intervention. Within deep learning, lies generative AI, that can create new content, such as text, audio and video, using different algorithms that have been trained on a pre-existing dataset.
From a macroeconomic standpoint, the adoption of AI into the economy could hardly have come at a better time. Using the UK as an example of a developed market economy: many employers are already struggling to find workers. This problem is unlikely to disappear soon.
By contrast, up until 2010 the growth in the UK working-age population was sizeable (see chart). Between 2010 and 2020, increases in the participation rate (aided by people outside of the workforce entering it) helped to mask the weaker total labour input. But since 2020, increasing numbers of older workers are really taking a toll on the workforce, as can be seen in our 10-year projections in Strategic Asset Allocation 2023 update.
In an economic system that is geared towards growth, something else needs to be found to make up for the shortfalls in labour production. AI, with its potential for productivity enhancements, is one of the most obvious solutions to this issue.
Stacked growth rates of the UK working age population (15-74), labour-force participation rate and average-hours worked per worker, as well as the aggregate growth rate of the labour input (hours total). Own projections from 2022 onwards
At a company level, AI could be a transformative force with significant benefits. The technology enables more effective automation, streamlining operations by handling repetitive tasks and allowing employees to focus on more complex and innovative endeavours5. For instance, ExxonMobil has become a pioneer in harnessing AI for autonomous drilling, more energy-efficient refining and less polluting processes in general for the energy sector.
Tech giants lead the way in using AI, particularly in narrow AI applications such as search and query responses. Hardware manufacturers, such as hardware manufacturers, that create the chips needed for AI applications, have seen their stock prices soar this year given the perceived prospects for their sales and future profitability.
More widely, industries ranging from healthcare, transportation and finance could be altered by AI. Indeed, it is expected to make many jobs largely redundant while creating a swathe of new types of employment.
Thinking about common human behavioural finance biases, and their ability to harm performance, it is easy to get excited about AI applications. Machines need not be attached to past investments (overcoming sunk costs), can estimate potential gains and losses accordingly (overcoming loss aversion) and are less prone to the risks of being overconfident.
However, AI will probably introduce its own set of biases. Any narrow AI application is only as good as the data it is fed. If the data used is biased, incomplete or worse still fake, then the resulting predictions could be flawed.
Investors have several ways to access the potential opportunity in the sector, such as by gaining exposure to stocks directly, as well as through technology funds that invest in a broad mix of related companies. Alternatively, electronic-trading funds that offer substantial exposure to both established tech giants and smaller companies poised to benefit from the long-term growth in the area.
While listed securities can provide exposure to such technology, it may not be part of their main business activities. Large companies tend to acquire AI businesses to help improve their own products. However, many companies focused solely in this field are not mature enough to be listed in the S&P 500 or other public markets. The most direct and focused way to gain AI exposure probably lies in private markets.
One way of investing indirectly (and with a medium-term horizon in mind) in AI could be to focus on companies expected to make efficiency gains from incorporating such applications. A recent study by consulting firm McKinsey & Company explores the disruptive potential of generative AI across sectors6.
While the industrial revolution mostly transformed manufacturing, the generative variant might be most disruptive in services, such as high-tech, banking, education, pharmaceuticals and telecommunications. This is due to its ability to interact with customers, develop software and manage administrative tasks more efficiently.
Investing in AI businesses also entails several possible risks that are worth considering. Ethical concerns around solutions that are seen as biased against select groups of people, invasive, or discriminatory could lead to legal and reputational risks, and could trigger a consumer backlash against relevant companies.
Limited regulation within the nascent industry brings its own risks. Companies are planning blind to potential new regulations that may emerge and limit their ability to operate. Data privacy and security pose other dangers, as the reliance on extensive sets of information makes companies vulnerable to data breaches and legal repercussions.
While semiconductor manufacturers are less likely to be hit by concerns around AI regulation and ethics, chipmakers are largely exposed to geopolitical risks as their profitability is dependent on delivering AI infrastructure globally. The current tit-for-tat trade disputes between the US and China are a case in point, risking a fall in revenues for companies that get caught in the crosshairs.
This impressive performance of the largest tech companies in 2023, has pumped up valuations, with several AI-themed stocks trading at elevated price-to-earnings (P/E) multiples. Many investors question the sustainability of the rally, especially given it is largely built on the promise of considerable revenues well into the future, drawing comparisons to the Dotcom bubble that popped up in the late 1990s and early 2000s. While there are similarities in investor behaviour, there seem to be many differences between the two eras.
While current valuations are stretched, P/E ratios are still significantly lower compared to the levels seen during the Dotcom bubble. For instance, in March 2000, the tech-heavy Nasdaq 100 had a P/E ratio exceeding 90 times earnings, whereas it stood at just above 30 times earnings in August (see chart (left hand side).
In addition, the AI rally is being propelled by established mega-cap tech companies, some from the Dotcom days, that have more stable earnings and stronger balance sheets relative to those, smaller, companies that led the rally in 1999. Earnings have risen consistently, with the share price, over the last few years, whereas there was clearly a huge disconnect during the Dotcom bubble (see chart (right hand side)).
Today’s commercialisation of AI is more rapid than what was witnessed in the Dotcom era, and this was highlighted when ChatGPT reached 100 million users just two months after being launched. Companies are swiftly adopting AI to enhance operational efficiency. A study by IBM found that 35% of businesses reported using AI and about 42% are exploring adopting AI7.
In the last forty years the NASDAQ 100 share price and earnings per share have largely moved in step, while looking at the index’s price-to-earnings ratio shows that the Dotcom bubble was more extreme than the current AI-driven run
As the promising potential for AI to revolutionise business, redraw the labour force and power profits becomes more apparent, substantial corporate and government investment in the field looks certain. Such a scenario, leaving aside the risks, is likely to positively impact AI-related stocks as companies seek to boost operational efficiency. In turn, investment opportunities might flourish for investors.
While there is no doubt that the technology will reshape work and industries in time, anticipating the potential barriers to capturing the full benefits will be more difficult. As such, a diversified approach to the sector probably provides the best hope of investing success over the long term.
Read our latest round-up of the global themes, trends and events currently influencing investors.
Microprocessor trend data, Karl Rupp, August 2023Return to reference
Artificial intelligence index report 2023, Stanford university, August 2023 [PDF, 24.1MB]Return to reference
Deep Blue computer beats world chess champion, The Guardian, 12 February 2021Return to reference
Will COVID-19 be the tipping point for the Intelligent Automation of work? A review of the debate and implications for research,ScienceDirect, December 2020Return to reference
The economic potential of generative AI: The next productivity frontierReturn to reference