How to spot investment opportunities in AI

The debate about the development of artificial intelligence can sometimes seem like a bungee jump off the Eiffel Tower.

One day the headlines are all about the existential threat that AI poses to humanity. The precise danger is not always spelt out but it is hard not to think of that Skynet program from the Terminator movies, which launches a nuclear war to eliminate its human masters. 

Another day, the forecasts are delirious with optimism. Take the view of Marc Andreessen, a US venture capitalist, who wrote recently that, as a result of AI: “Productivity growth throughout the economy will accelerate dramatically, driving economic growth, creation of new industries, creation of new jobs, and wage growth, and resulting in a new era of heightened material prosperity across the planet.”

So far, investors tend towards the optimistic camp. Manish Kabra of Société Générale calculates that without AI-related stocks, the S&P 500 index would have been up only 1 per cent in the first half of 2023, rather than 15 per cent. The rise in the S&P 500, says Dhaval Joshi, chief strategist at BCA Research, has been driven by “euphoria in a handful of megacap AI plays”. But he warns that “this euphoria is completely disassociated from the near-term evolution of the global economy”.

Productivity growth throughout the economy will accelerate dramatically, driving economic growth, creation of new industries, creation of new jobs, and wage growth

One of the hottest AI-related stocks is Nvidia, which makes graphics processing units (GPUs) that are best known for powering video games. It turns out that GPUs are also ideal for developing AI; Nvidia estimated that its second-quarter revenues would jump to $11bn, from $7.2bn in the first quarter. That has taken the company’s market value to $1tn. 

The UK has only a small tech sector, but there has been investor enthusiasm for shares in DotDigital, a group that helps businesses automate their marketing campaigns.

So how realistic is the hype? Investors over 40 may recall the dotcom boom of the late 1990s when money flowed into technology stocks. While some of today’s tech giants, such as Amazon, were around back then, others such as Alphabet and Meta (the parent companies of Google and Facebook) had yet to list.

Meanwhile, many early dotcom pioneers, such as Boo.com and Webvan, fell by the wayside. Those investors swept up by the enthusiasm were caught out when technology funds plunged in value between 2000 and 2002. The tech-heavy Nasdaq index rose fivefold between the start of 1996 and early 2000, only to fall 78 per cent by the autumn of 2002.

That episode should have taught investors that they need to be patient when new technology trends emerge. Morgan Stanley cites Amara’s law (named after Roy Amara, a US scientist) which states that “we tend to overestimate the effect of a technology in the short run and underestimate the effect in the long run”.

Certainly, the development of AI is potentially important for the global economy and thus for financial markets. But the scale of that importance is difficult to judge, whether on the financial, economic or societal level, because the technology is still at an early stage. Retail investors who want to profit from the new trend face a lack of investment options, as well as the uncertainty about the long-term trend.

The promised productivity boost

There are three ways in which AI might bring changes to the way we live. 

The first is linked to ChatGPT, Bard and other “large language models” that have received much publicity in recent months. Many people seem to find them useful: ChatGPT reached 1mn users within five days, a record for any application. These models can substitute for a lot of the “grunt work” that occupies the time of knowledge workers, such as creating computer code, summarising articles, and translating languages. The gains are not just linked to text; AI can also generate images. This could change the way goods are marketed; in May, WPP, the global advertising agency, signed an agreement with Nvidia to produce adverts using AI technology. 

A second development is the use of AI for efficiency gains. For example, it can be used for predictive maintenance in infrastructure or to enable farmers to use fertiliser more efficiently. In computers, AI can analyse code and look for vulnerabilities, or suggest experimental lines of code that humans can check.

The third change is hard to assess, but potentially very exciting. The brainpower of AI may be able to drive innovation because it can make faster calculations and perform deeper analysis than humans. In 2020 DeepMind, part of the Alphabet group, cracked the problem of predicting protein structures — a discovery that could, potentially, lead to more effective vaccines and drugs.

Add all this together and the gains could be substantial. In a research note released in March, analysts at Goldman Sachs said that AI adoption could lead to an annual productivity improvement of 1.5 percentage points in the US over the next decade. If correct, that would double the productivity growth rate since the 2008 financial crisis, a substantial economic gain.

The downside dangers

But AI’s potential gains also have drawbacks. First, AI may be a threat to jobs, particularly among administrative staff. Second, the information it provides can be unreliable. A US lawyer submitted a court brief on behalf of his client, citing a number of impressive precedents; unfortunately he had used ChatGPT to research the brief and the cited cases did not exist. Third, AI can be used for malign purposes, such as generating fake videos and news stories. Hackers can use AI to find weak spots in our technological infrastructure.

On the first concern, it is common to worry about jobs when new technology is introduced. Goldman estimates that around two-thirds of jobs are exposed to some kind of automation, and about a quarter of all “work tasks” could be automated. This is not all bad news; in many cases, AI will perform the most tedious jobs on employees’ behalf. In total, Goldman thinks that roughly 7 per cent of US jobs will eventually be replaced. And many of those workers may be re-employed if, as with previous technological advances, economic growth creates new jobs.

Still, even if new jobs are created, there is a risk that AI will lead to a division between a highly paid technological elite and a mass of low-paid, low-skilled workers. So far, there is no sign of that happening; an OECD survey found that four times as many AI users said the technology had improved their working conditions, compared with those who said their conditions had deteriorated. A study by the Massachusetts Institute of Technology found that low-skilled workers benefited most from using AI.

The malign uses of AI are harder to quantify. Because these programs learn as they go, they are “black boxes” which even their creators do not understand; so it can be hard to tell whether they are doing such things as discriminating against some members of society. It is worth remembering that the internet was initially described in very positive terms before the advent of spam email, fake news, and trolls.

Remember dotcom

To put AI into perspective, there have been three or four transformative technologies over the past 250 years. The first was the steam engine, which represented a massive improvement in productivity over human and animal power. The internal combustion engine then changed the layout of the landscape and our cities, the way people commute and enjoy their leisure. The third shift was electrification, which brought power and light to our homes and enabled the creation of many labour-saving devices and sources of entertainment. 

Those three technologies were clearly revolutionary. More recently, personal computers have changed the way we communicate and absorb information. However, apart from a period in the 1990s, computing does not seem to have delivered the same boost to productivity as the previous technological developments. Smartphones were introduced in 2007 and they are in some ways miraculous; huge computing power in a handheld device. But the smartphone era has coincided with sluggish productivity growth.

One reason may be that smartphones are as much a distraction as a useful tool. Another may be that the ability to, say, communicate with friends quickly, or check directions, is very useful but adds little to GDP. Similarly, large AI-powered language models will make it easier for office workers to write reports. Some of this extra paperwork may be useful but not all of it will be.

These uncertainties help explain why the overall stock market has not received an even bigger boost from the latest AI developments along the lines of the late 1990s dotcom boom. Even the optimistic Goldman report concludes that, given the uncertainties, “we are not incorporating our findings into our baseline economic forecasts at this time”.

In retrospect, AI may come to be seen as merely one phase in the computer revolution, rather than as a transformative technology in its own right. Indeed, AI is merely a more glamorous term for “machine learning” which has been around for a while.