Companies and investors are racing to make sense of the latest AI boom. But the verdict is still out on which artificial intelligence tools truly boost productivity and improve the bottom line.
Take any industry and chances are that at least a handful of companies are promising to reap stunning benefits from new-fangled artificial intelligence. These expected benefits range from increased productivity and efficiency to greater competitive advantage. Law firms, for example, say that with the help of AI, they will be able to more swiftly summarise mountains of lengthy documents, while marketing experts claim the technology will help them whip up convincing campaigns.
And the projections for the impact of generative AI are indeed impressive – everyone seems to agree on that. The only thing they don’t agree on is exactly how much value the technology will generate. McKinsey, for instance, estimates that generative AI could add value equal to or in excess of the United Kingdom’s GDP, which in 2021 stood at 3.1 trillion US dollars. It also believes that the majority of the rewards will be reaped in just four areas: customer operations, marketing and sales, software engineering and research and development.
ChatGPT, other generative AI tools
Investors have caught on to this potential, and are pouring money into this field that is not entirely new, but that has been electrified by the release of ChatGPT in late 2022. AI-related companies raised more than 25 billion US dollars in venture capital during the first half of 2023, which corresponds to 18% of all VC dollars spent worldwide.
This frenzy begs the question of what AI tools that can seemingly write and reason like a human mean for companies’ daily operations and business processes. There are clear and convincing examples of AI’s usefulness for sifting through large amounts of data, for example, to surface persistent problems in production lines or customer interactions. One example is the desire to finally close the previously not really closed loop between customer service, programmers and product development.. Rarely do call centre employees with old-fashioned tools have a holistic view of the unresolved tickets that are in the pipeline and the issues that are trending. And it is not easy for them to connect with coders and product managers to prioritise what should be improved or built next. An AI that looks across silo boundaries solves the problem.
A few studies now offer concrete insights into the impact that AI can have in the workplace. Researchers from MIT and Stanford studied call centre workers, for example, and discovered that access to a conversational assistant boosted productivity by about 14 per cent, with novice and low-skilled workers reaping even bigger benefits. An AI chatbot, they concluded, “improves customer sentiment, reduces requests for managerial intervention, and improves employee retention.”
Limits for machine learning and AI
But as long-term observer and entrepreneur Bob Goodson points out, there are limits to what AI can do. More than a decade ago, Goodson started a company called Quid that provides text-based data analysis for enterprise customers. Quid scours all sorts of written information – from news and patents to social media – to discover relevant trends before its clients’ competitors do.
This autumn, Goodson’s company integrated AI assistance into its search function, but with strict limitations. “Giving users direct access to a large language model (LLM) might at first seem like pure magic, but it comes with real problems”, he says, explaining this cautious approach. “There’s a lack of transparency. You don’t know how answers come about and how to verify them – even if they sound perfect.”
Goodson therefore suggests keeping AI tools on a leash. At Quid, for example, the technology is only used to help formulate searches. The software is fed a few keywords and then constructs a search query that would take a human up to half an hour. However, users can still check and edit the resulting query and drill down into the source material. “We turn a black box into a glass box”, Goodson says.
He thinks that AI tools that serve as add-ons will become “catalysts across most of the world’s existing software”, and will unlock vast productivity gains. The initial infatuation with generative AI, he explains, will wear off and eventually lead to a crash in terms of the public’s trust. But Goodson expects this correction to usher in a true boost in productivity. In other words, that generative AI will evolve the way technology usually does after the initial hype has died down and enterprise customers start to demand tangible value.
Wait and see over generative AI
Amit Joshi, Professor of AI, Analytics and Marketing Strategy at the International Institute for Management Development (IMD) in Lausanne, is equally cautious. He sees the current processes and the often outdated IT architecture of most companies as the reason. “The danger is that companies fall in love with the shiny new toy and forget the basics, such as adequate data collection and storage, data cleaning and compliance with regulation. That must come before any fascinating use cases can be explored”, Joshi recently wrote in an article entitled “Generative AI: go ahead, but proceed with caution.”
In an interview, he describes the space as being much like a crazy quilt. “There is a lot of scrambling by companies in terms of pure IT.” Overall, Joshi says he still sees too much “Wild West and experimentation” going on. “For the most part, I am advocating a wait and watch approach. The exact direction that generative AI will take is uncertain, so massive investments in any one LLM make no sense”, he says.
That’s why attention is shifting to vertical AI applications, the type of helpers that experts like Goodson espouse. Improvements to existing systems can, for example, better address the data security, compliance and audit trail requirements companies must adhere to. As MIT economist Erik Brynjolfsson recently said in an interview, augmentation “creates new capabilities and new products and services, ultimately generating far more value than merely human-like AI.” But while such applications may be more productive, they don’t make for nearly as exciting headlines as a potentially sentient chatbot.
What is LGT’s investment thesis around AI and generative AI in particular?
While AI is not new, ChatGPT is testimony to the progress that has been made in this area and its strong disruptive potential across industries. We now have the models and computing power to take advantage of the massive data explosion we’ve seen in recent years. We believe that after the PC, internet and smartphone era, AI will be the next step in the evolution of computing technologies. There are many business use cases that enhance productivity and innovation.
What sectors and specific companies offer the best upside?
In this early stage of adoption we don’t yet know who the industry winners and losers will be, so we prefer hardware providers and the Big Techs. Semiconductor companies provide the computing power to train and use models, regardless of who will benefit in the end markets. The Big Techs are well positioned across the entire AI value chain. At a later stage, when it comes to verticalisation, we believe companies with unique data sets will have an advantage.
What risk factors and unknowns could rain on AI’s parade?
Because of AI’s disruptive potential in many aspects of our lives, it requires an appropriate regulatory framework. There are many issues that need to be addressed, from legal to social, and that will take time. So despite the initial excitement, we should keep Amara’s Law in mind: “We tend to overestimate the effect of a technology in the short run and underestimate the effect in the long run.” There are also a number of technological and geopolitical risks.