Investors who feared that American tech giants were making overly ambitious bets on generative artificial intelligence (AI) were reassured by the latest quarterly results from the major players in the sector. Growth in enterprise demand for cloud services from Amazon, Microsoft and Google has been meteoric. Amazon CEO Andy Jassy said AI-driven revenue for Amazon Web Services (AWS) was growing in the triple digits – three times faster than AWS itself in the last years. first years following the creation, in 2006, of this pioneer of cloud computing.
Faster adoption by individuals
If we dig a little deeper, the situation is more nuanced. Generative AI seems to be one of those innovations, like email or smartphones, whose most enthusiastic first-time adopters are individuals. Businesses are much more hesitant.
39% of Americans say they use AI in their work, while many find their employers dragging their feet.
In the two years since OpenAI unveiled ChatGPT, generative AI has seen a faster rate of adoption than PCs or the Internet. According to a study carried out by Alexander Bick, of the Federal Reserve Bank of St Louis, and his co-authors, 39% of Americans now say they use it; 28% of them say they use it as part of their work, including 11% daily.
However, many of them are apparently clandestine “cyborgs,” using technology at work even as their employers drag their feet. Only 5% of U.S. businesses report using technology to produce goods or services, according to a U.S. Census Bureau survey.
“Pilotite” and limited income
Many companies seem to be suffering from a case of “acute pilotitis” – that is, they procrastinate on pilot projects without actually deploying the technology. In a recent 14-country survey by Deloitte, a professional services firm, only 8% of business leaders said their company had implemented more than half of their generative AI experiments.
Many companies suffer from “acute pilotitis”: they procrastinate on pilot projects without actually deploying the technology.
As a result, revenue from selling AI services to businesses remains limited. Andy Jassy may say that AWS now generates “several billion” dollars in revenue thanks to AI, but this sum represents only a tiny fraction of the $110 billion in annual revenue from its entire cloud computing business. . Accenture, a consulting giant that recently announced it would train 30,000 people to help companies adopt generative AI, said in September that it had booked $3 billion in orders related to the technology over the past few years. last twelve months, ten times more than the previous year. But compared to the company’s total revenue of more than $81 billion, this is still a drop in the ocean.
Why are so many bosses hesitant to adopt generative AI? One reason is that they fear the downsides. Listen to the tech giants and they will tell you – as Alphabet boss Sundar Pichai said in July – that “the risk of underinvestment is much greater than the risk of overinvestment.” Alphabet, Amazon, Microsoft and Meta are expected to spend at least $200 billion on AI-related investments this year. Bosses in other sectors are more circumspect. In a recent closed-door discussion, the head of a major U.S. business group discussed two types of concerns business leaders have about generative AI. The first is to be left behind if they adopt it too slowly. The other is to make a fool of themselves if they seize on it too quickly, thereby damaging their company’s reputation.
Regulatory risks and financial uncertainties
Legal and regulatory risks are significant. Lawsuits relating to privacy, discrimination and copyright infringement will soon hit the courts. In August, the European Union’s AI law came into force. Similar bills have been introduced in at least 40 US states this year. Bosses in highly regulated sectors, such as health and finance, are particularly cautious. While they see the potential for generative AI to transform their businesses, for example by accelerating drug discovery or fraud detection, they are acutely aware of the threats to privacy and security if medical data or financial matters of their clients were the subject of a violation.
Large-scale implementation of generative AI can increase revenue and reduce costs, but the benefits are not immediate.
Another issue is that the benefits of adopting generative AI may be uncertain. Access to large language models (LLMs) is expensive, whether through the company’s own servers (more secure) or through cloud service providers (simpler). Large-scale implementation of generative AI can increase revenue and reduce costs, but the benefits are not immediate, raising concerns about return on investment. In its recent survey, Deloitte found that the share of senior executives reporting a “high” or “very high” level of interest in generative AI fell to 63%, from 74% in the first quarter of the year, which suggests that the “glow of new technologies” may be fading. One executive sums up this skepticism by telling the story of a CIO who was asked by his boss to stop promising 20% productivity improvements unless he was first willing to cut by a fifth the staff of his own department.
Technical debt and the hunt for talent
Even when companies want to scale up their use of generative AI, they may face challenges. To reap the full benefits of technology, they must first upgrade their data, systems and train their staff, says Lan Guan, head of AI at Accenture. She believes that businesses are far less prepared for generative AI than they were for previous technological advances, such as the Internet or cloud computing.
One of the problems is the mess of data, scattered in different formats across various departments and software systems. Lan Guan cites the example of a telecommunications company that wanted to train a call center AI assistant by providing PDFs, manuals, call logs, etc. The robot discovered that instead of just one standard operating procedure – what Lan Guan calls “a single source of truth” – the company had 37, accumulated over decades. Not organizing data before using it to train a robot increases the risk of hallucinations [réponse fausse ou trompeuse présentée comme un fait, ndt] and errors, explains the specialist.
A sales rep with AI skills can earn $45,000 more per year than one without them.
Another problem, known as “technical debt,” arises from the fact that computer systems are often old and fragile. It can therefore be difficult to connect LLMs without causing malfunctions. Integrating semi-autonomous AI agents into systems designed for humans could also create security vulnerabilities.
Finally, there is the problem of skills. Many companies still struggle to find enough AI specialists. According to the research firm Lightcast, job offers linked to this technology have jumped 122% since the start of the year in the United States, compared to an increase of 18% in 2023. Elizabeth Crofoot, economist at Lightcast, indicates that this increase is mainly explained by generative AI, with job descriptions mentioning ChatGPT, “prompt engineering” (query engineering) and large language models increasing in number.
Companies are also looking for workers in other functions who know how to use generative AI. According to Elizabeth Crofoot, a sales rep with AI skills can earn $45,000 more per year than one without them. It is therefore not surprising that if some bosses are hesitant to deploy generative AI, their employees are all up for it.
The Economist
© 2024 The Economist Newspaper Limited. All rights reserved. Source The Economist, translation The new Economist, published under license. The article in original version: www.economist.com.
