Not a day goes by without reports of a new achievement, investment or national plan powered by artificial intelligence. AI is embedded in many of the apps and the software we use, and it is making functions such as voice interaction a reality.
Yet the adoption of AI itself is largely absent from most of the organisations with which we directly interact or work. While applications that were just a dream only a few years ago are now widespread, their development is still restricted to a handful of savvy companies.
For instance, Meta (formerly Facebook) is building the world’s largest supercomputer. The company has said its power was needed not to run the metaverse but to train AI models “that can learn from trillions of examples; work across hundreds of different languages; seamlessly analyse text, images, and video together . . . and much more”.
The number of days or months it takes to train an AI model can determine the extent of innovation and competitiveness. For a company such as Meta, shortening development time to enable quicker experimentation is essential to compete.
Financial Times Executive Education rankings 2022
The top 70 providers of customised courses and the top 65 for open-enrolment programmes in 2022. Plus the combined top 50 providers of both custom and open courses. Also, learn how the tables are compiled.
But AI uptake is painfully slow in most businesses, in contrast to other technological disruptions, from cars to digital photography to smartphones. The reason is the specific conditions required for AI adoption. For many technologies, it is simply a matter of purchasing an innovation, such as a smartphone, where local apps speed up processes.
In other cases, such as the internet or social media, significant local infrastructure and support are required to create relevant content and drive network effects and uptake.
This leads to relatively slower adoption curves. AI demands still more complex preconditions and the active involvement of companies. In “Artificial intelligence as augmenting automation: implications for employment” — my article with Feichin Ted Tschang in the Academy of Management Perspectives journal — we highlight how AI is allowing businesses to modularise and control routine work and, in the process, requires the transformation of their structures.
The disruption can be significant. In earlier eras of automation, loss of employment was offset by the growth of new sectors and jobs, and the loss of routine, middle-skilled work with a polarisation of jobs into high- and low-skilled. In an era of AI automation, this may be further aggravated.
Such obstacles mean we are witnessing two-speed adoption in which AI can seem to be everywhere except our own organisations. Yet AI is ever more essential to compete effectively, offering zero marginal cost and rapid scalability.
The consequence is a large productivity gap between “frontier” businesses and the rest. The quantity of information is greater in service industries than in manufacturing, so service companies are those in which differentiation is largest — and those without AI risk falling further behind.
The introduction of AI results in routines being translated into code, and the creation of new tasks impossible to achieve by other means. For many interconnected OECD countries, wage increases are being driven by inflation and worker mobility, meaning it will be essential for companies to tackle productivity with AI to stay competitive.
We are witnessing how AI is being embedded in products and solutions. Prime examples are warehouses using robots and the widespread deployment of recommendation engines, image recognition software, fraud detection and forecasting systems, and chatbots.
However, the adoption of AI requires changes in the business and operating models of organisations. This, together with an ever-accelerating pace, explains our two-speed world. It also explains why non-frontier organisations face increasingly tough competition.
Acceleration demands new capacities, including both sufficient AI talent and ways to foster innovative practices through a more supportive, “can-do” culture. Effective talent generation requires a network to produce, attract and retain skilled people. That can mean leading universities and research centres for training and nurturing expertise, and the prospect of high salaries and projects that are sufficiently stimulating to ensure specialist staff can be recruited and motivated to stay.
Computing power is also needed. While cloud platforms are now widely available, harnessing their potential also requires the presence of cloud-savvy universities and organisations.
Finally, organisations require specific data to give them a competitive advantage. This can be derived internally, which means it must be gathered and processed; or externally, in which case it must go beyond basic transactional data to be useful.
Capacity alone is insufficient. Progress in AI requires competitive clusters. While knowledge has gone global, innovation remains local. Without greater understanding of all these factors, a growing number of businesses will be left behind by the AI revolution.
But as AI-induced automation replaces more and more work, and much remaining employment is concentrated in a smaller, highly technical workforce, we must also reflect on how to use new technologies to promote sustainable forms of work and livelihoods.