Nordea On Your Mind: Generative AI

When Johan Trocmé and Viktor Sonebäck wrote about artificial intelligence (AI) for Nordea On Your Mind back in 2019, big data had improved the performance of AI algorithms to such a degree that there was a plethora of commercial applications and widespread media interest. Now, in response to a new wave of interest driven by the success of large language models (LLMs), the authors provide a recap of the history of AI, the drivers behind its commercial breakthrough in recent years as well as the new hype around natural language processing and LLMs such as ChatGPT.

The big new thing: Large language models

Since we last wrote about artificial intelligence (AI) in 2019, there has been a new surge in interest in the topic. The rapid evolution of large language models (LLMs) such as ChatGPT has created a new, easily accessible and intuitive user interface for AI applications. The impressive capabilities of LLMs are giving rise to high expectations for potential productivity-boosting use cases, as well as fears for human job losses or even for risks from an AI evolving into a superintelligence.

Sophisticated sentence completion applications

LLMs are a form of generative AI trained on huge text datasets. They are essentially word predictors that output responses based on an analysis of how words typically fit together. Interest has soared after the launch of OpenAI's ChatGPT in 2022, which gives anyone access to its GPT-3.5 LLM for free through a chat prompt. You ask or instruct it, and receive text responses which seem like human dialogue. LLMs can answer questions, create, or modify text, based on the existing knowledge and data which has been used to train them. They do not think, reason or create actual new information. But they can be a powerful tool for boosting work productivity, now with a friendly and easy user interface.

Should we be afraid for our jobs?

Generative AI could fuel a labour productivity boost, which is sorely needed after more than a decade of stagnating productivity in the OECD, as described in our 2020 Nordea On Your Mind report Industry 4.0. As an illustration, our simple analysis shows that as a gross impact, a 1.5pp boost to productivity with no boost to GDP growth could eliminate just over 1 million Scandinavian (Norway, Denmark and Sweden) jobs, or nearly 9% of the workforce, by 2028. But productivity drives GDP growth, and for example a combination of 0.6pp higher productivity growth and 1pp higher GDP growth would generate a net increase of 300,000 jobs (2.4% of the workforce) in Scandinavia on a five-year view. Labour markets have adapted to shocks from new technology through history, and we believe it will be no different with generative AI.

Existential threats and calls for AI regulation

There have been several recent warning calls and initiatives from influential people in AI and governments, owing to the rapid development of frontier AI models in the past year. The prevailing AI models are far from being any general artificial intelligence, and it remains to be seen whether there will ever be one. But a superintelligence remains a hypothetical risk which should not be dismissed, and there are more immediate potential existential threats from current AI capabilities within disinformation and the bioengineering of pathogens. We believe a global framework, similar to that which oversees nuclear non-proliferation, could emerge to manage the risks.

Expert views on the potential challenges and opportunities

For an example of industrial applications of generative AI, we interview Hugo Nordell. He is a serial entrepreneur from several Silicon Valley startups. He is the founder of Encube, which is developing a collaborative design review tool that aims to help industrial manufacturing companies minimise R&D bottlenecks with AI-powered solutions.

“AI, deployed meaningfully, is exceptionally good at doing repetitive work accurately, over and over. Often with significant time savings as a result. By deploying many finely tuned AI models to solve important but mundane and repetitive tasks in the hardware engineering cycle, Encube can free the hardware team to do what it does best: look at the big picture,” Hugo Nordell says in the interview.

We also interview Philip Widegren, Data Science Lead at Nordea's AI Hub (which aims to accelerate AI adoption and ensure safe and responsible use). He explains how generative AI, particularly LLMs, is making access and usage far easier for nonexperts, and how potential productivity gains are fuelling massive interest in its business use. 

“New technologies have always impacted our way of working. The first batch of changes will be similar to what we have faced before – there will be individuals and corporates who adopt to the new technology and others who don't. The ones who adopt it will most likely stay relevant and replace the ones who don't,” says Philip Widegren in the interview.

And who better to interview about AI, than an AI itself – OpenAI's chatbot ChatGPT. We ask it about AI and what it will mean for humanity, again and again, each time instructing it to answer as a different persona, such as Gollum and Donald Trump.