05-01-2024 10:45

AI is not likely to replace people, but people with AI may replace those without it

OpenAI’s launch of ChatGPT sparked a surge of interest in artificial intelligence (AI) and its productivity-boosting use cases, as well as fears of human job losses. Don’t panic, says Philip Widegren, Data Science Lead at Nordea’s AI Hub. New technologies have always reshaped our ways of working. Find out more in the interview from Nordea On Your Mind.
Robot finger meeting human finger

To understand the new wave of hype for AI, its drivers and potential implications, Nordea On Your Mind author Viktor Sonebäck turns to Philip Widegren, Data Science Lead at Nordea's AI Hub. He describes how generative AI has opened a new interface (through Large Language Models) making access and usage far easier for non-experts. Potential productivity gains are fuelling massive interest in business use. There are risks of new AI technologies being abused for disinformation and destabilisation, but the outlook for human employment is not necessarily bearish. AI is a productivity tool for humans.

Viktor Sonebäck (VS): Can you share your background and your current role at Nordea?

Philip Widegren (PW): My academic background includes an M.Sc. in Applied Mathematics from KTH Royal Institute of Technology. I initially started my career in the area of AI as a quantitative trader at SEB. My journey further led me to BCG and BCG Gamma, where over four years, I was part of driving large-scale digital transformations using AI, ML and Data Science across various industries and clients, with a special focus on areas such as pricing and personalisation. Since August 2022, I have worked as a data science lead for Nordea’s AI hub, focusing on commercial units and leading our Generative AI initiatives across the bank. Nordea's AI hub exists to accelerate AI adoption in the bank and to ensure safe and responsible usage.

VS: LLMs, and generative AI in general, have reached an unprecedented level of hype in recent times. Why is this? Why now?

PW: Generative AI and in particular Large Language Models have been one of the hottest, if not the hottest, topics in the industry since the release of ChatGPT back in November 2022. This is despite the fact that Generative AI only accounts for 30-40% of the economic impact of all AI technologies, according to McKinsey.

One of the primary reasons for the hype surrounding generative AI is its enhanced accessibility. Previously, developing AI applications to realise business value required highly skilled data scientists; however, OpenAI’s release of ChatGPT has significantly lowered such barriers to entry, meaning anyone may now access state-of-the-art AI technologies. Moreover, through leveraging ChatGPT and other generative AI applications, employees themselves are attaining productivity increases from directly utilising this technology.

Other factors include the fact that people can easily build new services. There is an ongoing battle, including Google, Microsoft, Meta on a global stage, to develop the best services and foundation models. Everyone is watching. It is important to also mention that LLMs are not that "smart". Essentially they just predict the next words based on the instructions you have provided, meaning they are very prone to hallucinations if you do not provide enough context.

If we for example zoom into how the software developer experience can be improved using generative AI then studies from GitHub regarding its copilot claim the following: software developers feel more productive, efficient and satisfied. They have measured a 55% productivity gain when writing code. It is important to emphasise that many software developers spend many hours a week on non-coding tasks. Also, navigating large complex codebases is much more challenging than creating new codebases from scratch. In the not too distant future, when companies recruit developers, the developers will assume that they will get a copilot and the question will rather be which copilot the company uses instead of if developers have a copilot, similar to how new hires today expect to get a computer and mobile phone when joining a company.

VS: How has the development progressed over the years? Where are we right now in terms of capabilities and where are we heading? Will we see continued explosive growth in this space or can one expect some kind of satiation?

PW: If we start with what has led us to where we are today, then numerous influential research papers have contributed significantly to advancing this technology. One of the most impactful works is a 2017 paper titled "Attention is all you need" which reported many pivotal findings. It demonstrated how words can be contextualised, such that "bank" refers either to a riverbank or a financial institution, depending on whether the full phrase is "the bank of the river" or "money in the bank". Prior to this work, distinguishing between such different meanings for the same words was challenging. Another important development is in distributed computing where enormous models can be trained in parallel across several GPUs. This enables models to have a larger number of parameters as well as bigger training datasets.

Today this technology has become so powerful that global thought leaders, people like (MIT physicist) Max Tegmark, are arguing that we should pause the training of any AI systems more powerful than GPT-4 (the latest version of ChatGPT) because so much capital is investing in increasingly more powerful models, while relatively limited funding goes to developing control systems for the models.

On the other hand, I believe we are close to the top of the hype curve if not on top of it. Many of the text models we have today are more than enough to cater for the business use cases corporates are coming across. Moreover, many corporates will be busy exploring and adopting this new technology over the upcoming years. Just because these models can perform a task doesn’t necessarily mean that they should, in terms of either positive ROI or ethical considerations. I would even argue that the focus on extracting high business value with this technology is now centred around integrations with data and knowledge rather than the capacity of these models.

It is important to focus on your business outcomes when you start building generative AI solutions. As with any other technology, generative AI is not the solution to all potential problems you have in your organisation. According to a joint study between Harvard and Boston Consulting Group, the consultants completed significantly more tasks at a much faster pace when using generative AI. The bottom performers had the biggest gain when using this technology. Importantly, they also find a significantly negative impact on the performance in the area of business problem solving for tasks outside of the tool’s current competence.

Other generative AI technologies such as image generation have made enormous progress over the last few years but it can still be difficult to get all details correct for an AI generated image.

VS: What is driving the current rapid and renewed focus on AI? Interesting/striking new features? Specific applications? Massive potential usefulness across society?

PW: For someone like me who has been part of many digital transformations using AI, I would say that the main driver of the renewed focus on AI is around the massive productivity gains you now can achieve. There is a win-win situation when employees feel that the usage of these tools will move the focus from more repetitive, time-consuming and sometimes boring tasks to more value-creative tasks.

If we focus on society overall, then as much as I truly love the potential of this technology, I must also reflect on how this technology can be abused. You can use the technology to drive business value, but there are also criminals who can and will use it to their advantage. One example being that you can now with just a few seconds of someone's voice create audio sounding like that person. Another example is the amount of fake news that this technology can create, it is very easy to create deep fakes where you provide disinformation in a realistic way. The ability to fabricate convincing videos of public figures, such as a prime minister, presents a significant risk of misinformation that could potentially destabilise a country.


Some jobs will become redundant and new ones will emerge. For instance, how many have heard of 'prompt engineering' before this year?

Philip Widegren, Data Science Lead, Nordea AI Hub

VS: How will people be impacted by AI? New services? New applications affecting one’s work? Process automation making some jobs redundant?

PW: 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.

The way I see it, this new technology enables us to focus more on the value-adding activities and less on repetitive and boring tasks. Some jobs will become redundant and new ones will emerge. For instance, how many have heard of 'prompt engineering' before this year? Another example is around creators versus editors. This technology has the capability to create new content at a much faster pace than any human being, meaning there will be much more content created. This might transition people from being creators to editors, etc.

With all of this in mind, it is difficult to not mention the co-pilot concept. The latest AI technology is here to help individuals and employees. I don't see that it will replace the human being in any short view, but people with AI might replace people without it.

Nordea On Your Mind

Nordea On Your Mind is the flagship publication of Nordea Investment Banking’s Thematics team, which produces research for large corporate and institutional clients. The research does not contain investment advice and typically covers topics of a strategic and long-term nature, which can affect corporate financial performance.

Top decision makers at Nordea’s large clients across the Nordic region receive Nordea On Your Mind around eight times per year. The publication’s themes vary widely, and many are selected from suggestions by clients. Examples of covered topics include artificial intelligence, wage inflation, M&A, e-commerce, income inequality, ESG, cybersecurity and corporate leverage.

See more from Nordea On Your Mind
Nordea On Your Mind