Nordic Business Forum 2023: notes on the AI theme

We spent the end of September at Helsinki Messukeskus for Nordic Business Forum 2023. The event was organized around four subthemes: leadership, innovation, resilience, and artificial intelligence. From what we saw, AI was clearly the most prominent of the four, both on stage and in audience conversations during the breaks.

General atmosphere: a mix of interest and concern

After the talks covering the current state of AI, a noticeably worried tone ran through many of the conversations we had. Some attendees openly said it was a relief whenever the program broke into lighter topics like leadership or organizational culture.

People we talked to brought up AI often when discussing their own work, but when we asked what they had actually done or tried, the answer almost always stayed at the level of ideas. The topic clearly mattered to them, but for most it hadn't yet translated into hands-on experience.

Mikko Hyppönen's "Underestimated AI"

The most concrete AI talk of the event was Mikko Hyppönen's. He brought along examples of the latest developments that most of the audience hadn't seen before: deepfake audio, generative music, output from image generators.

Hyppönen mentioned having first read about AI as a thirteen-year-old in 1983, which puts the promised revolution roughly forty years behind schedule. What finally made it possible was three technologies maturing at the same time:

The internet and digitalization moved enormous amounts of content into digital form, giving machine learning something to work with.
Cloud services provided the storage for the massive datasets.
Compute reached a level where specialized AI accelerators like NVIDIA's H100 enable the scale of today's generative AI. Hyppönen used the difficulty of 3-nanometer chip manufacturing to illustrate just how demanding modern silicon has become, and pointed out that the phone in your pocket today would have made the world's TOP500 supercomputer list twenty years ago.

On the scientific side, the key breakthrough was Google’s Attention is all you need paper and the Transformer architecture it introduced. That gave machine learning models a long enough attention span for generative AI to actually start working.

What generative AI looks like in practice

The principle behind a generative model is that it is trained on such a large body of material that it learns to produce similar content. Hyppönen illustrated this with Midjourney Batman images, "Barbie versions" of himself and others, and a deepfake of Aqua's Barbie Girl sung in Johnny Cash's voice. Behind the demos sit questions that remain unanswered: on what terms can models be trained on existing creators' work, and who owns what those models produce?

What this means for businesses

Hyppönen named OpenAI, Anthropic, TikTok, Google, and Meta as the key players in AI today, and noted that recommendation algorithms have already paid off clearly for services like TikTok, YouTube, and Instagram. His message to executives was blunt: companies that ignore the AI revolution will end up like the ones that ignored the internet. He gave practical examples ranging from sales automation to anomaly detection in physical security, and mentioned that he himself uses Anthropic's Claude to summarize research papers, including across languages.

The other side: risks

The same talk walked systematically through the downsides:

Rogue models like WormGPT and BlackGPT. Safety restrictions can be stripped from openly available or downloadable models such as LLaMA, and the modified versions are used to write phishing emails and malware.
Deepfake scams. For example, videos that look and sound like Elon Musk are being used to lure people into cryptocurrency fraud.
Future CEO impersonation. Hyppönen's prediction: a Teams call from "your boss" that looks and sounds authentic in real time.
Zero-day vulnerabilities. AI can find unknown security holes, which is valuable to defenders but equally arms attackers.

The core point: we're underestimating this

The talk's underlying argument was that the impact of AI is being systematically underestimated. The framing: average human IQ is 100, and right now people with IQ 100 are confidently predicting what will happen in twenty years when we will be facing AI with a vastly higher IQ. The outcome could be human extinction, utopia, or anything in between. The next comparable leap, Hyppönen suggested, may come from quantum computing, and his guess is that superhuman, self-aware AI will appear within his lifetime.

Other AI talks

AI also featured in three other talks worth noting:

Amy Webb framed AI as a tool to sit alongside more familiar technologies, and emphasized that before adoption you need to understand the data sources, the training process, and the biases that can carry over into the models.
Mo Gawdat spoke about the pace of development and the ethical questions it raises, and suggested thinking of AI as your most capable team member, with a portfolio of tools worth experimenting with in parallel rather than picking just one.
Scott Galloway took the economic and market angle, describing AI as a "time machine" that produces clear winners and losers, with the biggest winners being the players building infrastructure and LLMs.

What this means for us

For us, the event confirmed that our data-driven development work is heading in the right direction, and that the leading edge of this technology is worth following closely.

Looking for expert solutions? Discover Crestia's professional services today.