At the AI Strategy Summit, industry experts gathered to discuss the most pressing issues surrounding GenAI. The focus was on Europe's AI challenges, immense potential, responsible usage, and practical ways to get started. Here are my five key takeaways from the conference!

Key takeaway: Companies like OpenAI, Facebook, Salesforce, and Microsoft have invested billions of dollars and plan to increase their investments further to develop advanced, general AI models. For most companies, it's inefficient to build their own models or try to recruit similar expertise. Instead, they should focus on understanding, using, and applying these models to their specific business challenges, best suited to be solved with AI. On the bright side, most companies can avoid these enormous investments!

1. Sweden and the EU Lagging Behind in GenAI

Multiple speakers emphasized that Europe lags behind in global AI development. Only four of the world's top 100 AI companies are European. In Sweden, the situation is particularly challenging: only 18% of CEOs report having implemented generative AI, compared to the global average of 33%. In Silicon Valley, for example, over 260 AI events are held monthly.

The EU's upcoming regulations are seen as a hindrance to domestic innovation and the use of non-European AI solutions (e.g., OpenAI). The U.S. has extensive AI development and limited regulations, while the EU has low levels of development and extensive regulations. The outcome of this disparity remains to be seen.

2. Responsible Use of AI

Almost all speakers highlighted the importance of responsible AI use and the risks involved. Discussions centered on challenges like biases embedded in language models, hallucinations, and the risk of manipulated election results.

Sweden and the EU appear overly focused on the risks compared to the U.S., perhaps because the technology isn't developed here. The unknown can seem scary, leading to caution and strict regulations. However, by not fully embracing the technology, we risk losing competitiveness, and a balance needs to be evaluated.

3. Consensus on the Potential of AI, but...

A recurring theme was the tremendous potential of AI. Reports from companies like McKinsey, PwC, and Cognizant have estimated that GenAI could generate several trillion dollars in added GDP value by 2030. Although these estimates are uncertain, everyone agrees on the massive potential. Automation, improved customer experience, upskilling in various areas, and streamlined workflows were some of the highlighted opportunities.

Currently, the most significant value lies in "traditional AI," like machine learning and recommendation engines. For instance, 80% of what we see on Netflix results from their recommendation system, and 35% of Amazon's revenue comes from AI-based product recommendations.

Generative AI, on the other hand, is believed to have even greater potential. Since 80-90% of all data is unstructured, GenAI can transform it into structured data, generate new insights, and create valuable information. As a result, the world is quickly filling up with artificial data. Gartner estimates that by 2025, 10% of all information on the internet will be generated by AI, and by 2030, it could reach 99%. The full implications of this trend remain unknown.

Despite the hype, we have yet to see this value fully realized.

4. Lack of Real-World Use Cases

I've attended many events promising practical applications but mostly focused on early-stage tests and proof-of-concepts (POCs). This isn't surprising since the technology is still relatively new. However, this event stood out, as you'll see in the next point.

One speaker mentioned that 87% of large language model (LLM) projects fail, meaning they don't reach production. Instead of viewing this as a failure, it's a positive sign that so many are experimenting. Currently, LLMs are too slow, energy-intensive, and unreliable for many applications, but failing to prepare and learn would be a big mistake.

Some companies throw AI at every problem instead of focusing on areas where it's suitable, but experimentation is needed to identify the right applications. What seems unsuitable today might become possible with the next generation of AI models.

Examples from Bonnier:

As a company working extensively with text and images, Bonnier is well-positioned to leverage generative AI, even though investigative journalism still requires personal contact with sources.

• An HR bot that answers common questions about vacation and time off.

• A Slack bot that can quickly retrieve and summarize data from their article database.

• An editorial assistant.

• Personalized, AI-generated newsletters.

• Customer service support that increased the number of resolved cases from 40% to 70%.

Examples from Klarna:

Klarna has over 100 projects across the organization, and 75% of employees (targeting 100%) use AI tools, showing the extent of their adoption.

• They have a bot that can create SQL queries from plain text, democratizing and simplifying data access.

• Parts of customer service are automated, reducing the workforce by 700 agents while significantly shortening response times.

• Marketers use a copilot to create copy, with 80% of all copy now generated through the tool. AI-generated images are twice as common as real ones.

• An internal bot has answered 500,000 questions about internal processes.

  
  

5. Get Started with AI!

Many speakers stressed that companies must start using AI because many employees are already using it in their work, a phenomenon known as "shadow AI." Banning the use of AI tools is often ineffective because people follow the "path of least resistance." Instead, provide everyone with a copilot tailored to the user's needs and training to increase exposure and understanding of AI's potential.

Tips highlighted:

• Cross-functional collaboration between data, business, developers, and AI experts.

• Management understanding and support, which often lacks today.

• There was a debate on whether a centralized or decentralized strategy is best for AI adoption, but good examples were found with both approaches.

• Once the problems are identified, the next step is to find an "AI-problem fit" – problems suited to AI solutions. Avoid using AI just for the sake of it.

Do you need to develop your models? For 99% of companies, this isn't practical. OpenAI, Facebook, and Dell spend billions on training their models, and most companies can't keep up with their pace. Instead, leverage their investments and use their models. In the worst case, fine-tuning is enough and much cheaper. There's a broad selection of fast, smart, and open-source models available.

Consider how AI will affect your business, suppliers, and customers. A stock photo company might benefit from not paying royalties to photographers, but their customers may also prefer creating their own images instead of browsing the collection. Perhaps it's better to sell data for training models, but for how long will that remain relevant?

Ending notes

Start using AI today, not because it solves every problem now but because the technology improves rapidly! Thanks to all speakers and participants for an inspiring and insightful event!

Speakers: Martin Elwin, Lina Hallmer, Patrik Liu Tran, Göran Lindsjö, Fredrik Heintz, Jörgen Warborn, and Peter Kurzwelly.

"AI is Alfred, not Batman – a helpful assistant that solves problems rather than a savior!"