At Oslo Business Forum, Amy demonstrated what’s wrong with the way business leaders are thinking about AI today. She shared a framework designed to help them shift their thinking from short-term to long-term and make better decisions for the future.
Amy recounted a recent event where she was cornered by two successful businesspeople, one a venture capitalist and one a banker. In separate conversations, they each sought her reinforcement of an audacious theory they had developed: Super intelligent AI is going to become the world’s dominant power and replace the human race.
“These conversations exemplify two big things wrong with how business is thinking about AI,” Amy said.
It’s too early to predict how AI will impact business. “AI is a long-horizon trend with a lot of development still ahead,” Amy said. The term artificial intelligence is shorthand for many different technologies. There are interdependencies in various stages of development, and at this point, no one knows the impact AI will have in the future.
We’re so focused on the present that we’ve lost sight of the future. “AI is not a revolution; it’s an evolution,” Amy said. Many leaders are worried about the short-term impact of AI, but nothing will change overnight. As AI evolves, it will require some parts of our businesses to be reimagined.
Amy blames fundamental misunderstanding for the mindset many leaders are working with today. And that misunderstanding is logical. “This is the most complex operating environment I’ve seen in 20 years,” she said.
“AI is not a revolution, it’s an evolution.”
To help them shift their current mindset, Amy took leaders on a journey to ground them in what AI is, what it can and can’t do, and where she believes it’s headed in the future.
The Evolution—So Far
Conceptually, AI has been in some form of development for hundreds of years. Although we’ve long envisioned machines that might “think” the way humans do, the most important development has only happened over the past six years.
Amy pointed to a 2017 paper by Google titled “Attention is All You Need,” which asserted that our ability to communicate effectively depends on focusing on the most critical aspects of a conversation or text. It demonstrated how computers could comb data to determine which parts are most relevant to each other, in essence, becoming a hyper-efficient translator. This is the transformer that underpins the large language models (LLMs) used in generative AI today.
AI began to bubble years ago but only really broke the surface in November 2022 when OpenAI introduced ChatGPT. ChatGPT was the first form of generative AI widely available to the public.
Amy explained how ChatGPT and other generative AI tools have been “trained.” The first iteration of ChatGPT was trained on a database of 7,000 books in the public domain. The second version added millions of web pages. The third version added a massive corpus of books, websites, Wikipedia, social media posts, and more. Now, no one knows the data sources of GPT4. “As systems start to get commercialized, the tech companies are less interested in telling people how they built them,” said Amy.
Amy used a simple query seeking a recipe for farikal, a traditional Norwegian dish, to demonstrate how generative AI recognizes, translates, and predicts text and images. She first asked for the recipe in English, then in Norwegian, then in the voice of the author J.K. Rowling. Each query produced a new, more detailed response.
“This is an example of assistive computing,” Amy said. “This is what the future of AI really is; not wiping out jobs but changing jobs.”
The Implications
Amy pointed out that our lives are full of tools that are completely invisible to us. She used an example of a calculator, once a clunky tool that students carried around in their backpacks. Now, if you have a phone, you carry a calculator in your pocket every day. It’s become something we take entirely for granted.
“That’s where we’re headed with AI,” Amy said. “If you can reframe your thinking on AI as assistive, you’re going to see the opportunities and the risks very differently.”
Amy demonstrated three tasks using generative AI to show the benefits and drawbacks of these tools and foreshadow the implications as individuals and businesses more widely adopt them.
1. Tell Me About Cats
First, Amy searched for “cats” in Google and found abundant information. She pointed out that there’s no telling where the information came from, which indicates how we are increasingly removed from the source of the information we seek and receive. The implication? “This is the end of the internet as we know it,” she said. “It fundamentally changes the economics, mechanics, and use cases of how the internet works.” This is good for people who want information fast but bad for businesses’ traditional product and marketing strategies.
2. Search My Screenshots
Next, Amy described her penchant for taking screenshots and promptly forgetting why she took them. She was frustrated by the inability to search and decipher them—until there was an app for that. A new generative AI tool allows you to upload screenshots, which it then scans and summarizes. The implication? “Going forward, everything is searchable,” Amy said. This is good for people with information overload and helpful for gleaning insights. It is bad for our reading and critical thinking skills.
3. Solve My P&L Problem
Finally, Amy searched for a publicly available P&L statement and found one from a hospital. This hospital was operating at a significant loss. She uploaded its P&L statement to Bard from Google and asked the tool to find novel ways to reduce expenses and improve margins. In 32 seconds, she received a summary analysis. This is a task that would typically take days, if not weeks, for a human to perform. The implication? “Workers are going to have to learn how to delegate in new ways,” Amy said. This is good for businesses developing their AI capabilities to reclaim employees’ time, but bad for professional services firms that rely on the volume of billable hours.
AI isn’t Magic
“When we think something is magic, we stop investigating ‘how did they do it?’” Amy said. “‘How did they do it’ and ‘why did they do it’ are questions I want you as leaders to always be asking.”
To demonstrate why it’s so important that leaders remain vigilant, Amy reminded them of how large language models work. Today, AI relies on Reinforcement Learning with Human Feedback (RLHF). This means that LLMs require a lot of people to review and annotate data to train the systems. Much of this work is done by people in emerging economies. “All they’re doing is labeling data all day long,” she said. “It’s very tedious work and also very subjective.”
Amy showed leaders how important context and understanding are to label data accurately. Using a tool called Midjourney, she submitted a query for a CEO of a large hospital system. The tool returned images of four mature white men. Her second query, for a CEO of a mid-sized hospital system, returned similar results with slightly younger and more attractive men. Her third query, for a CEO of a small, rural hospital system still returned images of men. Wondering how she could get the tool to return an image of a woman, she searched for data on which city in the U.S. has the highest concentration of women. She used the answer, Jackson, Mississippi, to further refine her query for Midjourney. The result was still an image of a man.
Amy used this example to demonstrate that the people who are labeling and training generative AI systems have possibly never seen or heard of what you’re searching for. They have a different set of beliefs or understanding. This calls into question whether we can trust the data we receive from AI. But it also offers us reassurance that it is improbable for AI to replace highly specialized professionals, like healthcare providers.
“What are the next-order impacts for healthcare and similarly specialized fields?” Amy asked.
She encouraged leaders to reflect on questions like:
- How will you upskill your workforce?
- How does the educational system need to evolve?
- Are you ready to hand over your data to big tech? And what are you going to get in return?
“This is the near future,” Amy said. “You still have to focus on what’s coming—the farther future.”
From Science Fiction to Reality
To demonstrate the advances AI has made and those yet to come, Amy gave the audience a challenging cognitive task, asking them to determine how to stack multiple objects of different shapes, sizes, and materials. “This is analogous to something you as leaders need to be doing all the time,” she said. “The answer is not just on the internet.”
While most leaders were still trying to figure out their approach to the task, ChatGPT completed Amy’s request in less than 30 seconds, describing the order and manner in which it would stack the objects she had named. “This is what happens when AI becomes ubiquitous and invisible,” she said.
“What if answers in the world are frictionless and real-time all the time?” Amy asked. “Text is today, voice is tomorrow, ubiquitous information all the time is what comes next. This is the moment that we go from science fiction to reality.”
How to Build a Better Future: The Four Horizons Framework
Amy showed leaders how to shift their thinking to the farther future by leveraging the Four Horizons Framework. She believes that making decisions about a future we cannot predict requires us to “Think about time stretched out.”
Horizon 1: Operations
Horizon 2: Strategy
Horizon 3: Vision
Horizon 4: Transformation
In her opinion, too many leaders today are making decisions driven by fear. They are spending wasted time in Horizons 1 and 2, Operations and Strategy, when they should be spending their time in Horizons 3 and 4, Vision and Transformation.
“Stop spending your time on Horizons 1 and 2,” she said. “Start thinking further out and then connect that to your day-to-day strategies. Think about the fourth horizon.”
As they consider the implications of generative AI and other emerging technologies, Amy encourages leaders to ask, “What does this mean for the future?” They can then use their long-term vision and plan to determine what this evolution means for their strategy and current operations.
If leaders can think long-term, their fear will disappear, and they will feel like they have the time to make better decisions. “Regardless of what you hear, the future of AI is not determined,” she said. “It’s up to you and the choices that you make about the future today.”
Key Points
Questions to Consider
Want to be a part of the OBF community? Join Oslo Business Forum 2024: Courageous Leadership now!