Welcome to the Gilded Cage: who really rules America?
The tech broligarchs have invested heavily in Donald Trump but are not getting the payback they bargained for. Do big business and the markets still shape US government policy, or is the...
Artificial Intelligence
Artificial intelligence
AI will book your holiday, but won’t clean your kitchen… yet
Automated LLM-powered agents that talk to each and solve problems together represent the most exciting commercial frontier in AI, but a future of human-like machines remains a distant dream, says Michael Wooldridge...
Automated LLM-powered agents that communicate and collaborate to solve problems represent AI’s most exciting commercial frontier – but truly human-like machines remain a distant prospect, says Michael Wooldridge
I’ve spent most of my 30-plus-year career telling people to calm down about artificial intelligence (AI), but the progress of the past five years has been genuinely remarkable. The next wave of opportunity in the rapid rise of AI will see chatbots conversing with each other to uncover solutions and generate new types of content, opening up fresh business prospects.
However, despite popular belief, AI is still a very long way from accurately reproducing many simple human tasks, and even the most impressive large language model applications have limited real-world use. Significant risks and flaws also remain and must be addressed.
For me, the most eye-catching development is the emerging capability of AI to answer complex questions based on its analysis of vast reams of data and to provide real-world solutions and services. The development of “AI agents” represents a significant opportunity for businesses and consumers. An AI agent would not just help you plan a trip to Lausanne, but use web services to book the flight, ground transportation, and the hotel.
There are three levels of AI agents to watch out for, rising in levels of complexity and value, and they all have possible applications for business:
- Customer service agents that can, for example, handle banking enquiries and make changes to accounts based on customer instructions.
- Agents that automate processes using LLMs to give users ideas for a holiday and then book it on their behalf.
- Agents that can converse with each other to solve problems, such as improving regulation-heavy processes such as planning applications.
The benefits of AI are significant in many sectors
As these innovations materialize, all executives should already be embracing GenAI tools to save time at work and to boost productivity, especially for straightforward tasks such as creating presentations. I use LLMs as part of my daily workflow, and they make me more productive. If you are not doing so too, you are missing an obvious and easy win in your professional life – and in the performance of your organization.
It doesn’t stop there. AI can unlock hidden potential by leveraging the vast amounts of unstructured data stored across your firm’s computer systems. Try feeding all of your emails, policy documents, meeting minutes, and advertising brochures into an LLM and bring your corporate memory to life – the results may surprise you.
There is also tantalizing promise in supporting our creative endeavors. If you’re an ad executive looking for a new slogan for a banana-flavored milk drink, LLMs can generate almost limitless ideas. They might be a bit pathetic, but they could still spark your creativity to come up with something great.
Like all new waves of information technology, LLMs will undoubtedly create new categories of content. These will, in turn, lead to new businesses and services in ways that we can’t imagine. It may not be us that brings them to life, but they will become ubiquitous for future generations in the way that social media has become for ours.
As you adopt and experiment with LLMs in your own business and working life, it’s vital that you feed AI with the right inputs. The efficacy of AI, whether it’s focused on trained outputs or not, relies on high-quality, consistent data. So, if you want an effective AI strategy, you should first ask yourself the following questions: Where is the data coming from? What form is it in? Are we using consistent terminology? Are we applying the same standards across the board?
“Despite hugely impressive advances, we are nowhere near the point where AI could replicate human intelligence and behavior, even if it can be trained to solve complex mathematics problems in a matter of seconds.”
Complex math? Yes. Minimum wage work? No.
Despite hugely impressive advances, we are nowhere near the point where AI could replicate human intelligence and behavior, even if it can be trained to solve complex mathematics problems in a matter of seconds.
Yes, today’s AI can handle PhD-level math problems at speeds beyond human comprehension, but we don’t have robots that could walk into an unfamiliar house, find and clean the kitchen, and load the dishwasher. We don’t yet have AI that can do what a minimum-wage worker can. This is one of the quirky paradoxes of the technology – and one of the challenges of innovation: it’s moving fast, but it remains imperfect and patchy.
LLMs are powerful, and that power is available to bad actors as well as good ones.
Dangers to watch out for
More seriously, we are still grappling with key flaws and risks that leaders must be mindful of when it comes to adopting AI. Here are just a few:
- LLMs lie convincingly, and a lot: ChatGPT does not know what the truth is, and it can get things wrong in believable ways. This is doubly damaging because it’s not just wrong, it’s wrong in plausible ways that can easily mislead us.
- Bias and toxicity: LLMs are powerful, and that power is available to bad actors as well as good ones. While the models can be trained to reject obviously malicious requests, it is less effective at recognizing benign-sounding requests that result in harmful outcomes, such as a user asking for a poem about how to get away with killing a relative.
- Copyright and intellectual property: GenAI is trained on publicly available data and can mimic styles of music, art, and writing – raising valid and unresolved questions about ownership.
- Employment: Society is now facing serious questions about the social and economic impact of GenAI, particularly in roles that are automatable, such as coding and other repeatable tasks.
- Internet pollution: It is becoming increasingly likely that, at some point, most content on the internet will not only be generated by AI but trained on content generated by AI. This raises concerns about our ability to distinguish human from machine-generated content.
- Limited control of the tech: Will AI remain in the hands of a few tech giants and state entities? This concentration of power is already having consequences for the healthy functioning of economies and democracies.
And then, of course, there’s privacy. If AI is eating up everything we do and say online as part of its insatiable training diet, what should we share – and what should we keep to ourselves? My rule of thumb is to never disclose anything that I would be embarrassed to tell my neighbors or to hang on a banner outside of my house. That might sound a little extreme, but it’s better to be safe than sorry.
While we don’t have all the answers for these dilemmas, the next iterations of AI will continue to transform our world and working lives. The question isn’t whether you should be using it – it’s how.
This article is inspired by a keynote session at IMD’s signature Orchestrating Winning Performance program, which brings together executives from diverse sectors and geographies for a week of intense learning and sharing with IMD faculty and business experts.
Read the white paper on the 2025 IMD World Competitiveness Ranking.
Authors
Related
The end of corporate passivity: Why business leaders must shape geopolitics, not just react to it
To thrive in a fragmented world, business leaders must stop being passive spectators and start actively shaping geopolitics, urges IMD’s Arturo Bris...
How to thrive in a fractious world: a strategic playbook
As governments lock horns in our increasingly multipolar world, long-held assumptions are being upended. Forward-looking executives realize the next phase of globalization necessitates novel approaches....
Trump’s vision for America is not the 1950s – it’s the 1890s.
Donald Trump’s blueprint for American greatness isn’t the 1950s – it’s the 1890s – a time of high tariffs, hard power, and imperial ambition. To understand his vision of MAGA 2025, look...
Learn Brain Circuits
Join us for daily exercises focusing on issues from team building to developing an actionable sustainability plan to personal development. Go on - they only take five minutes.
Explore Leadership
What makes a great leader? Do you need charisma? How do you inspire your team? Our experts offer actionable insights through first-person narratives, behind-the-scenes interviews and The Help Desk.
Join Membership
Log in here to join in the conversation with the I by IMD community. Your subscription grants you access to the quarterly magazine plus daily articles, videos, podcasts and learning exercises.