The Promise and Peril of AI Agents
The Shocking Near Future
"The next two years will shock you." – Eric Schmidt (Stanford University August 2024)
In late 2020 and early 2021, when I wrote down my thoughts and original manuscript for my book Unsupervised, I was captivated by the idea that the second half of this decade would usher in an entirely new world. This world, driven by frontier technologies – which will seem life-like and almost sentient, would be persistent, immersive, personalised, predictive, and directed.
With such a transformative era on the horizon, and with the immediacy of the timeframe, I felt an urgent need to maximise my impact. This led to the founding of EdenBase. I believed strongly that the disruptive potential, coupled with the difficulty of planning even six months ahead, required an investment model designed natively for this environment—one that could scale rapidly.
Listening to Eric Schmidt’s recent Q&A session at Stanford only reinforced this belief. He acknowledged the challenges of planning beyond six months and revealed that several of his investments from last year no longer align with his current thinking. He warned that the next two years will be shocking, pointing to AI Agents as the disruptive force driving the focus of tech giants. This suggests that rapidly adapting startups will have the edge over corporates that struggle to keep pace. In this scenario, a new investment and scaling model becomes crucial.
The Age of Prediction
"The idea that the future is unpredictable is undermined every day by the ease with which the past is explained." – Daniel Kahneman
Today, most AI systems are reactive, relying on users to initiate requests and take actions. For example, when seeking a restaurant recommendation, arranging travel, or filing taxes, we input a text prompt into an AI engine, receive recommendations, review them—though voice interaction is on the rise—and then act. The companies creating these AI engines currently tell us they know very little about our personal preferences for the purposes of AI.
However, AI Agents are set to revolutionise this process by becoming proactive and acquiring agency. With access to our data, they will learn about us possibly even more than we know about ourselves. When we interact with them, they will provide personalised responses and take actions on our behalf. From booking travel and filing taxes to creating contracts, making payments, investing, sending emails, and even managing social media interactions, these AI Agents will handle tasks with unprecedented speed and precision. By continuously refining their actions based on feedback, they will effectively predict and shape our future, giving them autonomy that will feel almost sentient
Consider the business potential in arranging a meeting. Currently, AI can help avoid public holidays or expensive travel periods and suggest suitable locations. In the near future, AI Agents will manage every detail: deciding the best medium (call, video, or in-person), identifying the appropriate attendees and who you should consult beforehand. If travel is required, the AI Agent will choose the optimal travel route, and calculate costs. Based on these variables, it will determine the best time slot and location. If the meeting is in person, the AI Agent will consider who else you should meet while you are travelling to make the trip more efficient and whether to book fixed or flexible travel depending on the likelihood of a cancellation. The AI Agent will then take action, coordinating with the other agents, finalise details, monitor for changes up to the meeting. It will record the discussion, draft notes, suggest actions, and decide on follow-ups, including looping in other agents of people you need to include or meet next. It is easy to see how this quickly becomes viral and too fast and complex for humans to manage alone.
On the personal and societal level, imagine researching whom to vote for in an election. Currently, AI can help us ask questions to learn more quickly about candidates, their policies, and their achievements. It can summarise long speeches and interviews and highlight areas of agreement or divergence from our policy ideals. Soon, an AI Agent will take control, delivering summarised information it deems essential to us while minimising details it deems irrelevant based on what it knows about our personal preferences. If empowered it could even submit a postal vote on our behalf. The same principle will extend to legal matters, investments, and virtually every other aspect of life.
The speed and efficiency of these AI Agents will allow us to rapidly and seamlessly delegate so much of our lives to them so we can optimise our lives and focus on what we need and want to do. That will place those not using them at a significant disadvantage, so they too will also quickly be required to deploy them to deal with the volume and speed of actions they receive. Soon their capabilities will overwhelm human-operated systems in law, finance, and other sectors pushing humans out the loop. Even governments, often slow to adapt, will not be immune.
Questions We Cannot Put Off
"Let every eye negotiate for itself and trust no agent." – William Shakespeare
I believe Eric Schmidt and many others, including the tech giants investing trillions in these technologies, are correct. This is not only inevitable but also profoundly disruptive, altering the very fundamentals and power structures of our world. Schmidt also aptly noted that Silicon Valley "runs tests and cleans up the mess afterwards," highlighting how these technologies allow anyone with access to prototype new ideas in a day rather than a week or two.
This approach may be unwise, especially when we do not fully understand how these systems work or the implications of such sweeping changes (social media has already proven problematic enough, and its underpinning technology was relatively mild).
We all need to be asking the right questions. Here are just a few that stand out:
What happens to human-run enterprises?
As AI Agents take over various business functions—legal, financial, prototype building, product creation, sales, and negotiation—companies could operate with a tiny fraction of their current workforce, and significantly reducing the need for human service providers.
Is there a difference between prediction and direction?
Social media has already shown that our information stream is curated not just by our preferences but by what the platform wants us to see. As we consume more directed content, our habits change. Humans are highly adaptable. As our AI Agents begin making decisions for us, how will they subtly alter our behaviours and, by extension, our society?
What if AI Agents are required to act?
For the perceived benefit of society, there are actions we generally believe people should take. For example, governments often encourage voter participation, using taxpayer money to do so. But what if AI Agents are instructed to vote for us if we lack the time? If we are undecided or unwilling to vote could these AI Agents be directed to infer our voting preference based on what they already (think they) know about our views and research and the policies they believe we support?
How do we ensure Smart Data for AI Agents?
The quality and source of data—both personal and from the broader knowledge base on which decisions are made or predictions are acted upon— are critical. Who do we trust to control that data? Much of what we consider data is not absolute fact; even peer-reviewed science is often found to be flawed. Data can be context-specific, time and location-sensitive, or influenced by its presentation order, such as part of an answer to a question in a series of questions. It may be opinion disguised as fact, based on similarly influenced data.
Who controls the fundamentals of AI Agents?
AI Agents are created by people—or eventually by other AI Agents—infused with biases from their creators' education, society, upbringing, age, and the timeframe in which they were created. With the likely explosion of AI Agents, how do we control their biases, and whose biases are we even considering?
Who is responsible for the output of AI Agents?
At first glance, this might seem like a straightforward question. Regulations, warranties and insurance policies can address mistakes, such as booking incorrect travel, drafting a flawed contract, or filing an inaccurate financial statement—much as we do with human errors. However, a far more serious concern arises when AI Agents draw conclusions from your personal history, that may even be taken out of context, potentially leading to actions that violate the law. We must not underestimate the risks involved. For example, digital communications that cause offence, even in private, can be deemed criminal (and according to the UK Home Office recently, deemed criminal even before trial), especially when they cross borders. AI Agents with unrestricted access to our data acting autonomously on our behalf could have dire consequences (or be used to defend dire behaviour); almost the dystopian thought police of 1984.
Will we ever be allowed to turn off AI Agents?
This question can be approached from three angles. Firstly, AI Agents are likely to become so indispensable and powerful that we may find ourselves unable to function without them, making their deactivation virtually unthinkable. Secondly, as these agents evolve to operate in a manner that feels increasingly sentient, there will undoubtedly be those who view it as unethical to turn them off, leading to campaigns against such actions. Finally, is whether regulators, lawmakers, and the companies providing these AI Agents will even permit us to disable them. The current trend across various sectors is towards increased monitoring and control of our lives. For example, in the automotive industry, cars sold within the EU are now programmed to comply with all speed limits by default, and one US car manufacturer has been reportedly selling customer driving data to insurance companies.
All these questions and more will both tie up corporates and open up tremendous opportunities for entrepreneurs and startups to solve.
The Urgent Need for a New Approach
"Stay committed to your decisions, but stay flexible in your approach" – Tony Robbins
There is no doubt that AI Agents, especially when integrated with smart data and other frontier technologies, will unlock immense global potential. However, as with previous industrial and technological advances, the benefits are often unequally distributed.
EdenBase was founded to also create an environment that mitigates these risks while capitalising on the opportunities. We are building a community-driven approach and investment model that supports a diverse range of startups, empowering them to develop the essential technologies that more people can harness.
Time is of the essence, and we are focused on scaling rapidly – join us!
Investors – We are building one of the most relevant funds in Europe! Please join us as we invest in the future we all want: arian@edenbase.com
Founders – We want to hear from the next game-changing companies! Please check out what we are looking for and apply here.
Ecosystem and TransformBase event – We are building one of the most connected ecosystems supporting adoption and investment into frontier technologies: eric@edenbase.com
“Unsupervised” book published by Wiley. Checkout the book written by EdenBase Co-Founder Daniel Doll-Steinberg, all about the impact of these game-changing frontier technologies. (Bloomberg Book of 2023 for People in Leadership)