Becoming Dependent in the World of AI and Why do We Need a Don Quixote To Save Us.
We were not designed to do boring tasks.
Our nature demands and rewards exploration, constantly looking out for new challenges and expanding our understanding. Yet, there are many seemingly boring tasks our society, on the other hand, demands to be done. We are trying to outsource them all to AI, without considering the hidden costs and risks.
Algorithms can automate and speed up a lot of the boring and non-creative work, completing it faster and more accurately. Obviously, this is a great thing, and has enabled us to dramatically improve the quality of our lives, but we cannot let every task be overtaken by an algorithm, as there is a need for human supervision. More and more companies are buying enterprise versions of the most capable AI models and replacing whole departments of people, who have experience and agency in their job, with less experienced staff equipped with the latest and greatest models. They change labor costs to token costs.
Meanwhile, individuals get lost, running to increase their productivity in every possible way, trying to meet the expectations put on them by our society. They see AI as a perfect way to multiple their work output, handing over more of the mundane tasks they have to do, and increasingly more of the ambitious and demanding tasks as well.
When in Rome, do as the Romans do
We live in a society that expects constant growth and gains. There is an expectation that all companies will increase their revenue year by year. That’s how our economy works. The environment demands the perceived increase of productivity, and the companies need to comply by showing the people who lend them money that they are working on it, and they are working on it hard.
Businesses have went from being seen as innovative and forward looking when implementing AI, to being seen as too conservative and incapable of adapting to the current needs when failing to adopt AI across their business processes. This shift of perspective put a lot of pressure on the directors and managers of all companies to incentivize the use of AI tools to increase the company’s overall (perceived) productivity and as a consequence, the employees need to comply.
We have seen this trend with technologies in the past, companies rushed to make use of technologies like Cloud or experiment with Big Data. This time it is different, it is different because people have to let go of their understanding of core parts of what they do, in the name of perceived productivity gains, which are expected of them. This is a trade-off because they can’t spend as much time as previously on the same tasks, they can complete them faster with the help of AI assistants.
This paradigm shift did not stop at affecting businesses only. Students across schools and universities are incentivized to use AI more and more. Teachers themselves use it to grade their pupils’ work. Dead Internet theory, according to which there are no human interactions on the internet anymore, only bots interacting with each other, seems to be realized in classrooms across the globe.
Assignments get harder and more time-consuming as universities can’t find proper ways to cope with the rise of new technologies, raising the bar with the expectation of everyone cheating on their tasks. At the same time, students have more trouble understanding the content of the courses, but do not realize it until the exams are due. This problem also affects well-known universities.
Is there a puppet master?
In Cervantes’ Don Quixote, the main character and his trusty squire, Sancho Panza, settle down in an inn to spend a night, when a well-known puppeteer comes in and starts giving a puppet show. The courageous Don Quixote interrupts the show two times before launching a full-scale attack, massacring the puppets with his sword, barely missing the puppet master himself. He did so because he felt that some of the puppet characters needed his help. While Don Quixote admitted his attack was a mistake, regretting it after realizing what happened in reality, he surely did put the puppet show to an end.
While there might not be a single puppet master in our world today, there surely are powerful actors that have incentives in wanting to implement AI as fast as possible and wherever possible. As time goes on, more and more progress on this aim is being made, and there doesn’t seem to be an actor powerful enough to put this to a timely halt, so that we can consider what is really going on and how we can safely progress into the future. Individuals and whole economies are losing independence in doing work without using AI tools.
Several problems arise. Obviously, a situation where a few particularly strong companies have essentially a monopoly on the most powerful models, which will be running a big part of the world’s economies, is not ideal, given that they can raise their API prices or equivalently, cut down token limits as they wish. Then, there is also a bunch of concerns about the safety of such wide adoption of these tools, given that we can’t perfectly control their usage and don’t know their exact limitations. These systems can be implemented in a safety-critical environment without people paying too much attention. This year, for example, McKinsey’s internal chatbot got compromised, interestingly enough, by another AI agent leaking secret information about some of the biggest companies in the world.
Resisting the urge to use AI models gets harder because not only are the environments that surround us demanding it more, but the AI companies are giving ordinary users (or ordinary paying users) access to their more capable models for a period of time, only to lock them behind more powerful paywalls afterwards. This is how Antrophic, the company behind Claude, decided to release their new Fable model. They explicitly say their Pro plan users will have access to it for the 12 days after its release, and they will move it to a more expensive pricing model afterwards. This way of marketing is nothing new, but it is hard not to see the resemblance between such practices and drug dealers offering a sample of some new stuff to get their clients hooked on it.
Remember nuclear warheads?
With the outsourcing of doing the actual work to AI systems, we are losing important working knowledge and the talent who knows how to do key things. This is not a problem we haven’t seen before. Fogbank is a codename of a secret material, critical to craft nuclear warheads, whose production process was lost as the original engineers working on it retired or have deceased, without passing on the tacit knowledge needed to create it. The National Nuclear Security Administration spent 5 years and tens of millions of dollars to reverse engineer the material and how to produce it afterwards. The knowledge on how to create it was not written in any documents, and it could only be passed on by practicing engineers.
A similar process is prone to happen as we outsource real work across all sectors to AI and have fewer and fewer people actually thinking about it. Not documenting working knowledge properly, we are bound to see it disappear. After some time, we might find ourselves in a similar situation as the US military did in the early 2000s, noticing that we forgot some fundamental process to craft something we need. Only, unlike the US military, we might not have the time or the budget to make up for that loss.
Become irreplaceable
Knowing how to use AI seems to be an essential skill in today’s workplace, but knowing how to work without AI might be a critical skill in the workplace of the future. This should not be understood in the context of work only, as we have a tendency to take the easy route whenever possible. We should think hard about stuff that matters to us, and as I explained in one of my previous articles, don’t outsource it completely.