AI
Any company that wants to be here in five years will have to integrate AI. But be vigilant: use it for organizing and scaling, and never hand it the power to decide.
Any company that wants to still be here in the next five years will have to integrate AI, in one form or another, whether into its processes and systems or into its product, if it wants to survive. The way AI is developing right now is, I would say, fairly alarming, and we should be careful. I would not go so far as to call it an apocalyptic scenario for the survival or death of certain companies or markets. But everything will be transformed, with near certainty. I would be cautious, vigilant, that is the right word, because it is almost certain AI will disrupt many industries, many processes and systems, the ways we do things. The risk of not integrating it is probably similar to the risk certain companies faced before the era of digitalization. They stayed local, undigitized, and certainly lost power. Except this risk may be a bit larger.
The comparison to digitalization is worth taking seriously, because we have already run this experiment once and we know how it ended. The businesses that refused to digitize did not die dramatically. They faded, slowly, losing ground to competitors who moved faster and reached further, until one day they were simply irrelevant. AI is the same shape of transition, only faster and wider. The companies that treat it as optional will not be wiped out overnight. They will just be quietly out-competed by companies that fold it into how they work, until the gap is too large to close. Being late to this is not a dramatic death. It is a slow slide into irrelevance, which is worse, because you never get the shock that would have made you act.
At the same time, be careful how you integrate it, because today's AI is not always specific, and it does not always give the results we want in certain circumstances. Take a mental health platform with a chatbot, where you talk to the AI and it gives you specific answers. We need to train that AI to give specific answers, but we still have to collaborate with researchers, psychologists, specialists, to make sure those answers are correct. And even so, for reasons no one today fully understands, the AI's answer will not always be specific. At some point that accuracy will rise, and it will rise, and then this part of the thinking will become obsolete. But until then, be careful how you integrate AI into your systems.
Here is where it gets interesting, because people judge AI's output by the wrong standard. They expect it to be perfect. It is not, yet. But the quality of the output should be judged not against perfection but against the error rate of the alternative. Take the mental health app again. If we manage to build a dedicated psychologist for every single person, even if the answers carry a margin of error of three, or five, or ten percent, consider that in the offline world, working with ten thousand psychologists for ten thousand clients, the human margin of error would probably be around ten percent too. So if the AI's error is at that level, or lower, we have a clear benefit, and on top of it we have gained almost infinite scale. In some situations it may be better to have three percent wrong answers if it lets us serve, individually, a thousand times more people than we otherwise could. Because in the world there are clearly not enough experts in certain fields, and a trained, specialized AI can supply those experts, which can prove genuinely beneficial to humanity.
The perfection standard is not just wrong, it is a way of hiding from a real trade-off. Nobody demands that human experts be perfect, because we know they are not, and we accept their error rate because the alternative to a flawed expert is no expert at all. AI deserves the same honest comparison. The question is never whether the AI is perfect. It is whether the AI, at its real error rate, serving a thousand times more people, produces more total good than a scarce, expensive, equally imperfect human alternative that most people cannot access at all. When there are not enough experts to go around, a slightly imperfect expert available to everyone can be a profound improvement over a perfect expert available to almost no one.
And it is worth noticing something almost philosophical here. AI wants to mimic the human brain, or rather its behavior, and it has errors in judgment, which we also have as humans. So it manages to mimic that behavior of the human brain. We should not hold it to a standard we do not hold ourselves to. Whether AI will eventually understand, or rather mimic, human emotions and create emotion through its output, I believe it will. I think we need to think a bit differently from how AI is trained right now, and I think we are wasting a lot of resources, because we found one method that works and stopped hazarding others. There are surely other methods we have not yet understood. Why not bio-organic AI? A synergy between technology and the organic. There could be many ideas leading to a different kind of AI, one that understands emotion. But right now the market invests massively in training AI on existing models, which leads to very high costs. At some point, though, a small team somewhere might discover another algorithm that trains AI far more cheaply and understands differently.
Now to the practical question every founder is asking: where should you use AI, and where should you not? I would be skeptical about integrating AI into strategic decision-making. I would be skeptical about integrating it into decisions concerning product improvement. I would use AI for an organizing role, to help us organize, to sort information, filter information, and bring it up to the point where decisions are made. It is important to understand that right now AI will find it hard to make decisions on its own, because you cannot digitize everything that happens in a company. Human relationships, the conversations between people, would all have to be recorded constantly, uploaded, and used to train the AI to make the best decisions. We cannot do that right now, and I am not sure we want to. AI often works from outdated information, and on that basis it can help you make decisions. I would take it up to the point of decision, but I would not let it make the decision, not now. And when I have all the information, or a summary of it, and I feel it helps me make an important decision, I would make that decision without caring whether the AI recommended it or not. Even if the AI says it recommends this, if my intuition, based on business experience, says otherwise, it might be better. We do not know for certain yet. The odds suggest it might be better.
I want to be precise about what a decision even is, because people misunderstand it. A decision in business and in life is not only which strategic direction I will take this year. A decision can be anything, from a new piece of information I take from a chatbot and rely on, and then base my own further decisions on, to something that heavily influences me. So exercise caution, because we often talk to these systems and they give us information, let us say it is correct and not hallucinated, and we rely on it, when we should actually do more investigation. This applies even where AI might seem critical, like certain trading platforms or places where scenarios are created mathematically, where if A plus B equals C and we know it is a hundred percent so, it is logical to let the AI make that decision. But in most other scenarios, be careful about letting AI decide.
This broad definition of a decision is the part most people miss, and it is where the real danger hides. You think you are keeping AI out of your decisions because you did not ask it to choose your strategy. But you asked it a factual question this morning, believed the answer, and built three subsequent choices on top of it. That is the AI making decisions for you, quietly, one accepted premise at a time. The influence does not arrive as a recommendation you can weigh and reject. It arrives as a fact you did not think to question, sitting underneath a chain of your own reasoning. Which is exactly why caution has to extend not just to the big obvious decisions but to the small inputs you accept without checking.
The medical field is the clearest test of this, and opinions are split. Some doctors say AI is excellent, it helps them diagnose faster, search through more cases and symptoms. Others say it is not ethical to rely on it to treat or interact with patients. I believe, in the end, doctors will rely on AI at nearly a hundred percent, and it is logical that they will. Think of the analogy with a digital calculator versus doing sums by hand. Sometimes you need the calculator, and in general you need to move faster, and AI can help with that. It may not seem ethical now, but given that it can save many lives, it will help enormously. The major problem right now is simply that AI is not consistent in its output. You can ask it the same thing and get three different answers, which can help you, or stop you, or slow you, or send you down a false path. That is why caution is needed, and why we should not give AI decision-making power.
And when I say decision-making power, I do not only mean the decision itself, I mean the power to influence the decision. It is like an expert you know who gives you all the information, and then you gather more from several places and make the call yourself. But if you give that expert absolute power over the decision, or power to influence it, it is exactly the analogy with mentors. Mentors are good, but if you give them absolute power over your decisions, or over influencing them, they may not lead you down the path you want, or the good path, or the path with the highest chances of success, because you are the one with the original vision. The same is true of AI. It is a phenomenal expert to consult and a dangerous one to obey.
So integrate AI, because not integrating it is the pre-digitalization risk all over again, only larger. Use it for its real strength: organizing, sorting, filtering, and scaling, taking a job that needed ten thousand experts and serving a thousand times more people at an error rate no worse than the human alternative. But hold the line at the decision. Take AI right up to the edge of the choice, and then make the choice yourself, with your own intuition and your own vision, uninfluenced by a system that will hand you three different answers to the same question. The mentor analogy is the one to carry with you, because it settles the whole question cleanly. A great mentor is someone you consult intensely and obey never, because the moment you hand a mentor the power to decide, or even the power to heavily influence the decision, you have outsourced the one thing that must stay yours, the original vision. AI deserves exactly that relationship. Consult it constantly, feed on its ability to gather and organize, take it right up to the edge of the choice. And then, at the edge, remember that it is a consultant and not the founder, that it will hand you three different answers to the same question, and that the vision the company is steering by lives in you, not in the model.
AI-first, yes. AI-in-charge, not yet, and maybe not ever, because you are the one with the original vision.