If AI Is So Smart, Why Isn't the World Changing Faster?

So we live in an era of superintelligence, of AGI, where silicon-fabricated intelligence can drive cars, can solve novel math problems, and are better than humans at just about everything, at least in terms of intelligence. So why doesn't it feel that way? Why does superintelligence feel so slow, and why isn't the world changing physically? Our daily experience, right, in going about our work, our personal lives, our cities, when compared to this incredible rise of these hyperintelligent machines. Well, there is a reason why, and I think it's important to understand, because it frames how AI will evolve and permeate through society, not just now, but over the next 5, 10, 20, 50-plus years. That is going to be the video. Let's hear a word from the sponsor of today's video, and then we'll get into it. Now, if your week is back-to-back meetings and calls and random conversations where important things get said, you know the feeling where you get to Friday, or Monday, whatever day, right, and you're piecing together what you actually agreed to, and your to-dos, and the commitment, and these kind of half-finished notes in the Notes app or Google Docs, or just scored in your mind, and if that's you, then this is the MemoKit Gem. It's an AI wearable that captures your conversations all day, and it turns them into summaries, tasks, and connected notes automatically. Now, it weighs just 0.4 ounces, so once you put it on, you can get it to there, and you can wear it with a co-wear band that snaps it right next to your Apple Watch, and there's no second thing to remember to put on, or you can wear it as a wristband, a pendant, or pin it to your shirt, and the battery runs up to 20 hours of continued recording, and for most people, that's about a full day. And here's the thing, though, right? Most AI reporters summarize one meeting at a time, while MemoKit connects context across every conversation over days, weeks, and months, and you can ask the AI questions across all of it, so a full week of meetings becomes one clean report, and you don't have to type any of it. And nothing here gets stuck in another app, so your action items are sent to Google Calendar and Apple Reminders, and your notes automatically push into Notion and Slack, so everything stays in the tools you already use. Now, the gem is $1.99, and with 10% off, it will be $1.79. My code will get 10% off, and the code works on Amazon or directly at memokit.ai, and if you order on Amazon, you get Amazon shipping, so link and code for that are in the description. Go grab the discount, and back to the video. So the reason that superintelligence feels slow is the same reason that the weakest link in a chain determines its strength, and in a production line, the speed of the slowest component determines the output, right, the aggregate rate of that line. Basically, the way in which superintelligence affects you is going to be bottlenecked not by the intelligence itself, but by a host of other things, which impedes how fast it And what that's saying, right, it's not about the AI necessarily as much as the environment these AI intelligent machines live in, which is a world built and designed for and by humans, and what that means is they will be gated by the same measures that we as humans use to regulate ourselves. Now, a very basic example of that is, let's say you want to deploy a new app inside of a company. Now, maybe it used to take two months to code the app, right, and now it takes a day. Now you can vibe code it completely with plot or with codecs. So that two months just got compressed to a day, but inside a large company, that two months is part of a year-long process to actually deploy it, because after the initial build, you have to go through revisions, you have to go through approvals, legal has to approve it, it has to go up three chains of managers to decide whether strategically it's a good idea, then marketing has to make their marketing plan, all these different things have to happen. So maybe you took that two months and you compressed it down into two days of vibe coding, whatever the number is. But if managerially things are still, say, reviewed and approved on a quarterly basis, that doesn't actually impact the speed of approval. While super intelligence means that you can very easily code this app, the net effect is not any different, right? No change to that system or the volume of output. And that's not actually how it is, right? Things are getting done faster. You are getting a productivity lift across the board, or at least in some areas, but that concept applies and that increments out these productivity gains instead of making them these kind of step function increases where you wake up and everything's kind of radically shifted overnight. And I think primarily there are three gating mechanisms or bottlenecks that most kind of slow down the permeation of these systems. I already mentioned the first one, and that is human processes for approving things. And that ties somewhat into just hierarchies and how organizations are structured and individual incentives for, say, moving up in a career over time. But for whatever set of reasons, typically you have a structure in the vast majority of businesses, or at least once you get big enough, where there's things that need to be run by certain people to get approved, right? That's certain teams, it's certain managers. And so a lot of this work that AI does has at some point to still run through these incumbent approval managerial processes for review, revision, and then release. And then that kind of has parallels into, I think, bottleneck number two, which is the whole legal and liability side of these things. And that if you go and want to publish something, get it out to the world in a company, one of those boxes to check is going to be legal. It's going to be the liability, the risk, the insurance, that whole domain of things. And that's part of why those approval processes exist, to make sure you can get rid of some of the risk or just not get into a situation in the first place where you're exposed to a lot of risk by function of a faulty product or a product that works, but just has kind of error rates, right? It works 99% of the time, but 1% of the time, and maybe the engineers building the thing know this, it's just not going to work. And well, how dangerous is that one? And well, how dangerous is that 1%? And so a lot of that bundle of things, it's why it's hard to just rip out human decision making and approval of risk processes and stick AI in there because AI is not an entity that can be blamed or there can be repercussions toward. And so that just means the liability falls on the company. And if there's a lot of liability and AI is doing all the work or kind of fully flowing through automating, managing a system, well, that's a challenge. And so that's why you still have an enormous amount of human oversight in a lot of this implementation of AI and why people really, really, really take their time to implement things at the corporate level. And yeah, sure. You have startups that are AI native and that's great, right? But 90 plus percent of the economy are going to be these incumbent, these larger businesses that are the real way in which AI actually goes and affects a majority of workers of technology and of the systems that kind of already govern the world we live in. And the number three, and finally, it's just the physical world, right? In the physical world, things take time to build, to deploy. And that means regardless of the speed of AI in the digital world, you still have things that take a lot of time in the physical world. That's moving materials, it's permitting, it's environmental studies, putting together the labor, the capital required to build things, right? The most obvious case study there in the case of AI is energy, right? It is building more power to power data centers which go power the AI and building a plant that can produce power of whatever kind just takes a long time. It's a big deal, takes a lot of money, a lot of labor, a lot of paperwork, politics, the whole nine yards there. And so maybe AI will eventually be an unlock there with robotics and things like that and automation. But for now, it's that same deal of you could speed up some parts of that process by 90, 99, 100%. But even if a permit document submittal used to take two weeks for a human and is now done in five minutes by AI, it might still take four months to get that thing reviewed and then another year to build the thing. And like so, AI can speed up these parts of the process which are functionally meaningless as far as effective actual output or speed goes. And that's not a hard rule, but it does explain a lot of the ways in which AI have developed over time, the speed of progress currently and what's going to happen over the next 10, 15 years. So yes, there are going to be counterexamples to this. And if you have a one-person startup, that's different in a lot of ways versus more established entities. But the net concept applies, which is even if AI compresses certain parts of a process and the time, the effort involved in those sub-processes by 90%, 99%, the net effect can still be zero or it can be marginal. It can be 3% lift this year, 5% lift next year, 7% the year after. And that is why we live in this singularity of AGI and super intelligence that we keep moving the goalposts on. And yet, Rome wasn't built in a day, right? The world is not changing that much. And in a lot of ways, maybe that's a good thing, right? Maybe that is what paces the frontier and gives time for people in the world to adjust and is really good for things like earnings and for the economy. But that's a subject for another video. Either way, I hope you found this one useful or just food for thought. Appreciate you being here and I will see you in the next one. Cheers.