Most People Think This Is A Smart Way To Learn AI, But It’s Actually Making You Dumber
By now, you’ve heard some version of the news:
“AI is the most important technology ever.”
“In a few years, AI will be smarter than the smartest person.”
“AI will create a utopia.”
“Oh, wait. Actually, AI might destroy humanity… Or at least take all the jobs.”
Intrigued, you tried out a chatbot like ChatGPT and got interesting results, but probably not life-changing.
Sure, there are things that a chatbot can do better, cheaper, and more reliably than any human (e.g., translation, transcription, data entry, large-scale knowledge retrieval, and summary).
Sure, there are things it can do well enough to disrupt or augment many jobs right now even if it isn’t superhamn. For example, low-level 24/7 customer service jobs, line editor, market research, brainstorming, coding, or basic content generation.
But when you use it, the effort required to learn and incorporate a chatbot into your workflow was barely worth the squeeze. At the very least, your experience may not live up to your hopes.
Part of you doesn’t want to get lost in the hype. Another part feels FOMO.
So you vacillate.
You vacillate between periods of ignoring AI…And periods of binging on the latest news, demos, tools, prompts, and hacks. Then, you get bored and start the cycle again.
If any of the above resonates with you, you’re not alone. It’s par for the course for all of us as we figure out this AI thing.
Collectively, I call this the AI Adoption Dilemma, and it’s critical to understand. Deciding how and when to adopt AI may be one of the most important decisions we make in our career.
That’s why I’ve spent hundreds of hours trying to “solve” it. This article provides the most comprehensive overview of what I’ve learned so far so that you can shortcut the process…
Inside The AI Adoption Dilemma
“A problem well-stated is a problem half-solved.”
Charles Kettering, Former Head Of Innovation, GM
In short, the AI adoption dilemma is this:
WHAT WE KNOW
“It is not what you don't know that gets you into trouble. It is what you know for certain that just ain't so.”
Mark Twain
AI is the fastest-growing technology in human history.
AI is the most general technology in human history.
It is not guaranteed to keep improving at this rate, but it likely will for the next few years.
If it keeps evolving, it will have an enormous impact on all of us.
There will likely be winner-take-most dynamics where a few companies/models/tools win the market. (why tech markets are winner-take-most)
AI capabilities are growing much faster than AI safety.
WHAT WE DON’T KNOW
“To know how to deal with what you don't know is more important than anything you know because the world is so much more surprising than you can really be sure of.”
Ray Dalio
Trajectory. We don’t know if AI will slow down (bottleneck/regulations) or speed up (AI is used to improve AI’s capabilities).
Winners. We don’t know which tools, business models, companies, or use cases will be the ultimate winners.
Limits. We don’t know what the limits of AI are or when we will reach them.
Intelligence. We know surprisingly little about the human brain. And we know even less about computerized general intelligence, which is significantly smarter than us.
Work. We don’t know what people will do for work in a world where cheap AI can do everything a human can do.
Impact. We don’t know all of the second-order effects of how AI will impact politics, war, international relations, work, mental health, parenting, law, education, etc.
Safety. We don’t know how to predict or safeguard society from AI’s side effects.
This situation is like a puzzle with the following qualities:
There are many ways things can play out.
The stakes are high.
Things are moving fast.
There are a few things we can confidently say.
But, there is even more things we can’t.
We have to act (even indecision is a form of action).
Bottom Line:
First, The AI Adoption Dilemma is more important than most people realize. Not only are the stakes high. But, the dilemma is part of a larger dilemma of how we make technology adoption decisions throughout our entire life in a world where technology is becoming more important. It’s AI today, but it might brain and body augmentation tomorrow.
Second, The AI Adoption Dilemma is challenging on a deeper level than most people appreciate. And, it’s critical to understand it at that deeper level in order to truly solve it. With that said, below are two key nuances that will help you understand The AI Adoption Dilemma at a deeper level:
The Time Machine Test shows that understanding how general purpose technologies (like AI) will be used is extremely hard
Succeeeding based on predictions of the future is surprisingly hard according to the world’s top investors
#1. The Time Machine Test Shows That Understanding How General Purpose Technologies (Like AI) Will Be Used Is Extremely Hard
Imagine it’s Friday, December 16, 1994.
You wake up early when it's still dark, tiredly shuffle to your front porch in your pajamas, grab the newspaper, and sit on your couch to take in yesterday’s news.
As you stumble through the large, unwieldy pages, you come across a short article in the back that catches your attention. A startup named Netscape, created by a recent grad named Marc Andreessen, just launched the first graphical web browser.
You don’t quite get it. So you decide to check it out yourself by visiting a site called Yahoo, which was started earlier that year as a directory of pages on the Internet.
Suddenly, it hits you.
Surfing the Internet used to be the purview of academics and techies who had to learn command-line interfaces and protocols like FTP, Telnet, and Gopher in order to use the Internet. Now, anyone can easily use it.
“The Internet is going to be huge,” you think to yourself. “One day, everyone will use it.”
So, you go all in. You quit your job and just focus on how to capitalize on the Internet opportunity.
Given that you’re so early, you’re bound to succeed beyond your wildest dreams. Right?
It’s just like the old Biff in Back To The Future II, giving his younger self the Sports Almanac, so he can bet on the outcomes of all of the big sporting events…
And become rich…
Well, that’s not quite how it works.
Knowing what the Internet would be used for was surprisingly hard at the time. My favorite case in point is the famous 1995 Bill Gates interview on The Letterman Show, where Letterman just doesn’t get it:
Below is my favorite part of the back-and-forth where Gates keeps explaining what the Internet can be used for, and Letterman keeps pushing back that the offline alternatives are good enough:
Dave Letterman: What the hell is that [the Internet] exactly?
Bill Gates: Well, it has become a place where people are publishing information so everybody can have their own home page, companies are there, the latest information. It's wild what's going on. You can send electronic mail to people. It is the big new thing.
[…]
Dave Letterman: But I can remember a couple of months ago. There was like a big breakthrough announcement that on the internet or on some computer deal they were going to broadcast a baseball game. You could listen to a baseball game on your computer. And I just thought to myself, Does radio ring a bell?
[AUDIENCE LAUGHTER]
You know what I mean?
Bill Gates: There's a difference. It’s not a huge difference.
Dave Letterman: What is the difference?
Bill Gates: You can listen to the baseball game whenever you want.
Dave Letterman: All right. I see. So it's stored in one of your memory deals.
Bill Gates: Exactly
Dave Letterman: Do tape recorders ring a bell?
[AUDIENCE LAUGHTER]
Knowing me the little you do, what am I missing?
Bill Gates: Well, if you want to learn about the latest cigars or auto racing statistics…
Dave Letterman: Well, you know. I've got that covered. I subscribe to two
British magazines devoted entirely to motorsports, and I call the Quaker State Speed Line about two times a half-hour.
[AUDIENCE LAUGHTER]
Bottom line: Even if you correctly predict that a certain general purpose technology will be huge and you decide to go all-in, identifying how it will evolve and impact the world is extremely hard.
Not only that, it’s very, very easy to be skeptical of new technologies. At first, they seem like a toy with limited utility. But, what’s critical to understand is that EVERY transformative technology seems like a toy at first. For more on this pattern, read famous investor Chris Dixon’s classic essay, The next big thing will start out looking like a toy.
#2. Succeeeding Based On Predictions Of The Future Is Surprisingly Hard According To The World’s Top Investors
Between 2016-2020, I became obsessed with studying the world’s top investors with the longest track record of success.
My logic was simple and ultimately right.
Investing is one of the most transferrable skills there is. Although, most people don’t invest their money for a living, all of us invest our time and make difficult decisions about career opportunities and risks based on how the future might unfold.
One of the people I’ve learned the most from is self-made billionaire investor Howard Marks. The video clip below captures his deep wisdom on the topic:
Original video: Finance Simplified, Notes: Sweat Your Assets, Date: 2019
In short, Marks says this:
The future is really hard to predict.
“One of the main reasons for the difficulty of making predictions is the enormous influence of randomness.”
“Nothing is more common than investors who are right for the wrong reason, and they get famous. But of course we shouldn't follow them.”
You can get into a lot of trouble by being too confident about the future.
“If you look at the great investors of the world, virtually none of them got famous by being able to predict the future.”
“Decisions do not have to be made on the basis of guesses about the future. They can be made based on an understanding of the present.”
Mark’s experience is not an outlier. It’s backed up by data.
The brutal reality is that only a few Internet companies started in the 1990s survived and thrived past the Dotcom bust of 2001:
Google (search engine) founded in 1998.
Amazon (e-commerce) founded in 1994.
eBay (e-commerce) founded in 1995.
PayPal (payments) founded in 1998.
Yahoo (media/search) founded in 1994.
Priceline (travel) founded in 1997.
Craigslist (discussion board) founded in 1995.
Netflix (TV & movies) founded in 1997.
Alibaba (e-commerce) founded in 1999.
These companies that succeeded as stand-alone companies got almost all of the financial benefit of money invested in the 1990s. Now, think about it. For each of these category winners, hundreds of other companies launched products but ultimately failed.
Not only that, a startling 2012 research study from the Kauffman Foundation found that the, “The [venture capital] industry hasn’t returned the cash invested since 1997.” This is a big deal because venture capitalists specialize in investing in the most promising tech companies. So, if they aren’t collectively killing it, it should give us all pause.
Bottom line: Even if you did know that the Internet was going to be huge in 1994, capitalizing on it would’ve been hard. Similarly, just knowing that AI will be huge and investing your time and money into it doesn't guarantee you will succeed.
After really thinking about the AI Adoption Dilemma, I think there is one clear, universal solution…
This Is The Best Solution To The AI Adoption Dilemma That I’m Aware Of
In the Howard Marks video above, Makrs concludes that what we can all do is understand the situation today, because the present is known. Marks says we can ask ourselves the following questions:
Where are we today?
What does that imply for the future?
What does that imply for how we should behave?
Within Marks comments lies the solution to The AI Adoption Dilemma.
No matter what, it’s worth devoting some time now to learn about AI and to think through its implications.
This very simple idea is powerful on three levels:
First, this enables you to make an informed adoption decision rather than a purely reflexive, all-in or all-out decision.
If it turns out that acting now is the best approach for you, you can then act now. However, if you reflexively avoid AI and then choose to go all-in later, you may regret that you can’t turn back the clock.
There are ways to learn about AI without falling prey to the noise of narratives and predictions that don’t pan out.
But learning could mean almost anything. For example…
What is the best way to learn?
How much time should we devote to learning?
What is even worth learning?
The rest of this article is devoted to answering these questions.
Answering these questions well is particularly important, because the default approach that most people take to learning online results in junk learning.
Navigating The Noise Of AI Media
“The information we consume matters just as much as the food we put in our body. It affects our thinking, our behavior, how we understand our place in the world. And how we understand others.”
Evan Williams, Co-Founder of Twitter and Medium
Over the last year, AI news has exploded.
Now, there is a whole industry of newsletters, articles, AI influencer tweets, and podcasts that repeat each other and share some variation on the following:
Startup Funding. New funding rounds, valuations, acquisitions, and investments in the space.
Corporate Announcements. Partnerships, products, strategies and acquisitions.
Demos. Jaw-dropping demos that pull back the curtain on the future.
Features. New features for existing products.
Predictions. Pundits predicting the future.
Tools & Prompts. Resources to help people use AI today.
Gossip. Rivals poaching talent. Conspiracy theories. Lawsuits. Etc.
The brutal reality is that within weeks and months, almost none of this news will matter:
Most new tools will not gain traction
Most predictions will turn out to be wrong
Most new companies will fail
New versions of AI will obsolete many of today’s prompts
Not only that, much of the news just isn’t practical now or is biased:
Demos will take forever to turn into mainstream products
A lot of news ultimately isn’t relevant to your day-to-day life
It is overly optimistic or overly pessimistic, so you can find data to affirm whatever you already believe
Bottom line: It’s easy to get into an AI junk learning habit where you consume the same newsletters, blogs, and podcasts every day, thinking you’re getting smarter when the opposite is true. It’s like eating McDonald’s every day while thinking you’re getting healthier.
Just as quickly as you pile information in, it becomes outdated. It’s like steering a leaky boat:
Even with coverage that has the best intentions, the future is fundamentally unknowable.
The noise of AI media begs a question:
How do we most effectively learn about rapidly changing fields like AI?
And for this, I also have one simple answer…
AI Mental Models Are The Antidote To Throwaway, Biased, Irrelevant AI News
“Developing the habit of mastering the multiple models which underlie reality is the best thing you can do.”
Charlie Munger, Warren Buffett’s Longtime Business Partner
Mental models are an abstract concept. To clarify what they are, below are three ways I like to think about them:
Toolbox for Thinking: A mental model is like a toolbox for your brain. Each tool represents a different way of thinking about a problem or situation. Just as you’d use a hammer for a nail and a screwdriver for a screw, you use different mental models to tackle different challenges effectively.
Lenses for Viewing the World: Think of mental models as different types of augmented reality glasses you can wear. Each pair lets you see the world in a unique way. Some might clarify your vision, others might highlight certain features, and some could even show hidden details that aren’t visible to the naked eye. By switching these glasses, you can get a more complete and clear picture of reality.
Cause-Effect Relationships: Mental models clarify the complex cause-effect relationships in any domain. This understanding helps you more accurately predict the consequences of your actions. Said differently, the better you understand the cause-effect relationships in a domain, the more likely you are to do better in that domain.
Everyone unconsciously collects models of the world. For example, if you closed your eyes right now, you could still navigate around your house, because you have a model in your head of where all of the rooms are in your house. Similarly, you have mental models of people, concepts, domains, industries, and technologies. Mental models of your friends help you interpret what they say and determine what you should say back. And mental models of technology determine how and when you should use it.
Top mental modelers do three things differently:
They make mental modeling a conscious process.
They actively refine and stress-test their models.
They proactively collect and use the top mental models.
The top conceptual mental models generally have the following properities:
Timeless. You can use them for the rest of your life.
Universal. They can be used in lots of domains and contexts.
Useful. They help you independently reason about facts, predictions, and news without blindly following the herds rushing toward or away from AI.
Collecting the top AI mental models help solve the AI Adoption Dilemma in two ways:
First, mental models contextualize news and give you ever-green takeaways
Rather than taking in information on its own, I quickly connect it to existing mental models in order to generate insights:
For example, there are a lot of predictions about the future of AI.
Because of the Wisdom of Crowds mental model, I immediately looked for multiple predictions rather than just trusting one. Because of the Expert Political Judgement mental model, I downvote the weight of “hedgehog” expert predictions, and I looked for diverse sources of predictions (e.g., academic researchers, scaling laws, and online forecasting platforms). Because of the dialectical thinking mental model, I looked for contradicting predictions in order to understand the underlying reasons for their predictions.
On the one hand, these models me to find the AI Scaling Law, which helped me understand the most bullish predictions for AGI timelines. The scaling law has existed for the last 7 years or so, its limits on hardware, data, and parameters haven’t been reached, and the progress is speeding up. Because of the Lindy Effect and building a model of why the scaling law works, I got more confidence that the scaling law will continue at least for a few more years and likely longer.
On the other hand, understanding contradicting predictions also led me to long-time AI researcher Melanie Mitchell’s academic paper, Why AI is Harder Than We Think, which introduced past AI predictions that proved false and explained why. This paper has helped me not become too bullish as to be black & white in my thinking.
Takeaway: Most people read news on the surface level and forget it. With mental models, you extract deeper, ever-green takeaways. With AI mental models, you build a better and better model for understanding the future of AI and how to prepare for it. As the model gets stress-tested, it gets better, and as it gets better, it tends to become more stable. Thus, I’m left with a useful tool I can reuse over and over to contextualize what’s happening in the environment.
Second, mental models reduce overwhelm and increase memory
The first rule is that you can't really know anything if you just remember isolated facts and try and bang 'em back. If the facts don't hang together on a latticework of theory, you don't have them in a usable form.
—Charlie Munger (Warren Buffett’s longtime business partner)
While there are a bazillion facts about AI, there are only a small number of mental models.
Said differently, mental models reduce overwhelm in the same way that closets reduce clutter.
They take a pile of clothes (eg, information):
And provide you with hangers and drawers to put everything in:
Elon Musk summarizes this mental model benefit perfectly in a Reddit AMA:
It is important to view knowledge as sort of a semantic tree — make sure you understand the fundamental principles, ie the trunk and big branches, before you get into the leaves/details or there is nothing for them to hang on to.
—Elon Musk
Takeaway: When all of your knowledge is connected by deeper mental models, you feel less overwhelmed. By remembering a few mental models, you can more easily remember more facts.
Given the power of mental models, I’ve spent the last four months:
Collecting the top mental models most related to AI
Using them to contextualize the news
Stress-testing and updating them based on research and writing about them
I’ve already shared several of these models in articles, and I will be sharing more in the months to come as well as a longer, categorized list of my top AI models:
AI Scaling Law. This model helps us understand how AI capabilities evolve as a function of data, parameters, and compute.
Paradigm Shifts. Things often don’t change gradually in one direction. They jump from one paradigm to the next. Understanding how change happens helps you navigate it and predict it.
Productivity Chain. Technological change reverberates through society in a predictable sequence. Understanding this helps you understand that new technologies don’t just automate work overnight. It takes months and years for companies to overhaul their workflows.
Hocky Stick Rule. When a technology is growing exponentially for a long time, take the time to explore it now.
Bottom line: In my humble opinion, learning AI mental models is the best way for the average knowledge worker to learn about AI given the stage it’s currently at. With mental models, AI adoption moves from a dilemma to an opportunity with immediate benefits. No more FOMO. No more hype. No more overwhelm.
Try The Mental Model Club: I speak about the power of mental models from direct experience. I first learned about them in 2015 when I read Poor Charlie’s Almanac by Charlie Munger, Warren Buffett’s longtime business partner. They completely transformed how I saw the world and helped me make much better decisions in every area of my life. So, I became addicted. In 2018, I co-founded the Mental Model Club with Eben Pagan. Since then, we’ve created 50+ 10,000 word manuals of the top 50 mental models along with 90-minute on-demand classes on each. If you want to access those resources for just $1 for the first month, check out the website.
Get Access To My Current, Categorized List Of AI Models (Paid Subscribers)
If you want to go deeper right now, I invite you to check out my mind map that contains the models I’ve collected and find repeatedly helpful. As I find more, I will update the chart.