The History of AI Explained: Crash Course Futures of AI #1

Computer Science
96 followers
14 Video Views·Jan 1, 2026
In 1965, Moore’s Law predicted how computers would become smaller, faster, and more powerful than ever before. But here in 2025, we’re on the brink of an even bigger revolution. In this episode, we explore where AI's been, what it can do, and where it might be going. AI benchmarks and scaling laws help us understand what AI is, and could be, capable of.
For half a century, computing power grew exponentially. In the 1960s, most computers
were the size of a refrigerator and they took a long time to do pretty much anything.
But by the 2010s, hundreds of millions of people were walking around with personal computers
right in their pockets and near-infinite information available at the swipe of a fingertip.
The change was unbelievably fast. It was like nothing history had ever seen.
And yet, it's possible that it won't even hold a candle to what's happening now.
Artificial intelligence.
Hi, I'm Kushin Avidar and this is Crash Course Futures of AI.
Back in the 60s, that exponential growth in computing power seemed virtually impossible
to most people. But one guy saw it coming. An engineer named Gordon Moore. In 1965,
he predicted that every two years, we'd find a way to get twice as many transistors onto
a computer chip as the previous one, making future computers capable of things Moore and
his colleagues couldn't even imagine. And he turned out to be totally right,
and his prediction came to be known as Moore's Law. And it held true for just about 50 years.
Today, progress has shifted from transistors and chips to the new frontier of artificial
intelligence. Artificial intelligence, or AI, is really just a type of computer system,
only it's a really advanced computer system that tries to emulate intelligent behavior.
AI is kind of an umbrella term that includes some pretty far-out-there stuff that doesn't
even exist yet, but also much more basic programs like the facial recognition software that unlocks
your smartphone. As long as you don't grow out your beard, or shave your head, or ditch your
glasses thanks to the wonders of LASIK. Ask me how I know. AI is sometimes defined by its ability
to do stuff kind of like what humans can do, like learning from errors, or making predictions based
on prior knowledge, or drawing hands. But it can also do stuff humans can't, like remembering every
single thing it's ever heard, processing millions of points of data in seconds, and making hyper
realistic fake videos of Tom Cruise covering Dave Matthews Band. But this capability is a
pretty new thing, and it's taken decades to get to this point. Back in the 1950s, computers couldn't
even help you with your homework or play a game of chess. At least, not until 1957. That is the year
IBM created the Bernstein chess program. The Bernstein chess program was an example of narrow
AI. It's a kind of artificial intelligence that only does one thing, in this case play a very
good game of chess. Basically, it used algorithms that mimicked human strategies, evaluating the
board, picking out plausible moves, and simulating them to figure out which one would work the best.
Eventually, it got good enough to beat an inexperienced human opponent. But it would have
to be a pretty patient human opponent. The thing took like eight minutes to make a single move.
Although, when I used to play chess with my dad, sometimes we'd take like 20 minutes to make a
single move, and that makes the Bernstein chess program seem like a certifiable speed racer.
We were thinking like really hard. In the decades to come though, computers got faster and the chess
bots got better and better. But they were still no match for good human players. In the 70s, a
Russian program named Kaiza became the world champion of computer chess engines, but it still
lost to human chess master David Levy. In the 80s, Grandmaster Garry Kasparov, the highest ranked chess
player in the world, played against 32 bots at once and beat them all. Then came the 90s and
Deep Blue. Building on their success with their earlier chess bot, IBM created Deep Blue. It could
evaluate 200 million chess positions per second. And at that rate, Deep Blue could burn through
all those different simulations to pick the best possible one. And in 1997, Deep Blue beat Kasparov
in a six game match. It was the first time a machine had beaten a human chess world champion.
But Deep Blue, like all those other early chess bots, was still an example of symbolic AI. It's
a kind of AI that relies on the logic coded directly into its hardware and makes moves by
hard-coded search and evaluation. Meaning it was never actually learning or improving. It was just
following the same human instructions it was programmed with. And while that can be good,
it wasn't great. But what if there was a system that operated a little less like an instruction
manual and a little more like a human brain? In 2020, a chess bot called Stockfish got something
new, an Efficiently Updateable Neural Network, or NNUE, and it totally changed the game. Well,
the game was still chess, but you know what I mean. Neural networks are a special kind
of computer architecture that works kind of like the human brain, processing information
with layers of connected nodes that pass information between them along weighted
connection pathways. Unlike symbolic AI, bots that run on any type of neural networks can
change based on experience. Like, if during practice a chess move leads to a checkmate in
three, that pathway will stay and the AI will probably repeat that pattern in the future.
But if it costs the network its queen, that AI is probably going to revise its strategy.
That process is called deep learning, and it depends on three things. One, the amount of
data or information available for AIs to learn from. Two, an algorithm to instruct its learning
process. And three, what your friend who's like really into AI calls compute, which includes
processing power, memory, and storage. And because Stockfish is an efficiently updatable neural
network, it's got even more space for that deep learning than a regular neural network.
While other AIs would have to re-evaluate the whole chess board every single move, Stockfish
can update incrementally, keeping up with every move it plays. And that frees up a lot of
processing space and it lets Stockfish think even harder about its next move, which turns out to be
really helpful. Since 2020, it's won basically every computer chess championship there is.
Although it hasn't played me yet. And not to brag, but I have an account on chess.com.
Since the deep learning revolution of the 2010s, neural networks have been the foundation for
basically all our AI up until right now, as I'm filming this episode. And that includes the rise
of general purpose AI. Unlike narrow AI, these systems can take on lots of different tasks to
achieve lots of different goals, like writing books, generating images, summarizing articles,
recognizing faces, driving cars, and making Will Smith eat spaghetti.
General purpose AI isn't only able to engage in deep learning, it's able to do so really,
really fast. That's thanks to a tool called a transformer, which lets certain kinds of AI
process entire sequences of data at once, rather than reading word by word or number by number.
That means general purpose AI is getting really good at this stuff really quickly.
So quickly that some people are starting to wonder, how can we keep up?
Enter AI benchmarks and scaling laws. In AI, benchmarks are standardized tests
that let us measure how well different AI models perform. For narrow AI systems,
benchmarks can be pretty straightforward. Like chess engines have a lot of the same
benchmarks as human chess players, getting certain fee day ratings or beating other players or
engines. But for more generalized AI, benchmarks have to evaluate performance across lots of
different domains, like large language models, which are neural networks that use transformers
to process language, are supposed to be able to do lots of different things. So they might be
tested on whether they can answer academic questions and accurately summarize things
they've read and write a 10 minute romantic comedy in the style of when Harry met Sally.
AI agents like self-driving cars might have to prove they can drive safely in the rain and not
hit a pedestrian on the street. When AI models can consistently perform well on a benchmark,
it's considered saturated. But these days, benchmark saturation is happening kind of
freaky fast. In a world where AI can identify faces, compose music, drive cars, and take PhD
level exams, the question is less what can AI do and more what can't it do? And it's something we
want to figure out before we're staring down the barrel of the AI apocalypse. The question is how?
Thankfully, the things we already know about AI, like past benchmarks, can help with that.
If you look at how different AI systems perform on different benchmarks, you can see certain
patterns, and one leaps out right away. In general, AI neural networks that are fed more
data perform better. The more books a large language model reads, the better it'll be at
writing its own. The more videos of Tom Cruise a video generator watches, the more lifelike
it's fake Tom Cruise will look. He's beautiful. Particularly if that's paired with more compute
and other scaling increases. If we squint, we can actually put some numbers on those patterns and
come up with formulas that describe exactly how an AI's performance seems to change with different
amounts of data, computer, both. We call those formulas scaling laws. And so far, they help us
predict what future AI with even more data, even more compute might be able to do. Full disclosure,
the formulas themselves kind of look like nonsense for everyone, but the most saturated AI nerd.
But they all mean the same thing. Bigger is better. In the world of neural networks and deep
learning, being able to find patterns and learn from huge quantities of data is gonna work better
than waiting for a human expert to code them in. And the more compute these machines have,
the faster and the more powerful they'll be. Now, of course, there are caveats. China's deep
seek is able to compete with models like chat GPT, despite having a fraction of the compute.
Plus, throwing all that compute at pre-training isn't showing the returns it once was. But still,
scaling laws explain why some of the highest performing generalized AI models belong to huge
companies like Microsoft and Google, since they've got access to tons of data and tons of processing
power. And following that pattern into the future, scaling laws can help us predict how capable AI
might become, gaining the power to transform our economy, or hand us over to authoritarian
dictators, or make all our likenesses sing covers of the Dave Matthews Band.
Crash course into me. Yeah, baby.
Are you feeling those tingles of excitement yet? Imagine how AI feels.
Just kidding. It can't feel. Probably.
Just like Moore's law foretold unimaginably small and fast computers,
scaling laws see a big future for AI. As algorithms improve and data and compute
keep increasing, AI could deep learn its way to some pretty advanced stuff,
like finding new treatments for cancer or achieving super intelligence.
Moore's law held up for a good 50 years, bringing us into this new era of computing
before slowing down. And there's no guarantee that scaling laws will last even that long,
thanks to potential bottlenecks down the road. But any day now,
AI could turn some new corner and scaling laws might not be enough to tell if it's still
harmlessly marching pawns around or coming for us all.
Next time we'll look closer at what AI could really do in our society,
other than lock us out of our phones if and when emo bangs come back in style.
Crash Course Futures of AI was produced in partnership with the Future of Life Institute.
This episode was filmed at our studio in Indianapolis, Indiana and was made with the
help of all these nice people. If you want to help keep Crash Course free for everyone forever,
you can join our community on Patreon.
