Unless you’ve been living in a secluded hut on the side of a remote mountain in Nepal, you’ve probably heard major chatter about AI in the last few months.
First, we had GPT-3 back in late 2022.
Then, we got AI-generated images with tools like Stable Diffusion.
Now, just yesterday, OpenAI released GPT-4, a multimodal model that can work with both image inputs and text inputs.
“We’ve created GPT-4, the latest milestone in OpenAI’s effort in scaling up deep learning. GPT-4 is a large multimodal model (accepting image and text inputs, emitting text outputs) that, while less capable than humans in many real-world scenarios, exhibits human-level performance on various professional and academic benchmarks.”
And, it’s very powerful.
Moreover, according to the press release, GPT-4 ranks well on a variety of academic benchmarks, like college entrance exams and AP tests.
GPT-4 is “intelligent” and capable in a way that we’ve never seen before from any AI system.
So although we’ll likely find plenty of problems with the current system …
And even though there’s still a lot more work to be done on these types of AI systems …
We’ve almost certainly entered a new era of human history. One that will change politics, economics, science, business, and most of our institutions.
Welcome to the Age of AI.
A New Era of History
I want to avoid playing into the hype too much, but, I suspect that we really are entering a new era: an era that’s marked by the use of AI and dominated by those who wield it.
The obvious analogy is the Industrial Revolution.
Before the Industrial Revolution, most physical goods were created by hand by skilled artisans wielding hand tools. Most cities were small and many people lived in small villages. Physical power was mostly generated by humans or domesticated animals.
And then, it all changed.
Over a few decades, innovations like the steam engine and the factory, along with advances in energy production and material science fundamentally changed the world.
According to one book on the Industrial Revolution, output per worker doubled within 80 years, from 1770 to 1850.
Standards of living increased.
People changed how they lived and worked (for example, urbanization).
And also, new problems emerged.
It was a fundamental shift that impacted work, economics, consumption, demographics, politics, and more.
The introduction of “intelligent” systems is likely to produce a shift similar in size and scope as the Industrial Revolution.
We’re in for some very big changes.
A Few Features of the Age of AI
Let’s talk about where things are likely to go from here.
Now, to be clear, the list of things that I’m about to discuss is limited. I’m personally interested in how AI will change the nature of work and how it will change the Tech industry.
And since this is a data science blog, I’m interested in how it will change data science work, in particular.
Additionally, most of the things I’m about to say are things that I have modest confidence in. Meaning: I think that they might be accurate and might come to pass, but there’s a probability that they might not. Things are changing fast, so I reserve the right to change my mind on these things as this technology matures.
How AI will Change Data Science and the Tech Industry
Here are a few predictions I’ll make about the impact of AI over the next few years.
- Machine learning eats software
- AI becomes a productivity enhancement tool
- AI becomes a “pair programmer”
- Small teams create massive wealth
Let’s quickly discuss each of these in a little more detail.
Machine learning eats software
A couple of years ago, I read a blog post by Andrej Karpathy, the former director of Artificial Intelligence at Tesla.
This blog post described what Karpathy called Software 2.0.
In his description, Software 1.0 is the traditional software that we know: explicit, hard-coded instructions given to a computer in programming languages such as C++, Java, and Python. Software 1.0 requires the programmer to tell the computer exactly what to do.
But Software 2.0 is something different. This type of software learns what to do. Software 2.0 learns how to perform well on a given task by being exposed to data. It is “Software that Learns from Data” or “Data Driven Software.”
Said differently, Software 2.0 is software that relies heavily on machine learning.
I think that this is where a lot of software will go in the future. Increasingly, software will involve more and more AI and machine learning techniques.
Part of this trend will be software that explicitly uses external AI services like GPT-4 via an API.
But in other cases, these new Software 2.0 systems will be custom-built machine learning systems that have been trained on proprietary (i.e., private) data sources.
Either way, “software” will increasingly use machine learning for better performance.
To quote Karpathy, and also echo Marc Andreesen …
… “Software is eating the world, but AI will eat software.”
AI becomes a Productivity Tool
More broadly, AI will be a major productivity tool.
At yesterday’s GPT-4 demo, OpenAI’s Co-Founder, Greg Brockman described their AI as follows:
“[GPT-4 is an] amplifying tool and together you can reach new heights.”
– Greg Brockman, OpenAI Co-Founder
Echoing this sentiment, I’ve heard it stated that “AI will not replace people, but people who use AI will replace those who don’t.”
Speaking broadly, AI tools will enable people to do more work, faster. And this will be true across a variety of disciplines.
Tools like Chat-GPT will enable writers to write more text and copy in a much smaller period of time. For example, copywriters will probably use tools like Chat-GPT to assist with copywriting tasks.
Non-fiction writers will use these tools for things like technical writing.
Even inside of a business, people will use text-AI systems to do things like write emails, reports, and memos. I already know of one friend of mine who has done this!
Beyond the large language model systems, tools like Stable Diffusion – which generate images – will be used by graphic designers and artists.
Overall, AI will enable people like designers, artists, and writers to do a large amount of work in a much smaller amount of time.
AI becomes a pair programmer
Concerning “AI as a productivity tool” there’s a special case related to programming and data science.
Going forward, software developers and data scientists will use AI to help write software.
We’re already seeing this with AI coding assistants such as Microsoft Copilot.
Additionally, more general tools like Chat-GPT can also write code. If you “chat” with Chat-GPT, and ask it to write some code to do a particular thing, then it will write the code for you.
In fact, about a week ago, I set up my Anaconda environment to connect to Chat-GPT. If you do this, you can interact with Chat-GPT and ask it to write code for you. You can also ask it more general questions right in your programming environment. For example, I asked it to explain some machine learning concepts. I’m going to write more about this in the future, but it was pretty useful.
Now, I would still recommend some caution around this. I think that it’s a mistake to rely entirely on AI code generating systems. You need to understand the code yourself. If you fail to understand it, you might build something that operates in unexpected ways, or does things that are otherwise bad.
But setting aside these concerns, AI coding assistants are very powerful right now, and they will become an important tool for any programmer or data scientist.
Small Teams Will Create Massive Value
Finally, one last point that connects to the previous points about AI as a productivity tool.
AI will enable small teams to create massive value.
Valuable products and valuable companies.
To explain this point, let’s quickly review what happened about 10 years ago with Instagram.
Back in 2012, Instagram was sold to Facebook for $1 billion.
Do you know how many people that Instagram had on its team?
13 people built a billion dollar company.
Before it happened, such an accomplishment would have seemed impossible to most people.
But through a combination of smart people and technology (which made those people more productive), the small Instagram team was able to create a very valuable company.
Now today, modern AI tools will make this type of story a lot more common.
Think about it: get a few software developers that use AI-assisted coding tools to large amounts of software, quickly.
Hire designers that use AI image generators to create images, designs, and mockups.
Hire marketers that use AI text generators to generate copy.
In the not-so-distant future, small teams are likely to be able to do 10x amounts of work, at very high quality.
It’s likely that within this decade, we’ll see another $1 billion dollar company created with fewer than 10 people.
There’s a Lot More to Come
I’m sure that I’ve only scratched the surface here, in terms of the impact of AI on tech, society, and everything.
And these systems are improving fast. Within 6-12 months, we’ll probably have new systems with even more capabilities.
But what seems clear to me is that AI will change almost every aspect of our society, from our work, to our politics and economic structure.
And with that in mind, I’ll be writing a lot more about this new Age of AI.
In particular, I’ll be writing more about how to build AI (i.e., more about machine learning) and how to leverage these new systems.
What do you think?
What do you think?
Are we in a fundamentally new era of history?
How else will AI impact our world generally, or data science specifically?
I want to hear your thoughts.
Leave your comments in the comments section below.