Let's Talk About Types of AI

“I’m sorry, Dave. I’m afraid I can’t do that.”

In 2023, we’ve seen an awful lot of change in terms of what the public thinks AI can do. Before the recent LLM (Large Language Model) revolution featuring superstar ChatGPT, most people working on AI predicted that AGI (Artificial General Intelligence) would take decades to arrive, if indeed it ever arrived at all.

Now, many—if not most—are placing it much, much sooner, possibly in this decade. Metaculus puts it almost exactly 10 years out, and you can see the trend line dropping dramatically a few times, as transformers were introduced, and then as ChatGPT rolled out:

Today, I thought it might be useful to talk about the different types of AI and what they mean. Language is a funny thing, and AI pontiffs and scientists are just like everyone else—they argue about terminology almost as though it’s a sport. I want to talk about the concepts underlying these definitions, and why they’re so useful.

This piece aims to unpack the four key points on this continuum, painting a picture of where we are now and where we might be headed. Remember, there’s a lot of necessary arbitrary classification here, but the main thing to pay attention to is the concepts.

Automation: Doing What Humans Do, Only Not As Well

If you've ever watched a Roomba bounce around your living room, you've witnessed automation in action. These machines aren't AI in the same sense as some of the more advanced examples we'll explore. Rather, they're an example of systems designed to perform tasks traditionally done by humans, albeit not quite as well.

From self-checkout kiosks at grocery stores to basic email spam filters, we live in a world saturated with automation. These systems, while not perfect, provide significant benefits. By taking over tasks that are mundane, repetitive, or time-consuming, they free us humans to devote our energies to more complex and meaningful pursuits.

So, while an automated system might not clean your living room as well as you could, its efforts still save you time and energy—benefits that should not be underestimated.

Now, some automated tasks might be done better than a human, and I don’t want to split hairs here. I’m just saying that there are loads of tasks out there that don’t take a tremendous amount of mental energy, aren’t stimulating in the least bit, and which we’re all too happy to hand over to machines.

As we climb the AI evolutionary ladder, we encounter systems that can do more than simply replicate human tasks. They can actually outperform us in specific areas, introducing a whole new level of usefulness and complexity.

Narrow AI: Excelling Where Humans Fall Short

Here's where things start to get interesting. Narrow AI, also known as Artificial Narrow Intelligence (ANI), is designed to excel in one specific area, performing tasks better than any human could. This is the domain of IBM's Deep Blue defeating world chess champion Garry Kasparov or Google's AlphaGo destroying Go world champion Lee Sedol.

Unlike automation, which typically takes on tasks humans find mundane or too time-consuming, narrow AI takes on tasks that require deep expertise or intense computation—tasks where human capability can be outmatched by machine precision and processing power.

Narrow AI can predict weather patterns, diagnose diseases from medical images, and optimize complex logistics problems. I mentioned spam filters with automation, but these have become more advanced over time, finding more and more creative ways to classify things as spam, and the “arms race” nature of spammers using better and better technology keeps spam filters at the forefront of ANI.

These tasks, which would be incredibly time-consuming and potentially error-prone for a human, are performed by narrow AI quickly and efficiently.

It's important to remember that while these AI systems excel in their specific areas, they are typically limited to one task.

Artificial General Intelligence: Mastering Any Task A Human Can Do

Unlike its narrow counterpart, AGI refers to a type of AI that can understand, learn, and apply its intelligence to any intellectual task that a human being can. It's the sort of AI that can plan, reason, and even comprehend emotions and social interactions.

It's crucial to note that, as of 2023, AGI remains a concept rather than a reality. The AI we interact with today, no matter how sophisticated, is still fundamentally narrow. Even ChatGPT, with its impressive language processing abilities, is not truly "understanding" the conversation it's engaged in or the content it's producing. It's operating based on patterns, not comprehension.

The arrival of AGI represents a monumental leap in technology, bringing with it opportunities and challenges that are hard to fully grasp. From potentially replacing jobs across a wide spectrum of industries to bringing about new innovations and solutions, the advent of AGI could transform society in ways we can barely imagine.

If you want to understand why the shift in sentiment has been so dramatic, and why all these AI researchers and programmers are now changing their tune so much about when AGI might roll out, I just wrote this the other day:

Yet, AGI doesn't represent the end point of AI development. There's one more stage beyond it, a stage where AI doesn't just match human abilities—it surpasses them.

Artificial Superintelligence: Beyond Human Comprehension

When we cross the threshold of AGI, we come to the final stage of AI evolution: Artificial Superintelligence (ASI). ASI refers to an AI that doesn't just match human capabilities—it vastly surpasses them in practically every relevant aspect, from general wisdom and creativity to problem-solving skills.

If AGI is the point at which an AI becomes as smart as a human, ASI is the point at which it becomes smarter than all of humanity combined. It would be capable of making discoveries that humans can't even comprehend, solving problems that seem insurmountable to us, and predicting outcomes in ways that we find utterly unfathomable.

The idea of ASI is both exciting and terrifying. On one hand, it could solve some of the world's most pressing problems, from climate change to incurable diseases. On the other, it raises serious questions about control and ethical considerations. Would an ASI be benevolent, malevolent, or something entirely beyond our understanding?

How would society change in the face of an intelligence so profound?

We simply have no idea.

As of 2023, ASI is purely speculative. While we've made incredible strides in AI development, we're still far from creating a system that exhibits AGI, let alone ASI. Yet, given the speed at which the field is advancing, it's worth contemplating the implications of such an unprecedented leap in intelligence.

Think about the whole spectrum of possibilities and what they could mean for our world. It's about reshaping the world as we know it, challenging our ideas about intelligence, and defining a new future.

From an AI that's no smarter than a Roomba to one that could outstrip human intelligence by orders of magnitude, we're entering an era of innovation that could redefine every aspect of our lives. The landscape of AI is shifting, and with each new breakthrough, we're inching closer to turning science fiction into reality.

AI is already changing the world as we know it. And as it continues to evolve, so will the potential for uncharted, transformative changes.

That’s why you’re here, and that’s why this place exists. We are at a serious inflection point for the human species, and we need more smart people to take a serious look at what’s happening out there. Help me do this by sharing this piece today: