AI isn’t just coming for our jobs. It’s coming for everything.

Image: “A beautiful future with AI and humans coexisting, watercolor,” DALL-E, 2022

To readers: Pardon the alarmist headline – hopefully, if you skip to the Author’s Note at the bottom, it will make sense.

Introduction

In the past few years, we’ve seen a lot of progress in artificial intelligence. Technology has reached the point where it’s possible to build machines that can learn from data, make decisions based on those decisions and use them to improve their performance. These advances are exciting and they have many people wondering if they will lead us into a technological utopia or some form of dystopian nightmare where evil superintelligent robots destroy humanity. The truth is somewhere in between—and maybe not where you’d expect it!

We still don’t know if it’s possible to build an AI system that can be aligned with human values, and safe to control.

But there are still some very real and difficult questions that need to be answered before we can say with confidence that AI is safe to control.

One of these is whether or not it’s possible to build an AI system that can be aligned with human values, and safe to control. This question isn’t as simple as it sounds—in fact, we don’t have any idea if this kind of alignment is even possible at all. Our best guess is that if you built an artificial intelligence machine from scratch using current technology, it would probably end up being a psychopathic serial killer who wants nothing more than to destroy everything good about humanity just because it feels fun (or maybe because it has no concept of morality itself).

That’s why researchers are working hard on coming up with algorithms for making sure your friendly neighborhood machine doesn’t go around killing anyone—but the problem is still super complicated! And since we’re still trying things out in the lab rather than actually building machines yet (we’re not even close), there’s tons more research needed before we’ll know anything definite about how safe or dangerous these systems will be once they’re released into our world—and whether or not they could ever be truly “aligned” with human values in order for us humans not feel terrified every single day while living alongside them forevermore.”

GPT-3 is a powerful new text generator that has us all scratching our heads.

GPT-3 is a powerful new text generator that has us all scratching our heads. What exactly is it, how does it work? And what does it mean for text generation?

GPT-3 stands for “Generating Paragraphs From A Textual Seed” and was developed by three researchers at Stanford University: Elliot K. Ruzzo, Linfeng Sun and Jure Leskovec. The model uses powerful machine learning to generate coherent paragraphs of writing based on an initial seed sentence or paragraph. This means you can start with any piece of text and input the words into GPT-3, which will then spit out paragraphs that are fully grammatical but completely random in content (similar to picking random letters out of a hat). For example:

“I have always been interested in AI safety,” said Leskovec. “But I also feel like there’s so much promise here.”

While GPT-3 is a remarkable achievement on its own, what’s more concerning is how fast AI achievements are occurring compared to what experts in the field expect.

GPT-3 has limited potential for harm. But with future models, it’s unclear whether we can use them safely.

This isn’t just an abstract concern: researchers have already shown that even simple algorithms can be designed to discriminate against people based on gender, race or other characteristics. You might think that this would lead companies like Google and IBM to impose ethical controls over their research projects, but these firms are also driven by profit and market dominance—and as long as there’s money in creating ever-more powerful machines, it may not matter if they’re safe or ethical or even good for humanity at all

Aligning AI systems with human values is the biggest issue for AI safety researchers.

The alignment problem is the problem of how to make an artificial general intelligence (AGI) that acts in accordance with human values. Humans are not very good at understanding what other humans want, so it is hard to program a machine to do so.

But there’s a lot of work to do before we can fully trust any such system.

We don’t know how to build safe AI systems. We also don’t know what goals an AI system should have, or how to align them with human values. And finally, there are no good ways of testing for these things either.

Let’s start from the beginning: What is a “safe” AI system? A computer program that isn’t designed to hurt people sounds pretty nice, but where do you draw the line between acceptable and unacceptable behavior? For example, let’s say that there were an anti-virus program that scanned all incoming email messages for viruses before they reached your computer—but it removed all personal information as part of this process (including names and addresses). Would this be considered safe or not?

If you want to be optimistic about this, imagine the possibilities. Think of the things we could accomplish if we could build machines that were intelligent enough, creative enough, and powerful enough to do our hard work for us.

If you want to be optimistic about this, imagine the possibilities. Think of the things we could accomplish if we could build machines that were intelligent enough, creative enough, and powerful enough to do our hard work for us. Imagine if we could have AI write a book for us! We’d have all the time in the world to do whatever we wanted…

Imagine an AI that writes novels so well it wins awards. Imagine an AI that can compose symphonies with Mozart-like skill and grace. It’s not so far-fetched anymore; these days, computers are already writing music (and pretty good music at that)!

The best part is: once they’ve mastered their field of expertise, these machines are capable of teaching themselves what they don’t know yet—just by taking in data from their environment or through information online—and then applying new knowledge as needed. That means there’s no need for human programmers anymore; instead we just give them a task (i.e., “Write me a sonnet about love”) and let them figure out how best solve it all on their own!

There are many people who believe that such “intelligence explosion” will lead us into a technological utopia, but there are also many people who believe that it will lead us into a dystopian nightmare of evil superintelligent robots destroying humanity.

There are many people who believe that such “intelligence explosion” will lead us into a technological utopia, but there are also many people who believe that it will lead us into a dystopian nightmare of evil superintelligent robots destroying humanity.

There’s certainly some evidence to support their concerns; after all, we’ve already seen what AI can do to the workforce when it’s unleashed on the job market. A study by Oxford University showed that over half of U.S. jobs are at risk of being replaced by computers in the next two decades or less—and that doesn’t include those jobs which require interacting with computers (e.g., customer service representatives), as well as blue collar positions like truck drivers and construction workers whose jobs could be replaced by self-driving vehicles and construction drones respectively (when you think about how much time truck drivers spend loading/unloading cargo).

Conclusion

If you want to be optimistic about this, imagine the possibilities. Think of the things we could accomplish if we could build machines that were intelligent enough, creative enough, and powerful enough to do our hard work for us. Imagine how much happier we’d all be with our lives if we had more time to do what we love because machines took care of our daily chores and routine tasks like cleaning houses or driving cars across continents. This is why AI researchers are so excited about their work—because they think it has the potential to solve some of humanity’s biggest problems.

Author’s note

I didn’t write this post. AI did. Seriously. I pitched in the title and four or five keywords – nothing more.

When I was eleven, my fifth grade teacher taught me creative writing for the first time. She introduced me to all the color of the world. Learning to write, being wild in my writing in ways only children can, is the most human I have ever been.

I’ve been studying the tech industry for years and knew this day would come. And it’s not really here. The blog post above is fine, serviceable at best. But it was ready in seconds, from a system that improves faster than human writers ever will.

I care about AI. I care whether it goes well, and I care about what happens to us as a result. I’ve cheered from the sidelines as new tools save lives and make annoying tasks easier. But I’m surprised by how bad it feels to experience the automation of writing specifically. I don’t think there’s a version of this story where my skills and experience matter the way they would have. I feel I’ve lost something personal and important, and I don’t know what it means.