Just how big is this new generative AI? Think internet-level disruption
There's this odd feeling that starts at the back of the neck. It feels like the hairs are raising up slightly.
The first time I felt it was in the mid-70s. I was in high school. I was sitting in front of an ASR-33 teletype machine and I hit something, probably the RE-TURN key. That's how it was spelled. The keys were all very small and round, and RE- was on the top and TURN on the bottom of the key. I hit the key, and it said "READY."
No machine had ever responded to me interactively before. I was hooked. Although I originally planned to go into nuclear engineering as a career, the interactive nature of that very early computing sucked me into computer science. I was never the same.
Also: Watch a young Kevin Costner's very 80s Apple commercial
Fast forward two decades. The year was 1997 and I was playing the game Diablo (the first version of what's since become a classic). It was a dungeon crawler. I'd played many variations on the theme. I was somewhere deep inside a cavern getting beaten by the monsters. Suddenly, another fighter came over to me, and offered me healing potions and stronger weapons.
It was the very first time I'd ever played an online game where there was another person in there with me. Experiencing how real people could play together in an online world gave me The Feeling. Things were changing. Of course, the idea that someone could be nice in a virtual world is hard to believe in today's very cranky world, but it happened.
Then there was Amazon. Not the Amazon we have today, but the Amazon that could get you any book you wanted. I grew up cherishing book store experiences. If I was in a new city and there was a big bookstore, I had to go. That's because books were only available in libraries and certain bookstores.
But then, the early Amazon hit the internets. Suddenly, you could get almost any book you wanted. If the company didn't stock it, they would actually try to get it for you. Thing changed, of course. Now, we can routinely order just about anything from the cloud, even food, and have it in hours or days. But back in the early Amazon days, I got The Feeling.
Also: These experts are racing to protect AI from hackers. Time is running out
Another time I had The Feeling was fifteen years ago, with the iPhone app store. Suddenly, non-techies could install their own software. And apps could be taken everywhere. I quickly went out, bought myself a Mac, and wrote a bunch of very silly pinpoint iPhone apps. I could feel it. The tech world was changing.
I'm getting The Feeling again. As I've gotten older, I've gotten better at identifying the components of that feeling. Part of it is amazement. Here's something we couldn't do before. Wow. Part of it is annoyance. Here's something we should have been able to do, and it's just now possible. Why so long? And part of it is fear. This is going to cause disruption, break things, cost jobs, and possibly kill companies. Uh oh.
In each case, I felt a great disturbance in The Force.
Generative AI is giving me The Feeling
As many of you know, I've been digging pretty hard into ChatGPT and other generative AI tools.
See:
- I asked ChatGPT to write a short Star Trek episode. It actually succeeded
- 6 things ChatGPT can't do (and another 20 it refuses to do)
- l used ChatGPT to rewrite my text in the style of Shakespeare, C3PO, and Harry Potter
- I'm using ChatGPT to help me fix code faster, but at what cost?
- I asked ChatGPT to write a WordPress plugin I needed. It did it in less than 5 minutes
- Can Al detectors save us from ChatGPT? I tried 3 online tools to find out
- Can Notion Al writing helper write this article?
I've been involved with AI since I did a thesis on it back in college. But while AI has always been interesting, it's never sparked The Feeling. But now, I'm feeling The Feeling, and it's in overdrive.
Back in the boom days of the internet, investors were always asking if ideas were disruptive. Their belief was that if something disrupted the status quo, there was a chance for big money to be made.
To investors, disruptive investments were the holy grail. But they didn't take into account what disruptive meant. Disruptive meant destruction (or at least major damage) to existing industries.
Also: The best AI chatbots: ChatGPT and other interesting alternatives to try
Craigslist was disruptive. So much so that the biggest income source for newspapers, classifieds, shrunk to a shadow of its former glory. Nobody thinks the print newspaper business is a growth opportunity anymore.
Amazon was disruptive, in many ways. Its AWS cloud computing service helped removed servers from on-premises installations. Its huge distribution network helped kill off retail stores and shopping malls. And free delivery with Prime made it much harder for every other mail order and online seller to compete.
Netflix. One word: Blockbuster. 'Nuff said.
Uber. Airbnb. Google. Facebook. Spotify (and before it, Napster). All of these companies became behemoths and danced on the graves of older industries and the people they employed.
ChatGPT is disruptive. Brutally so. Terrifyingly so.
ChatGPT is going to replace a lot of knowledge workers
I've been experimenting a lot with ChatGPT. So far, I've asked it to write code for me, to change writing styles, to provide technical support, and even to write an episode of a popular TV show. In all these cases, it succeeded.
It definitely showed limits. It appears to break down on answers longer than 800-1500 words. It sometimes loses the thread of a discussion after repeated questioning. And it can get some things spectacularly wrong, as shown vividly in this actual response:
One word similar to "devolve" that begins with a B is "debase".
One fundamental limit to ChatGPT's capabilities is the hard stop on its knowledge. Its corpus ends in 2021. ChatGPT is oblivious to any information, trends, issues, or popular cultural themes that have occurred since 2021.
But even so, ChatGPT is good enough for a lot of tasks.
Publishers, the human purchasers of written content, in the main, don't have a high threshold for knowledge quality. They'll put up with less pithy headlines if they score more Google juice. They'll often put up with lower quality articles if they generate revenue quicker. They'll accept good enough if it helps them fill out their offerings in areas where they're weak.
Also: What is Microsoft's new Bing with ChatGPT? Here's everything we know
And why do publishers often accept "good enough"? Because human consumers of information often lack discernment. Readers will put up with patently ludicrous theories that support a cherished world view. They'll accept as fact made-up data because they have neither the time nor the inclination to cross-validate every assertion. Social proof has become the accepted and even preferred substitute for expertise.
To some degree, this is why political adherents on both sides have pushed back against ChatGPT. It's not about whether the AI engine can produce material that might displace jobs. No, it's that some of ChatGPT's responses don't parrot the themes and stances that those political adherents prefer as their doctrinal versions of social proof.
Also: How to trade in your old tech for Amazon gift cards
But even so, ChatGPT is good enough. It's good enough for writing short articles, flyers, stories, blog posts, and so much more. It does an adequate job. Unfortunately, the work product expected by employers and clients from most entry level and lower level knowledge workers also hovers somewhere around adequate. As such, ChatGPT may be able to replace those workers, or at the very least, consolidate a bunch of jobs into that of one worker who can write prompts and edit AI output.
ChatGPT is going to empower a lot of mediocre writers
Don't read that header as a criticism. It's actually powerful stuff. To put this into perspective, I need to tell you two personal stories.
Back in the early 1990s, I started my first company, a software publisher. Initially, it was self-funded, although Apple did eventually put in a few bucks to give us some runway. But back in the self-funded days, I needed a lot of original art -- for logos, packaging (software was sold in stores back then), marketing (we used printed flyers at trade shows), and so much more.
When I went through the process of finding and contracting with a graphic designer, I found the cost to be prohibitive. My entire marketing budget was taken up by the cost of designing one ad. So I decided I needed to become my own graphic designer. So began a 20 year quest to learn the tools and skills of graphic design.
Also: The best AI art generators: DALL-E 2 and other fun alternatives to try
Today, many of those tools -- and the skills -- are embodied in services like Adobe Express, Canva, Prezi, and Pixlr, which provide design templates, graphics, matching color schemes, and more. No longer does someone need to spend years learning the skills of a graphic designer. Instead, for a few bucks a month, they can produce good enough graphics on their own.
Look at how even those shelves are. I could never have done that myself.
Another example is the CNC. This is a device that cuts and carves wood based on digital plans. I am very, very good at making digital designs. I am very, very bad at cutting wood in a straight line. Recently, I wanted to make a cabinet to store materials. It needed 25 shelves, each an inch and a half from the next. That, in turn, required me to cut 50 exactly matching slots in two slabs of wood.
To say that was beyond my woodworking skill set is an understatement. But with the help of the CNC, I designed the slots. The CNC cut them perfectly.
Tools like Adobe Express and the CNC do enable experienced users to do more, more quickly. But they also open the doors to people who otherwise couldn't produce "good enough" results.
Review: 3018 Pro CNC: This tiny, under-$150 CNC is surprisingly fun and useful
ChatGPT is going to do the same thing. Perhaps it will help a mechanic who isn't particularly good at writing, but who needs some ad copy for their shop. Perhaps it can aid someone for whom English is a second language in generating a blog post or article. Maybe the AI can assist someone with a disability produce an otherwise unobtainable work product.
An AI like ChatGPT could be transformative to a lot of people's work output, the way the CNC was to my ability to create shelf slots that actually worked.
And yet, I worry
Disruption is not gentle. As much as the internet opened doors to new industries, it killed (or at least severely diminished) others. Power and money became more concentrated. While resources and information availability increased by an unprecedented degree, so too did misinformation, scams, and criminal activity.
One of the biggest problems with ChatGPT is that it presents completely wrong information as eloquently and confidently as it presents accurate information. Unless requested, it doesn't provide sources or cite where that information came from. Because it aggregates a tremendous amount of free-form information, it's often impossible to trace how it comes by its knowledge and assertions.
Artificial Intelligence
This makes it ripe for corruption and gaming. At some point, AI designers will need to open their systems to the broader internet. When they do, oh boy, it's going to be rough.
Today, there are entire industries dedicated to manipulating Google search results. I'm often required by clients to put my articles through software applications that weigh each word and phrase against how much Google oomph it produces, and then I'm asked to change what I write to appeal more to the Google algorithms.
Can you imagine what will happen when SEO geniuses, political operatives, and criminals start feeding information into the internet-as-corpus to manipulate what generative AI produces?
Remember, most readers do not discern or qualify the information they consume. They merely compare it to their own closely-held beliefs and worldview, and wolf down that confirmation bias like it was a plate of homemade cookies.
My concern is that generative AI can feed, nurture, and amplify two worrisome aspects of human nature: the desire for confirmation bias and social "truth," and the lazy willingness to accept "good enough" as good enough.
Mediocre quality, mixed with fake facts, and presented authoritatively and confidently has already produced chilling results on the world stage. It worries me that, with the power and convenience of generative AI, that level of corrupted and warped discourse will be amplified beyond all reason.
But hey, it's worth it. After all, I did manage to get ChatGPT to talk like a pirate and rewrite the preamble to the US Constitution. So, there's that.
Are you worried about generative AI? Have you tried it? Do you see chilling potential or empowering benefits? Let us know in the comments below.
You can follow my day-to-day project updates on social media. Be sure to follow me on Twitter at @DavidGewirtz, on Facebook at Facebook.com/DavidGewirtz, on Instagram at Instagram.com/DavidGewirtz, and on YouTube at YouTube.com/DavidGewirtzTV.