Playing with AI

As a young software developer, I was intrigued by artificial intelligence research. I spent long hours reading the books of people like Pamela McCorduck, Douglas Hofstadter, Daniel Dennet, Joseph Weizenbaum and others, and for a while I thought I might want to find a way to work in this growing field. New applications for neural networks were appearing constantly, and rapid improvements in expert systems in the early 1980s made it feel like we were getting close to creating machines who think, to borrow the title of McCorduck's book.

But I got busy with other things, diving into computer graphics instead of artificial intelligence, and wound up ignoring the topic of artificial intelligence for the next few decades.

Then, around 2022-2023, just as I was retiring, AI was back in the news and bigger than ever, with the onslaught of LLMs (large language models): ChatGPT, Gemini, Claude, CoPilot, and their ilk. I've written before about how LLMs are merely pattern regurgitators, so I won't revisit that topic here. Today, I'd like to talk about AI and video instead.

I've always taken many photos of our dogs, and in the last couple of years I've started capturing videos of them as well. I quickly learned that this meant I needed to get proficient with some type of video editing software, and after exploring a few options I settled on Cyberlink's PowerDirector. It's a good fit for what I want to do and how I like to work, and the price is right ($5/month for the plan I use).

Like most software these days, PowerDirector has started adding various AI features. I had been ignoring those, but this morning I decided to take a peak at what they can do. I started by trying out the "text to video" AI feature, which prompts you to type a text description of a video, and then AI will generate a video that matches your description.

Remembering how our dogs Nancy and George spent five days and nights out in the Pioneer Mountains wilderness this spring, I wrote this prompt:

Two white Samoyed dogs are exploring a vast forest wilderness in the mountains, with some snow on the ground and areas without snow. The dogs chase a deer, then they are chased by a grizzly bear. The dogs fall asleep at night, huddled together to keep warm under a sky full of stars. The next day, they run with a pack of coyotes. The video ends with a high shot from behind of the two dogs walking down a dirt road, and the camera moves high above them to show that the road they're on is headed down out of the mountains.

I typed that in, clicked on Generate Video, and a few minutes later I received a notification that my video was ready. Here it is, with a title that I added:

Was this fun do? Sure. Would I do it again? Probably not. I wouldn’t want to spend my time or my money in this way.

I enjoy capturing the personalities and antics of our dogs, but I don't get any satisfaction or joy out of watching make-believe versions of the dogs doing make-believe things in a make-believe setting. Maybe I would have found that amusing when I was six years old, but those days are long gone.

Regarding spending money, that video cost me $33.60. When I clicked on Generate Video, my prompt was sent to an AI program running in a data center somewhere, and the delay of a few minutes was the time the AI program spent generating the video. Data centers are expensive, and you have to pay for the use of them.

PowerDirector, like many AI-connected programs these days, hides the exact cost of using AI features behind a "credits" system. You can buy credits, and then when you use an AI feature you burn up some of your credits. I bought a block of 500 credits for $35 (7 cents per credit), and creating this video burned 480 credits, which added up to $33.60.

Investors in AI are betting that millions of people will spend money regularly on these sorts of AI capabilities. I have no idea whether that's true or not, but I know I won't be one of those people. The thought of repeatedly buying credits and using them to generate little videos reminds me of people sitting in front of digital slot machines, inserting money to pay for the thrill of watching fake reels of fake fruit pretend to spin around. Whatever floats your boat, but neither is for me.

Thus concludes my foray into AI, for now. The next time I'm in the mood, I think I'll try out PowerDirector' image-to-video AI feature. Maybe I'll upload a photo of myself and ask it to make me dance like the Gurning Rave Guy. Stay tuned.

“Machines will be capable, within twenty years, of doing any work a man can do.”
– Herbert Simon, 1965

“Within a generation ... the problem of creating 'artificial intelligence' will substantially be solved.”
– Marvin Minsky, 1967

“If you define AGI as smarter than the smartest human, I think it's probably next year, within two years.”
– Elon Musk, 2024

“AI could surpass humans at almost everything within 2-3 years.”
– Dario Amodei, 2025

“AGI (artificial general intelligence) could arrive within 2-3 years.”
– Demis Hassabis, 2026