Where is AI going?

What I've been up to in 2024

I’ve been absent from Substack the past year. I’ll write more about my adventures later, but one of the biggest areas of focus has been using AI for everything in my professional, volunteer, and maker work in 2024. This post is a pontification of the future of ubiquitous AI.

A layperson’s intro to AI

AI is changing everything, but depending on what you do for work, you may not have much need to use it daily. So, for those folks, here’s a simple explanation.

The easiest way to think about the state of AI today is to imagine having a brilliant, well-read, thoughtful, and insightful colleague to talk to. This colleague isn’t perfect and makes mistakes sometimes, but they also possess a staggering amount of knowledge on a very broad range of topics. Because of this, you know when you ask them something—from a recipe for meatloaf to advice on starting a business to helping you build a website—you will get incredibly useful and detailed information to start you off in the right direction.

During this conversation with an AI, if you’re not convinced about something, e.g. “Are you sure I should add mustard in my meatloaf?” you can ask them, and they’ll recognize their mistake and correct themselves. You can talk back and forth about a problem, picking apart different aspects, and they’ll have a lengthy conversation with you.

It's difficult to appreciate how helpful AI can be until you use an AI that’s been trained in a specific problem space. I use AI for 1) editing written content, 2) assisting with my startup advisory work and coaching, and 3) writing software. While I won’t get into the nitty-gritty details of these three areas, AIs have saved me a tremendous amount of time, especially in software development, when I need to learn a bunch of new things very quickly.

A quick technical example

To illustrate the point, I was having trouble with one of my websites not loading when I was on certain internet service providers. The site loaded fine on AT&T and Verizon’s networks but wouldn’t load on Consolidated Communications or Spectrum. The error in the browser was obvious: the DNS lookup was failing. I have decent DNS knowledge, but I’m not an expert. In about 10 minutes of back and forth with Claude.ai, we ran some dig commands and identified a bad DNSSEC record left over from when I migrated the domain from one crappy registrar (who shall remain nameless) to AWS Route 53. I knew nothing about DNSSEC prior to this, and before AI, I would have spent several days googling around, finding and learning how to use dig, ultimately pondering the DNSSEC record, and then figuring out that that record wasn’t with the rest of the DNS records in Route 53’s Hosted Zones console tab.

This may sound like Greek to you, and it was Greek to me at first. But I spent a few minutes getting educated by Claude.ai and diagnosing the problem together, and we (well, really Claude.ai) solved the problem. And that’s the point. The speed of learning, problem-solving, and creating can be vastly accelerated by using AI once you learn to write good prompts, compare answers from different AIs, and do some classic googling to fill in the blanks.

Yes, there are downsides. AIs get confused, notably on deeply technical coding issues and especially the longer your conversation goes on. Claude.ai gets hopelessly lost in CSS when I iterate too long on a website design with it. But I believe these limitations with AI hallucinations are temporary while the models get better.

Some AI truths

Here are the four truths I believe about AI.

1. AI is a revolutionary technology, like the Printing Press, the Steam Engine, and the Transistor. Revolutionary technology disrupts people’s lives in both good and bad ways. Some jobs, companies, and industries will disappear because of AI. Other jobs, companies, and new industries will spring up. (Ask ChatGPT “What are some revolutionary technologies throughout history that fundamentally changed the world?” if you want to see an excellent list.)

2. AI can be a force for both good and evil. This is true of every revolutionary technology, and the outcomes are largely dependent on how humans use the technology. The difference between AI vs. other revolutionary technologies is that the impact of AI is being realized much faster than technological revolutions that came before it.

3. If you’re any kind of professional or knowledge worker, you must incorporate AI into your work if you want to remain competitive (or even employed). I believe this statement became true when ChatGPT-3 was released a few years ago, meaning if you’re not using it now, you better get moving.

4. While it looks like AI is in the hands of only a few companies, AI is actually a very democratic piece of technology. Deepseek built a ChatGPT-competitive model for a fraction of the training cost using open-source models, including Meta’s Llama open-source model.

What does the future hold

I like to talk about the future and the big changes that might happen.

And speaking of Aliens....

AI is particularly interesting because the future ranges from Utopia to Armageddon. I suspect people felt that way about the Printing Press in the mid-1400s, and people still feel that way about Nuclear Power/Weapons today. So maybe that’s the mark of a revolutionary technology: its ability to produce outcomes that span the extremes.

Here are my predictions.

Short term

• Job loss/migration. As I said above, use it or lose it. If you’re in the tech industry like me, you already know that AI is partially the reason the job market is so tough right now. But jobs are going to migrate more than disappear.

• Cheaper chips and training. DeepSeek is concerning because of potential censorship and security/privacy concerns, and it’s one of the reasons I won’t put anything sensitive in there. However, they have demonstrated that NVIDIA won’t be dominant forever, and models will get cheaper and cheaper to train.

• Lots of new startups. VC Investment is pouring into AI, and where the money goes, the companies spring up. Think about when SaaS first showed up on the scene. Do you even use shrink wrap software anymore, bro?

• Increased military, criminal, and pornography usage. Had to put this one in here because all new tech is used by (if not partially driven by) the military, criminal, and porn industries. The military has been experimenting with AI forever. My 1995 Masters Thesis on a rudimentary AI algorithm was funded by the Navy. Those increasingly realistic spam texts you get are AIs, and it’s going to get ever more difficult to tell who’s real and who isn’t. And porn…well see these SFW articles [1], [2], [3] for more predictions there.

Medium term

• Ubiquitous robots. Sometime in the not-so-distant future, robots like Marty, the Stop and Shop robot who looks for spilled groceries, and your Roomba will seem quaint. Robots are already in heavy use in manufacturing and warehouse logistics. Eventually, they will be an integral part of your household, like Serge from Battlestar Galactica, and they will be autonomous and very smart.

• Skilled professional AIs. Lawyers, Doctors, Engineers, and Customer Support of all flavors will have AI versions. Much law is boilerplate and can easily be handled by AI. Medical diagnoses that your primary care physician does can be automated by AI. If you combine the Da Vinci surgical robot and an AI, you will get an “autodoc” capable of performing routine surgeries. These jobs are already getting augmented and replaced by AI. We’re just at the beginning of this transformation.

• Rapid advancements in science and medicine. The bigger the artificial brain, the more it’s going to find things. AI is helping diagnose Breast Cancer earlier. We’re just seeing the tip of the iceberg on advancements.

• Haves and have nots. New technology isn’t immediately accessible to everyone. Democratizing AI and the products and services AI produces might be the singular biggest challenge as this technology becomes ubiquitous.

Long term

The Singularity. The singularity is defined as “a hypothetical future point in time at which technological growth becomes uncontrollable and irreversible, resulting in unforeseeable consequences for human civilization” [1]. This could be superintelligent AIs taking over, humans merging into machines, peaceful coexistence, or perhaps something we can’t even imagine yet. It sounds sci-fi, of course, but anyone who uses AI regularly and sees the power of the models is already thinking about this.

I think the only thing that would prevent us from reaching some kind of singularity is humans killing each other off first, which these days feels a lot more likely.

Don’t worry, though. Kurzweil is still predicting 2045 for the singularity [2], so you’ve still got 20 years to get your affairs in order.