AI Hands-On

As a follow up to my AI article, I decided to play with the chat.openai site a little bit. The first attempt told me that the servers were overloaded and I should go get a coffee. First mistake. I don’t drink coffee. If you’re going to rule the world with people’s private information, you have to get the basic facts right.

It came back with some clever sonnet thing. The words show up as if they are being typed by someone, with an old school flashing box cursor and all. A little odd maybe for something that’s supposed to be the future of everything.

Remember the Twitter Fail Whale? When Twitter was new-ish, the servers got overloaded all the time, and you’d just see the big blue Fail Whale. Anyway. Progress, huh? Not that I do twitter anymore. They actually kicked me out for automatically tweeting links to my blog. I guess I was considered a bot. Anyway, fail whale.

They offer to notify you by email when the service is available again. Again going old school. We’ve heard for years about the demise of email, but it refuses to die. Just like 2D drawings.

Every time I refreshed the screen to see if it was available, it gave me a new sonnet, or long excuse for why it wasn’t yet my turn to play with it. Different every time. At least this is better than the “Your call is (not) important to us, please remain on the line while we serve other customers.”

So I did something else and came back and it was up and working again.

The site is just an input box where you type a question or a command. It has certain limits, which I guess is a good thing. It looks like this AI has been trained with information pre-2021, so there is certain stuff that it’s not going to understand at all. I wanted to get a feel for what it could do, and what the whole thing is about. So I started with a simple task, to see if I can just use it to write my blog for me. A pretty generic request for something I would never actually write.

---

Write a blog post on the topic of the best cad software.

---

Well, obviously it didn’t capitalize SOLIDWORKS in marketing case, that’s sure to annoy someone. It’s a pretty generic answer to a pretty generic question. I could watch it type out the answer as it went along. Notice it doesn’t really answer the question directly. It kind of beats around the bush, giving several CAD tools that exist, without any sort of value judgment. Just facts, actually. Interesting. But it’s only interesting once.

Next I asked it to do something a bit more specific: “Write a detailed explanation of the use of bodies in Solidworks.”

---

---

It’s interesting that it can understand something so specific without knowing a lot about it. Some of this sounds like I wrote it, because maybe I did. It probably read my book, and is spouting phrases of it back at me. But when I wrote about bodies, I wrote a 30 page chapter, and I did it after years of actual experience with the software and the function. The response is still very generic, and not detailed in any way. But it sounds familiar. I mean, you see a lot of writing these days that is this insipid, this devoid of actual actionable information.

One concern I’m getting from this is that AI will read any book you write, and then supposedly have all of the knowledge in that book available. This would lead to people not writing books any more. There would be no detailed information created. AI wouldn’t kill off books because it could write them, but because there wouldn’t be any incentive to write them any more, since the AI would just give away your book to people who didn’t read it. They found a way to just give away music, maybe some day they will think of writers as human beings worthy of making a living too.

Or maybe AI counts on the inherent willingness of people who know to tell. In the same way that publishers have made it nearly pointless to write a book (because they take all the profit out of it and count on people to just essentially give away content), AI will just take all the content that people are so eager to throw out there and give it away. If Google has not been able to drive people with technical knowledge into bankruptcy, AI might be able to do that. Even today a lot of technical books are just full of senseless drivel.

Also, inevitably when I do a web search on how to do something in Facebook, my computer’s OS or in a browser, the step-by-step is inevitably wrong. Usually right from the beginning, the steps don’t match the interface I see. It’s probably a version thing. This doesn’t bode well for AI learning from the web.

So how do we make sure your knowledge is ever worth anything going forward? Back to the Matrix where we have to disconnect? I’m sure these questions are going to work themselves out, but I’m also sure they aren’t going to work themselves out in my favor.

---

Next I thought I’d throw it something that I was sure it couldn’t do, like “Design a 15 story hospital”.

At first it started throwing out text, but it started with some very generic platitudes, and then it started telling me how to design a hospital in very generic terms, it wasn’t designing a hospital.

Eventually I got this:

Maybe it took it a while to realize what I already knew. That it couldn’t design a hospital, and even if it could tell me how to do it, it still wasn’t an answer to the question.

There’s a lot of optimism about AI. The thing I see about it is that we’re going to have to be more careful than ever about trusting stuff we read. Whether it’s stuff off of Wikipedia, stuff on a blog or forum, or generic blather generated by AI, there is no guarantee anything is correct.

What happens when it does figure out how to finally design a 15 story hospital? Is it working from a template or several small templates, putting multiple ideas together? Is there an upside to letting a machine think for us? Does it really free us to do other more important things? What’s more important than thinking and creating and expressing? We’re left with human relations, but we will be bereft of thoughts and experiences.

I think it needs a short leash. And I don’t know who I would trust to wield this. Will the Google model work? Google simply strips every connected device of information, and if it can be stripped, it belongs to Google. This part of Google is certainly evil for people who create stuff. If you create images or words, Google owns your life’s work, and never pays you a cent.

After asking it some more questions, it’s easy to see a pattern to the answers. It’s a piece of programming with a lot of data. As with anything, GIGO (garbage in garbage out). If this were trained with incorrect data, it would give you incorrect answers. If you train it with biased data, it will be biased. It can take on a personality of sorts depending on how it was programmed. Do you trust a bunch of hoodie-wearing tech geeks who live in their grandmother’s basement to establish what kind of information you get fed? No. No more than I trust google to give relevant search results. It’s just like in real life, you talk to one friend and don’t get the answer you want, so you talk to another friend.

The thing that’s difficult here is people have to remember not to just automatically trust technology. We fall into this trap sometimes with CAD – sometimes we design things that absolutely cannot work or exist or be made.

So who is AI a threat to? Right now, no one. It might put some Russian trolls out of business, or proliferate their style of writing. But real writers who have opinions or draw conclusions or do something beyond strictly stating facts have nothing to worry about. AI right now can’t really solve problems, it just regurgitates bromides and cliches, and has whatever biases it was (consciously or unconsciously) programmed to have.

Just as an experiment, I input a statement on a politically sensitive topic which I won’t repeat here for obvious reasons. It started by giving me just the facts on the topic. But then it diverged into a paragraph on something that was definitely not factual, and exhibited a definite point of view. I proceeded to point out that it was taking sides and exhibiting an opinion, and it replied with the familiar “as a language model I do not have opinions”.

I pressed the point, continuing to argue with the computer, and eventually it admitted that it is programmed by humans and “biases or opinions may be present”.

Just to say that if it sounds intelligent, and speaks like a person, it’s not. You still have to judge what it says for yourself, and realize it’s just like going into Reddit and asking a bunch of random strangers for an opinion. You’re just as likely to get stupidity as wisdom. Remember, it has been trained to think using the internet. Kind of like your 13 year old.

Next I asked for personal information of myself. Reading what it wrote seems so common sense, but it is definitely at odds with the way all companies treat our personal data. AI is supposed to compete at some level with google. I think google could use some competition, even if what they are competing for is kind of unethical to begin with. “Providing personal information could also be illegal.” Really. Even AI can say this with a straight face while the entire electronic corporate world is falling over themselves to collect, steal, purchase if necessary personal information that they have no business having and we mostly don’t want them to have anyway.

So what’s my conclusion? What opinion did this non-AI writer come to about AI? AI is as fallible as the people who programmed it. It is very likely to be used to push agendas, and while it claims that it’s just about the facts, it clearly veers into opinion and you can’t just assume because it’s a computer it has an objective point of view. It’s not likely to take my job any time soon, but as with other market forces such as publishers and google, stuff that you write or images that you create don’t have much value because if it’s electronically available, it instantly belongs to everyone/no one but certainly not you.