Welcome to the 1st edition of AI Office Hours, your go-to monthly segment for everything conversational AI. First of all, I was really happy to see so many questions finding its way to my inbox. Good questions, too, great questions, some the best questions I’ve ever seen.
And, if you’re thinking “Wait, what!? Why did I not know about this?”, don’t worry. I’ll be doing an AI Office Hours like this every month, so feel free to email me or leave a comment below.
Before we dive in, I thought it’d be fun to start with AI-written joke. What’s a monthly segment without its own ritual? So, here it comes, by none other than ChatGPT:
Divine Negotiations
Death shows up late to his weekly chess game with God, looking unusually cheerful. God raises an eyebrow and asks, “What’s got you in such a good mood?”
Death grins. “I just thought of a new rule for humans: Every time someone says ‘I'd die for a coffee right now,’ I take them up on their offer.”
God chuckles, setting up the chessboard. “That’s quite the policy change. How about for every time someone says ‘This is heaven!’ I send them back to you for a reality check?”
Death laughs, moving his pawn. “Deal, but I get to send them back if they complain about the coffee in heaven.”
Okay, enough with the tomfoolery. Let’s answer some hard-hitting questions.
Question 1:
Full explainability — simply too revealing...? Is it in the interests of big tech to reveal what goes on in the black box or are we’re all being hoaxed with “Anonymous Indians”?
By:
Starting off, I think it’s worth explaining the phrase Anonymous Indians. For those who aren’t familiar, this tweet might shed some light:
If you’re still clueless, here’s a recent news story for you. Last week, it was reported that Amazon’s ‘Just Walk Out’-technology, touted for entirely relying on computer vision, actually relied almost entirely on Indian human workers doing the work. As many as 1,000 workers in India manually reviewed what the customers picked up and walked out with from Amazon’s Just Walk Out-enabled stores.
From there it’s a small leap to the anonymous Kenyans that were hired by OpenAI’s to train ChatGPT to be less toxic. Crowd workers, data labelers, Facebook-moderators, Foxconn’s factory workers — the reality is that big tech, and now AI, relies on cheap labor and hidden jobs that make the technology work and it can’t do without.
This all ties in nicely to the original question, “is it in the interests of big tech to reveal what goes on in the black box?”. I’m afraid the answer is no. Transparency about how these systems are built, what they’re trained on and whether that data was licensed data or not; all of it is being kept secret under the guise of “competitive advantage”. In reality some of it just can’t bear the light of day — and they’re well aware.
The New York Times reported on the 6th of April that OpenAI, desperate for training data, transcribing over a million hours of YouTube videos to train GPT-4. Not only was OpenAI president Greg Brockman personally involved in collecting videos that were used, the company also knew it to be legally questionable and proceeded anyway. Need I say more?
Question 2:
As a graphic designer, I encounter a lot of AI in the commercial world. Many artists are currently in revolt because their work is being used/scanned to create AI art. There are even galleries that already prohibit AI art. Will there be a way to recognize this kind of work in the future? And are there laws or regulations for creative expression, for example around watermarking AI-generated images?
By:
Last year, I wrote a short piece about the German photographer Boris Eldagsen, who was declared the winner of the prestigious Sony World Photography Awards. Soon after, he revealed that the prize-winning image was created using AI and proceeded to not accept the award.
With his action, Boris Eldagsen tried to address a pressing issue: the fact that we can’t tell the difference anymore between what’s real and what’s fake. We may think we can, until you realize that, by definition, you can only be aware of the all times you are not fooled by AI.
It’s an impossible conundrum, especially considering techniques like watermarking AI-generated images will never be 100% reliable. There will always be tools that allow you to either remove or create images without such markings.
The real problem we’re facing is that the progress in AI has been so incredible fast, on so many levels, that society is kind of forced to play catch up. Artists are standing up against companies like Stability, Midjourney, and OpenAI for profiting off their IP, as evidenced by the lawsuits that are building up — and I think rightfully so.
Is there a place for AI in the art world? Sure. But I don’t think AI is ever going to replace art. A comparison I like to make is chess. When IBM’s Deep Blue beat chess champion Gary Kasparov in 1997, some believed that chess as a sport was doomed. Funny enough, today, more people play and watch chess than ever before in the history of mankind. And we’re not watching computers play chess against computers, no no. We’re watching people play.
That’s because people like people. So even though some of the craftsmanship that’s part of being an artist or a graphic designer may disappear as a result of AI, the audience will always want to see the human behind the work. That’s just the way we are.
Final question:
Why isn't anyone talking about ideogram.ai? 🤯
By:
Honestly, I hadn’t heard about ideogram.ai either. A quick Google search tells me Ideogram is a Toronto-based startup. It raised $80 million in its Series A funding. They’ve flown a bit under the radar with DALL-E, Midjourney, and Stable Diffusion taking most of the spotlight.
Ideogram apparently stands out because of its ability to render coherent text in images, a well-known weakness of image generators. This is what I got when I asked for a Pixar-style movie poster promoting this newsletter:
Pretty cool, huh?
Before you go…
As always, thank you for reading Teaching computers how to talk. Support for this newsletter is now possible via donations. Any contribution, however small, will help me allocate more time to the newsletter and put out the best articles I can.
For now, I can only offer fixed amounts, so if you’re looking to make a odd-shaped or sized donation, shoot me a message and we’ll work something out.
PS. You can pause or cancel donations anytime.
PPS. If you do end up donating, the universe will grant you eternal happiness until the end of time. No joke.
Join the conversation 🗣
Leave comment with your thoughts or like this article if it resonated.
Get in touch 📥
Shoot me an email at jurgen@cdisglobal.com.
Many thanks for reply to my Q Jurgen. I guess we won't be seeing full explainability anytime soon. A shame really, as it may have shed light on how it is that Claude et al, Claude in particular, have acquired an ability to bewitch some eminent and extremely intelligent commentators.🤖