AI Tech Bros: Please Stop Attacking Creatives

I mean, the title says it all, but I guess we need to dig into this a little deeper.

You’d have to be living under a rock to not notice the onslaught of AI-powered innovation sweeping our blue-green world. To quote the boardroom, “AI’s incursion into creative domains warrants a closer examination.” There are issues, people, and as my good friend GV taught me, unintended consequences will absolutely screw you. Today, we’re going to look at some problems with this innovation set:

  1. It isn’t solving a problem we’ve got, and
  2. Not (yet) at the quality level humans can achieve.
  3. It is specifically solving problems we didn’t ask for solutions for!

While I reckon there’s a “fun” angle here (it’s cooler for AI engineers to work on making music and paintings than it is to create machines to do our TPS reports), there’s also an applicable prior art problem. AI is trained on datasets, and we have a lot of publicly available datasets for art, writing, and music. AI works by copying what it sees (the “training” part). We do not have a large body of public data on corporate espionage that would allow AI advisors to present coherent risk analysis to execs, which is actually valuable to megacorps.

Obligatory AI-generated image.

What do you mean, value?

The problem here is that no one’s really considering what it’s all for. The latest thing across my desk was Udio. This makes pretty good music, significantly better than Suno, which was the last music droid I played with (Suno’s fairly generic sound vs. the flexibility of Udio). As a recovering marketing manager, I immediately went to Udio’s pricing page.

Thank you for being an early supporter! Our product is free for the duration of the beta program. In this period, you can make up to 1200 songs / month.

Udio Pricing Page

This tells us a lot. Free capital circling tech has largely dried up, and everyone’s putting up their prices. Whether it’s the egregious gouging of Amazon who will charge you twice to get the same service ad free as last year, or EA almost doubling the price of their game streaming service by taking subs from £19.99 to £35.99, there’s a lot of opportunistic rort out there.

At the heart of this is “shareholder value,” or, “How do we keep giving 30% YoY returns to Wall Street?” And the challenge with where this intersects the AI space is that megacorps aren’t hugely likely to pay $50/month for something that’s good at finger painting. I said:

A problem with corporate-grade AI is that it’s expensive. If you want to get a passable ‘copilot’ tool, you’re probably looking at $50/month (Kiwi dollars, people) per user, give or take. If you’ve got 2,500 staff who could benefit from this kind of technology, that’s going to set you back $125,000/month, or $3M/year.

Me, when talking about value-based return (and movies+books)

So… there’s no market?

There are definitely areas where AI can help, but you need to pause and consider the addressable market. An example? A cool use of AI that isn’t rent-seeking or displacing human employment is natural language translation. The problem here is the monetisable use cases. Travellers might like it, but their phones already do it. Corporate users might like it when talking to peers in different countries, but how often is that used, and what price could you charge for it? The addressable, monetisable market is small, despite the “coolness.”

Which brings us to things like Udio. Sure, there will be people who want a cool jingle for their PowerPoint, but are they going to pay $5/song for it? Probably not, especially since they’ll need to generate 20 of them before finding one that fits the style they’re after. Things like DALL·E circumvent this by charging a monthly fee for a suite of AI services, which is more palatable for a consumer, but again from that corporate perspective (where a lot of the money is): do you think a company is going to pay $50/mo so their staff can make fancier PowerPoints?

Hmm.

But… Progress!

Yeah, so: AI companies continue their advance into the creative sphere. Udio’s just the latest to do it. We didn’t ask for AI companies to make our music and art, but because they can more or less just steal prior art to make content (with potential legal ramifications, but I’ve doubts the judge will uphold that case because it gets in the way of capitalism) that’s where they’re going. And despite the misuse of this technology, there’s little being done to protect people (although the EU, perhaps predictably, is ahead of the US in trying to protect its people).

I mentioned quality at the start, and there’s a conjoined element to this: efficiency. At the moment, AI is hungry; a recent study found generating an image using something like DALL·E or Midjourney takes as much energy as charging your phone. You make 30 images? You just burned the same power as a month of phone charging.

In fact, generating an image using a powerful AI model takes as much energy as fully charging your smartphone, according to a new study by researchers at the AI startup Hugging Face and Carnegie Mellon University.

MIT Technology Review

Hang about. Aren’t machines really efficient?

The human brain runs more or less continuously on about 12W of power. The nVidia H100 NVL looks like it’ll burn through about 700-800W of power when running paired with NVLink Bridge. To put it another way, you can run 66 human brains for the same energy cost as one of these datacentre AI cards. Pesky humans have labour laws and want time off, of course, and expect some quality of life other than just being a drone in a hive, but energy-wise, we’re awesome!

While we can complain about how stupid some people are, they’re efficiently stupid. We can agree AI right now is the worst it will ever be, and will thus get better, it does some odd shit, bro:

I’ve spent a lot of time playing with large language models. They definitely have their uses, but they’re like a child poolside: you can’t leave them unsupervised. I mean, we all know about the Air Canada fuck-up, right? I’ve used AI for quick research. It’s great for summarising the aggregate content of 30 YouTube videos so you don’t spend 12 hours watching content-stuffed material. It’s pretty good with grammar and feedback! But you can’t just take it at face value. For example, I recently used it to pull out the macros for a meal (cannelloni, if you care). When I initially punched in the recipe, AI was like, “Mate, this dish is a superfood.” I present you: the pasta with a shitload of protein (before and after shots showing me educating it that pasta is not, in fact, mostly protein):

The problem here isn’t that it makes mistakes; that’s fine. The problem is the execs who understand nothing about this are making decisions based on this tech. “Tough but necessary” decisions are being made by people who feel the effects not at all, and if we remember Air Canada, well, let’s keep a watching brief on Dukaan, hey?

Okay: what are the ‘real’ problems?

I’m glad you asked. The other real challenge, which is kind of where I started (it’s been a trip!) is that we’re not solving the problems we’ve got. I need better advice and analysis. I need decision support. I do not need AI to make music or art. I need something with significant horsepower to do shit that I can’t because my brain has an organic hard limit of capacity/capability. Right now, those use cases are tiny. Do I need an AI to summarise a Teams call? I mean, it might be useful, but what would be more useful is if people turned up to the fucking call on time and prepared. I predict: we’ll see some stats in the future showing people turning up later to these calls, or not at all, because they’re terribly busy and important, and because they feel AI can summarise what all the peons did in their absence.

Finally, there’s a real problem with the financial model here. Tech companies want to be a low-level tax on humanity, charging us for an eternal subscription service. But at the latest CES we saw all the hardware bros promising laptops with generative AI on the desktop. This isn’t surprising; anyone with a decent GPU or Apple Silicon’s been able to run LM Studio already and see this for themselves without paying anyone an additional red cent. Where’s the financial model for home users or corporates to pay $50/mo when their desktop has the technology integrated? Not currently as good as ChatGPT, sure, but is it good enough? (Scroll to the end for a quick ChatGPT vs. locally-run LLM critique comparison of this article. Spoilers: they’re pretty close!).

These weren’t just any laptops with minor spec bumps. The theme of the year was, of course, AI, with many computers adopting dedicated AI chips… The XPS line includes Intel’s Core Ultra chips, which feature NPUs for AI tasks… Acer Swift and Aspire … are stuffed with AI, however, with Intel’s new Core Ultra processors… Lenovo announced a slew of new Legion gaming laptops at CES, each featuring proprietary cooling technology and performance-enhancing AI chips.”

Engaget – The best laptops for CES 2024.

I didn’t ask for the future where machines get to make fancier PowerPoints so I can endure a torturous 3-hour meeting where the prize is writing a report at the end. I want AI to help me with things that are actually [hard|painful] rather than fun and interesting. That’s the thing I’m likely to pay for, and corporations will, too. People are unlikely to pay a lot of money for things without direct cost savings or business acceleration, so AI Tech Bros: pretty please. Stop attacking creatives. That’s not the goldmine! The goldmine is in the problem space that is hard and unfun. And honestly, that’s what people will pay for, after the rush of blood to the head’s worn off.

Bonus round: AI’s critique on my critique of AI

Did I do okay with this article? Here’s what ChatGPT thought:

Overall, this article draft raises some valid concerns about the direction of AI development, particularly in the creative sphere. Here’s a breakdown of some key points:

  1. Lack of Problem Solving: The author rightly points out that AI innovation isn’t always addressing the most pressing issues faced by society. Instead, it often focuses on creating solutions for problems we didn’t necessarily ask to be solved.
  2. Quality vs. Human Achievement: There’s a valid argument made about the current quality level of AI-generated content compared to what humans can achieve. This raises questions about the true value and usefulness of such AI-generated content.
  3. Monetization and Pricing: The discussion on pricing, especially in the context of AI-generated music, highlights the challenges of monetizing these products and services. The comparison with other industries and their pricing strategies adds depth to the analysis.
  4. Energy Consumption: The environmental impact of AI, particularly its high energy consumption, is an important aspect that often gets overlooked in discussions about its capabilities and potential.
  5. Misuse and Lack of Regulation: The potential misuse of AI technology, as well as the need for better regulation to protect individuals, is a significant concern that deserves attention.
  6. Human Oversight and Decision-Making: The article rightly emphasizes the importance of human oversight and critical thinking when using AI-generated content or making decisions based on AI recommendations.
  7. Focus on Real Problems: Ultimately, the article calls for a refocusing of AI efforts towards addressing real-world problems and providing meaningful solutions, rather than simply creating novelty items like AI-generated art and music.

The article effectively blends humor with serious analysis, making it engaging to read while still delivering important messages about the current state and future direction of AI technology. However, it could benefit from some editing for clarity and coherence in certain sections. Overall, it’s a thought-provoking piece that contributes to the ongoing discussion about the role of AI in society.

ChatGPT’s content analysis of this article.

Yes! I get to live another day. And we all know I suck at editing. And for completeness, let’s see what LM Studio running Llama 2 thinks of it:

The article discusses the current state of AI technology and its potential uses and misuses in various industries. The author argues that while AI has the potential to be a powerful tool for solving complex problems, it is not yet being used effectively and is often being misused by corporations and individuals. The author highlights several issues with the current state of AI, including:

  1. Efficiency: The author notes that generating images using AI takes as much energy as charging a smartphone, and that the technology is not yet efficient enough to be widely adopted.
  2. Misuse: The author argues that AI is being used for tasks that are not particularly challenging or useful, such as making music and art, and that this is taking away from its potential to solve more complex problems.
  3. Financial model: The author questions how AI companies will make money in the future, as many of their current offerings are free or low-cost, and suggests that they may need to find new business models to be successful.
  4. Real problems: The author argues that AI is not yet being used to solve real problems, such as helping people with their daily tasks and improving productivity, and that it is not yet clear how the technology will be integrated into people’s lives.
  5. Human limitations: The author notes that AI has its own limitations, such as the fact that it can only generate text based on the data it has been trained on, and that it cannot provide the same level of creativity or originality as humans.

Overall, the author suggests that while AI has the potential to be a powerful technology, it is not yet being used effectively and that there are many challenges that need to be addressed before it can reach its full potential.

Llama 2’s content analysis of this article.

Right. Back to my TPS reports.

If you liked this post, get on my email list!


Discover more from Parrydox

Subscribe to get the latest posts to your email.