TWIG 768: Google A/I

Beep boop - this is a robot. A new show has been posted to TWiT…

What are your thoughts about today’s show? We’d love to hear from you!

The question to ask is why humans have to be in the loop at all ? Google shows artists using the tools, like musicians and filmmakers, yet Suno.ai and every other AI can generate entire complete compositions of songs, which in about six months will be indistinguishable from human produced music. What then is the point of having human input ?

The fact that AI generated content can have the same emotional impact as human created content but generate it at a scale that no human artist is actually capable of, means that human crafted music (even if they somehow can maintain a job doing it like people who make music for video games ) then they’d just get outcompeted in the marketplace. People’s response to this is to try to rub the back of this problem and say that of course humans will always prefer content made by other humans, but the evidence is in that’s simply not the case because AI. If someone hears a song on spotify or sees a photo/video that gives them an emotional response then it doesn’t matter who originated that content.

As for incorporating search with AI, it’s enshitiffication one prompt at a time. Now some, like Jeff seem to believe is that the AI is going to find cogent quality websites and serve them to you. But given that the AI’s can also generate the content of said websites, it means those are indexed and referenced by AI as readily as the human created ones. This leads to the same problem where (like the example of HouseFresh and online reviews ) the AI generated content spreads at a higher velocity whether it’s high quality or not is irrelevant, because the jobs get lost either way. A boss in a company is looking at these tools not as a way to enhance the productivity of the workers but to replace them.

These creative jobs are require a decade to become proficient in. So think about a child, lets say they have an inkling to draw, their parents buy them a computer, the child makes a few squiggles and the computer translates those squiggles into a photoreal unicorn. The incentive to improve is lessened. I have first hand experience, and if you ask many artists they’ll pretty much say the same thing, part of the reason they kept trying to improve their artistic skills was because when they drew something it didn’t look as good as what they saw in media/comic books, this creates a motivating feedback look which motivates you to continue learning. The Ai doesn’t nurture creativity it says instead, step aside, let me do it for you. It’s like building a machine for your gym that before you lift the heavyweight, the machine steps in and does it for you.

Anyway, my point is that for the first time in history the computer can do the creative works for us, it doesn’t need a human to be in the loop (as Suno.ai has shown ) nor will it need so in the future. In fact the base case for this technology is not to have humans in the mix at all but instead to do the task for you based on a simple prompt. This is going to accelerate, not decelerate, job loss across multiple sectors, starting with artists who are the most vulnerable to it right now. (As it needs to be said again, AI has already replaced voice actors and concept designers in video games ).

As anyone who is young and trying to break into the industry and you’ll see they’re a lot less excited for this than exuberant tech journalists who report on the Sam Altman’s of the world whose goal is to hoover up any future value that creative can generate in the future. It’s all really depressing and bleak right now for anyone wanting to start their creative career.

Because art is about the human condition. It is the artists experiences, their loves, their tradgedies, the wonderful things they have seen and the terrible things they have seen and experienced. That is why blues music is mainly about tradgedy and “blues” music that is light hearted doesn’t seem to gel.

You are listening to music or looking at a picture to feel those experiences for yourself. Advertising might be different, for example, but “real” music is an expression of the human condition. As the AI hasn’t experienced those things, it can’t make music with that same sort of feeling, the emotions are missing, or at best, synthetic. The harmonies might work and it is good for a laugh or a “look how good AI can produce a tune,” but it isn’t the reason we listen to music.

Likewise, art (drawings, paintings etc.) is about what the artist saw, how they saw it, what emotions it released in them. Hero images for a blog are one thing, but you won’t go to an art gallery to experience the AI.

And this is why AI content will fail, in the long run. We don’t want millions of songs about nothing, we want heartfelt ballards about something special to the producer. I have 10s of thousands of songs in my collection, but I probably listen to a few hundred songs on a regular basis, maybe a couple of thousand get listened to more than a couple of times. The masses of AI created content will be like those bad b-sides, they are there are filler, but they aren’t really content.

At the moment, there are a lot of AI created “classic” car photos of cars from the 1930s through 1970s that never existed. They are interesting, the lighting is well done, the image quality is great, but they are a halucination, they aren’t real. They don’t represent anything real and they don’t hold my attention, I gloss over them and go on to real photos of real vehicles.

I am all for AI helping artists express themselves, provide people with tools to help them get their ideas out, but I don’t see the point of pure AI generated art, because the soul is missing.

I agree. It is the human condition again, this striving to improve, the drive to perfection. The AI is like the parent that looks at their child’s mound of homework and is frustrated that the child can’t answer the math questions, so they do all the homework for them. The child learns nothing but still gets a gold star. The child just learns to cheat and let others do the work for them at an early. age.

My teachers were great teachers, great at imparting knowledge, but the system meant they couldn’t do it at an individual level. I was ahead of many of my classmates and the system taught me to rest on my laurels. If I finished the excercises early, I was told to do my homework, I wasn’t given more complex problems to solve, I wasn’t pushed to better myself, I was taught to keep quite and do whatever I wanted, until the rest had caught up.

In fact, thinking about it, AI teachers might be one area where it might work. The AI can have the whole curriculum and when the physical teacher has given the child everything that was planned for the day, the AI could provide extrapolated problems to provide more excercises in that direction, for example.

Very much so. In fact, it is worse than that, it isn’t just taking jobs, it is taking skills. In a couple of generations, many of these jobs won’t even exist and people won’t know how to do them. Like the GPS spoofing discussed on TWIG and TWIT last week, aircraft couldn’t fly or land, famers were left twiddling their thumbs, because their equipment wasn’t working.

They didn’t just switch to manual mode and carry on, they just waited for the problem to go away, so the machines could carry on.

When I was trained as an analyst programmer in the 80s, the first thing you were taught was, when you replace a manual system with a new automated system is, that whilst the new system is more efficient - you don’t just automate the manual system, you optimise the process and then automate the optimised process - the system can fail, so you also detail a manual procedure to carry out the tasks, when the automated system fails. People can still work, maybe not at 100% efficiency, but you don’t have a standstill until the system is back online.

These days, in many areas, everything stops until the system works again. We no longer have those manual workarounds. We no longer have people with the manual skills. If the automated wood saw breaks down, the operator can’t pick up a hand saw and carry on, at a slower rate, until the automated saw is repaired, he takes a coffee break and waits for an engineer to turn up and repair the saw.

AI will extrapolate that out, so that even the people who know these skills will no longer bother with them and future generations won’t be taught them, “because there is no point, the machine does it better.” Until the machines break and nobody can do anything…

As you say, those coming into some industries today are less enthusiastic about their jobs and their chances, but they are still there. Give it a generation and they won’t even be there to be disappointed by the job opportunities in that industry.

On the other hand, even before this, we were seeing a resurgance in people looking for real art, from real artists, not cheap, mass produced reproductions. They want hand made furniture, not mass produced pressboard rubbish from the likes of Ikea…

We need to look at where it makes sense to implement AI to aid people in their jobs or their crafts, not to replace them completely. I am not against AI in general, I am against general AI.

2 Likes

I gather you’re unaware of what Pop Music is all about ? :wink:

1 Like

On a slightly divergent note, a colleague tried the new chatGPT 4 O, he is generally impressed, but the translate function, at least between English and German, is as abysmal as Google’s and all the other translation websites and AIs I’ve tried.

1 Like

Has anyone tried the real-time AI translator that’s been added to the Android phone app yet?

I got the AI updates for my old S22 this week, real-time translation when on the phone or messaging is one of many features added. Pretty impressed Samsung are extending the new AI updates to the older phones now.

I find it interesting that people rave about the translation (both things like Google Translate and on device translation), yet whenever my colleagues or I have tried it, the results are laughable to dangerous. But they are usually doing Spanish, or I believe in the Google IO presentation, Italian?

Maybe it is just English ↔ German that is the problem…

1 Like

I’ve found Google Translate (the web page) pretty decent for English to French and French to English. It’s not perfect, but it definitely has been useful for a learner who needs hints.

2 Likes

As someone who can speak both German and English fluently, to the point that I was reading over the contacts my employer was making with partners and suppliers, in both German and English, I was more looking at the translation tools as a time saver, but they are so wildly inaccurate that, after I stop laughing, I just write it out by hand.

Classic example:
“Do not open the case, high voltage inside,” Google dropped the word “not” completely from the translation, not an unimportant word, in that context.

Likewise, “do not open the case, no user serviceable parts inside” translated to “ Gehäuse öffnen, nichts drin” (Open the case, nothing inside), not really what you want to read, when paying 5,000€ for a piece of industrial equipment.

If you try it today, it works, because I raised a ticket with Google and I trained the translation engine. I think it had problems with formal English, using “don’t” worked correctly, but it got “do not” wrong 100% of the time.

Given that that is very often used in safety instructions, it makes such a translation service dangerous, when not deadly.

I tend to use leo.org and linguee.de when doing translations. They work more like a traditional 2 language dictionary and a useful reference work, showing uses of the words in context, respectively.

More than one time the Google Translate page has given me the sense it is somehow crowd sourced. It sometimes prompts me to agree with a translation or to offer a correction. It could be the case that it needs more valid input in German to/from English to get better? I know that doesn’t help you with your needs, but it might explain why it’s not been good.

1 Like

Computers can play better chess than any human, but no one wants to watch computers play against each other. On the other hand, watching humans play each other, or better yet, playing against a human yourself, is still incredibly engrossing (well for a certain type of person I guess!).

I get the point about the Gumbo query, but all of these products are extremely context dependent. Even generally “great” coding AIs can be horrible depending on the language or frameworks you’re focusing on. The letter based queries seemed to be all the rage a year ago, but I wouldn’t be surprised if all these services stopped focusing training on this quite early on. These models drift over time.

Re: the itinerary. I think there is a huge difference between a word predictor creating an itinerary and a word predictor doing the same task when it’s hooked up to the most comprehensive database of public locations (Google Maps)

Interestingly, the real value here is not the AI but the API.

Except automating concept art saves money to studios therefore they don’t care if a human produced it or not, they’ll use the machine. Or a voice actor for video game, they already use AI voices in games.