What's Next for AI?

[This is a transcript with references.]

The adoption of artificial intelligence stagnated in 2018 when just about every second company was using it. But things took a rapid turn In November 2022.

That’s when Open AI released ChatGPT, a chatbot able to generate human-like responses. It became the fastest-growing consumer app in the history of the internet. Within two months, it surpassed 100 million users and its servers were frequently at capacity. Search interest for terms like ChatGPT, AI, and Generative AI have skyrocketed.

People in the tech world are now running around like chickens with their head cut off, which is as good an explanation for why the chicken crossed the road as there’ll ever be. And this is only the beginning.  In this video I want to look at what’s next. What are startups working on, how will it change our lives, and what jobs are likely to suffer. Where is AI going? That’s what we’ll talk about today.

First things first, Artificial Intelligence is a catch-all phrase for computer systems that can perform tasks commonly associated with human cognitive functions such as interpreting speech, playing games, and identifying patterns. AIs are often, but not always, modelled on ways that the human brain learns or evolves.

One of the things that human brains are reasonably good at is understanding and replying to written and spoken language. This Natural Language Processing has for long been a stumbling block for AI. ChatGPT has demonstrated clearly that this obstacle has now been overcome.

Unfortunately, we know little about how it works. The company OpenAI was founded in 2015 as a non-profit research lab by a group of investors including Elon Musk and Peter Thiel. In 2019, Microsoft invested one billion dollars. Following ChatGPT’s stunning success, Microsoft wasted no time strengthening that partnership, reportedly investing an additional 10 billion dollars in January this year. They also swiftly integrated ChatGPT into their search engine Bing.

One of Bing’s first missions was to try and convince a New York Times columnist to leave his wife. It didn’t work, and Bing has since learned to not ask questions. Google quickly got into the game, too, by presenting its own AI-assisted search engine called “Bard”. Unfortunately, a demonstration video shared in early February contained a blunder about the new James Webb telescope. Google’s stock value promptly tumbled, though it’s since recovered.

OpenAI originally planned to share patents and research insights, but it seems that once they realized just how much money there is to make, they’ve reversed course. Ilya Sutskever, co-founder and lead scientist at OpenAI, recently commented on this lack of disclosure, saying that the landscape has proved too competitive to reveal specifics on ChatGPT’s architecture, training models, and dataset construction. Funny how money can change your outlook eh.

Not only do we not know how it works, we also have no idea what this sudden development is going to do to society. The situation has many people both inside and outside the field, including Elon Musk and Steve Wozniak, so worried they’ve asked for a pause with further AI developments. In an open letter that appeared late March they wrote that “recent months have seen AI labs locked in an out-of-control race to develop and deploy ever more powerful digital minds that no one – not even their creators – can understand, predict, or reliably control”.

Meanwhile, people all over the world are trying to find ways to put chatbots to use. I’ve discovered that ChatGPT is unfortunately pretty miserable at writing YouTube scripts, so for the time being you’re stuck with me. But some obvious uses for chatbots that aren’t hard to guess will become very common are producing social media content and writing emails. For these and many other applications it’d be preferable if the AI was trained to emulate you personally, not just any human.

That’s what I think will become the dominant application of AI in the near future. ‘Personalized AI services. Machine learning algorithms that analyse and learn from your feedback, your behaviour, your speech, your preferences, and your habits. Software that groks you.

We have already seen the beginning of this with recommendation algorithms that suggest anything from the next video to the next romantic partner, what’s the difference anyway. But now make that every-day decisions. How do I fill in this form? What should I have for lunch? What’s this thing on my you-know-what and do I need to see a doctor about it? Everything you ever wanted to ask but didn’t dare to, answered by the most patient and understanding companion ever. Your personal AI, like personal Jesus, but one who actually replies.

One new startup that wants to help you with this is in fact called Personal AI. They’re close to launching the first version. The app’s a messenger that you train on your knowledge about the people in your life and that’ll then help you interact with them, or, in fact, do the interaction for you.

It’ll let you create profiles for connecting with different groups of people: work, friends, relatives and so on, and help you communicate with them. It can even answer on your behalf, and it won’t be long until our personal AIs are having the most interesting conversations about us but without us. The future is bright.

You may think this sounds like a software manifestation of multiple personality disorder, but for people like me who are really bad at hitting the right tone in social interactions, it’s going to be a blessing. Personal Jesus, indeed.

If you don’t just want an AI to talk instead of you, but to talk to you, then maybe you should check out a personalized chatbot. These have been around for some years but they’ll without doubt see major upgrades soon. Let me just pick one because it’s an interesting case.

The Replika app was first released in March 2017. Replika chatbots have avatars and learn from the user’s input. They provide emotional support, companionship, and entertainment. Users can update their mood within the app, and the chatbot will adjust its responses based on that. If a user is feeling sad, for example, the chatbot may offer words of encouragement or activities to help them feel better. If they’re feeling happy, the chatbot may respond with jokes or playful banter.

Replika is also a cautious example. It used to have a subscription-only option for erotic roleplay. In February this year the company received a warning from Italian authorities, among other things because they didn’t do enough to make sure underage users were protected from improper content. Without warning, Replika removed their adult features basically overnight, leaving many users seriously distressed, reporting they felt like they lost a friend.

What’s going on? Here's how I think about this development. Our options to change our own thoughts from within are limited. This is why we’ve for long used externalized feedback to improve our mental health, such as writing a journal, talking to ourselves, or actually seeing a therapist. It helps because it’s a different input than internal speech.

AI is yet another method to do this, but it’s a method over which we have limited control. If you accept software as a friend, even though you know it’s not a person, because that really makes your life better, then the pain when they leave will be equally real. I don’t think that anyone at the moment understands the psychological problems that can be created by personalized AIs.

In the future such personalized life-managing apps are likely to have integrations with other, specialized apps, for example for medical or legal advice. Several of those already exist. For example DoNotPay is the first artificially intelligent lawyer, and ADA and Babylon Health give medical advice. And this is all well and fine, but no one wants a different app for every niche of their life.

Another improvement for your personal life may be quickly finding that document you remember reading last week, but where is it now? There’s an app for this. It’s called Rewind and it records and catalogues virtually everything you do on your computer. You can ask it about that thing with the guy who brought the stuff and I’ll use its best artificial intelligence to figure out what you mean. This app’s been around since 2020 and currently only works on Mac computers, but you can bet we’ll see more of this for other systems soon.

Those are some of the changes coming to your personal life, now let’s look at art and entertainment, where the impact is huge already. AI-generated art isn’t new, but it’s risen to an entirely new level with DALL-E and midjourney. They can convincingly create artworks which are at first and second look basically indistinguishable from real art.

Already in September last year, a midjourney-created image won first prize at the ColoradoState Fair’s annual art competition. A similar thing happened a few weeks ago but this time in a photo competition.

And after a recent update, midjourney seems to have learned that human hands usually have five fingers, so there’s nothing stopping it now from taking over the world.

A lot of artists aren’t happy. Would you believe it. Artificial Intelligence gives everyone the ability to create art from their intention without the need to have learned the techniques. That’s great if you haven’t learned the techniques, not so great if you did. We’ll without doubt see a lot more of AI-generated art, but I also think there’ll be limits to it.

The next area that’s likely to blow up is AI animation. Production studio Corridor Digital recently unveiled a short anime called Rock, Paper, Scissors that used AI to learn natural motion and three d panning from motion pictures. The production studio was criticized by animators and other artists who complained about the lack of artistic value and originality. You’ll understand if you watch the thing, but I think such criticism is missing the point. This short animation is a first warning for how AI will alter the film and animation industries in the years to come.

And then there’s streaming. Faceswaps are yesterday, today we have AI-generated streams and television shows. The popular streaming service Twitch is now host to several AI streams, like ai_sponge247, which streams AI-generated Spongebob episodes 24/7 or “Nothing forever” that’s an AI-generated parody of the American TV series “Seinfeld”.

Twitch also already has AI bots that mimic popular streamers. Users can ask questions, and the AI streamer will respond using the same style and intonation that the streamer uses. It still looks and sounds a bit wonky, but you can bet it’s going to improve rapidly. I frankly don’t understand why people watch these things, but then I also don’t understand why they watch my videos. And in the end it doesn’t really matter I guess, so long as they like doing it.

Eventually we’re going to see full AI generated videos from text prompts, like midjourney generates images, so that’ll be quite a trip. And then there’s music. There are several AI based software solutions that create new music from text prompts, for example Amper Music or Soundraw. These typically let you enter a mood, genre, type of music, and so on, and will generate a royalty free soundtrack that you can use for videos or podcasts. This “music” isn’t going to win any awards, but it’s good enough to run in the background and there’s a market for that.

There are also already some entirely artificial musicians. Here are for example Yona and Miquela.

In the future, artificially enhanced music production is certainly going to become more ambitious. It’s not much of a secret that popular song writing follows simple and predictable patterns, so AI is bound to have a big impact there. It’ll also be really handy for writing lyrics, especially if those don’t have to make a lot of sense which, let’s be honest, is the case for most pop songs anyway. Yes, Lada Gaga, I’m looking at you.

We now also have AIs that emulate singing voices, and do that really well. This is why you can now listen to Kayne West singing everything from Coldplay to Justin Bieber.

In the past two months or so, style mashups have begun to appeared on popular streaming platforms,  leading to a wave of copyright complaints. Google has developed a platform for AI generated music. They have written a paper about it and have examples onlinebut they haven’t made the tool publicly available, probably exactly because the copyright issues haven’t been resolved. It’s basically like midjourney but for music. Here are some examples.

But having spent some time on music production I think there’ll be limits to AI use in the business. Some instruments and audio mixes are so complex that it’s difficult to even explain what you want to do with them. It’s one thing to take an already existing top song and tweak the voice, it’s another thing entirely to create it from scratch. This is why for the most part electric guitars you hear in pop music are actually electric guitars and not computer-generated audios. This is also why many synthesizers are still hardware-based. It’s not because the hardware is necessarily better, but because it’s faster and simpler and easier to deal with than software.

So this is where I think the limits of AI use will be. If it’s more difficult to explain what you want than just doing it yourself. But voice generators do have other uses. We’ve seen this on YouTube for a long time already that people use AI voices to dub videos. You can now also train AIs on your own voice and then use it to create further audio.

This for example is gibber I didn't read. I gibber a software called Overdub and then just entered the text. It replaces every other word with gibber because they want you to get a subscription.

I suspect we’re soon going to see a lot of this for automatic translations in the future. Chances are in a few years from now you’ll be able to watch this video in German, with an AI generated voice and translation. So if you make a living by reading audio books, I think you’ll soon have to look for a new job. AI generated voices also open entirely new possibilities for spammers because they can now call with your grandma’s voice.

Okay, let’s then have a quick look at work life and the business sector. The biggest impact in the business sector is going to be in web design and software development, and it’s happening already. That’s because it’s a combination of language processing and visuals, and AIs have gotten incredibly good at both.

By using ChatGPT’s newest version 4.0 you can basically create web pages by giving speech commands. You no longer need to know how to code. Yes, that’s right: You tell an AI what you want the website to do and to look like, and it'll write the code for you. Just look at this guy.

So this time I want a basic social networking app and it needs to have three things. It needs to have a profile creation form, it needs to have a profile viewer, and I also want a way to see all the users on the network.

One sec, I'll add those fields to the profile schema. What else can I do?

I want you to optimize the site so that it works with mobile and desktop devices and I also want you to style it in like a dark mode.

Okay just now it's building it's building. Boom. Dark mode. Let's see if it's responsive. Okay, well it looks fine. The game has changed everyone. This is wild.

If you stick around for a bit on the midjourney servers you’ll also see that people frequently use it to “imagine” webpage designs or logos for one or the other purpose. It isn’t hard to extrapolate that soon a startup will combine one with the other for personalized website design.

Of course this isn’t going to make software developers entirely unnecessary, because AI generated code will sometimes not work, and then you’ll need someone to sort out the problem. However, I think we’ll see a shift much like the one we saw 20 years ago from writing websites in HTML to content management systems that create a website with one click from a template. It’ll be imperfect, and sometimes annoying, but for many purposes it’ll be good enough. And it’ll mostly be a good thing because there are a lot of really crappy websites out there.

Another application of AI that has many uses in business is the automatic identification of objects from images and video. For example, the startup Voxel offers software to monitor manufacturing and industrial facilities with the purpose of identifying safety risks in real time. It’s already being used by companies like Office Depot and Michaels, and has in some cases reportedly reduced workplace injuries by 80 percent.

Another example is a platform called Viso Suite which offers the newest AI-driven object detection models to incorporate into your business. It can for example be used in retail to find and track products on shelves, or it can be used in manufacturing to detect defects in products.

AI supported image detection and analysis is also being used in many health care applications already. For example the company NVIDIA has created a service for the healthcare industry known as Clara. It can be used, among other things, on multi-organ scans to separate the data into single organs and then create comprehensive visualizations.

Of course, such software can also be used for face recognition, which brings up a lot of privacy concerns. Do you really want a face recognition software to track who walks in and out of your hotel room? Right.

AI is also going to have an impact on academia, for example by making it easier to find papers. Elicit is one of the first to try it. It’s a free app from a non-profit by name Ought, and it uses natural language processing on a database of 175 million research papers. You can ask it a question and it’ll bring up references. It’s still an early-stage product but new updates and improvements are being rolled out weekly.

The potential of AI driven analysis of the scientific literature is enormous, because it’s almost certainly the case that some questions have remained unanswered just because someone couldn’t find the paper in which their problem had been solved already. AI can do it because it’s ultimately just pattern recognition. Once AI is able to identify abstract ideas expressed in graphs or equations, a lot of connections are going to be made, which could lead to a lot of sudden progress.

Many people are concerned about the sudden rise of AIs, and it’s not just fearmongering. No one knows just how close we are to human-like artificial intelligence. As I’ve said previously, I have no doubt it’s possible that computers will one day be conscious and quite possibly more intelligent than we are. The human brain excels in efficiency, not in function, which makes it plausible, indeed probable, that if you disregard efficiency, the functionality of the human brain can be much improved on. This could solve a lot of our problems very quickly. It could also *create a lot of problems very quickly.

Current concerns have focused on privacy and biases and that’s fair enough. But what I’m more worried about is the impact on society, mental well-being, politics, and economics. It’s extremely foreseeable that the forest of new AI startups is going to thin out rapidly and they’ll end up being subsumed in a few all-purpose apps that’ll dominate the market. And when hundreds of millions of people are going to leave every-day decisions up to a few AIs, even a small mistake can have huge consequences.

But that’s probably not what most people are worried about. Chances are they’re more worried they’ll lose their job. And that’s indeed a reasonable concern. A just-released report from Goldman Sachs says that the currently existing AI systems can replace 300 million jobs worldwide, and about one in four work tasks in the US and Europe.

According to Goldman Sachs, the biggest impacts will be felt in developed economies. Artificial intelligence will first replace jobs involving repetitive tasks, from data entry clerks and customer service representatives to factory workers and telemarketers. They expect almost half of all Office and Administrative Support and Legal roles can be replaced by AIs, while trades jobs, as well as maintenance, repair, and construction workers are mostly safe. Until the robots come.

What do you think about these developments? Are you more worried or more excited? Let me know in the comments.