WTF is the Problem with AI?

The phrase ‘AI’ is bandied about so much, these days, it is in danger of replacing ‘iconic’ as the most overused and misunderstood word in common parlance.

You’re probably already bored of hearing TV presenters, podcasters, politicians and scientists banging on about it. So, WTF is the problem with AI?

Be advised, we’re going to enter this particular minefield very carefully, because any article about the inexorable rise of AI immediately divides the audience into two opposing factions faster than beef tracks by Drake and Kendrick Lamar.

This issue is like a dog whistle wrapped in a red-flag.

So, consider this you’re trigger warning – we’re going in!

There are those who feel that AI is the biggest deal since knives were introduced to bread, and there are those that seemingly feel that AI will achieve sentience, build Terminators and declare war on us.

Annoyingly, there’s value in both points of view! So, let’s try and share a little of our perspective.

First off – let’s define our terms: WTF is AI?

Well, let’s start with what AI isn’t: It isn’t Data from Star Trek. it isn’t JOSHUA from WarGames. It also isn’t The Matrix!

All of that is where Artificial General Intelligence (AGI) comes in. We’ll come to that in a mo. But, let’s break down the key differences between these two concepts in a way that even a mere flesh-and-blood human can understand.

WTF is AI?

Think of AI as a toolbox filled with specialised tools. Each tool is designed for a specific task, like a hammer or a screwdriver or a spirit level. Each tool is really good at one thing. The hammer is absolutely the OG when it comes to knocking in nails, but you can’t really turn screws with it.

Similarly, AI systems are trained to perform specific tasks exceptionally well. For instance, an AI system like DeepMind’s AlphaFold can analyse and understand the structure of proteins in an unprecedented way, leading to real medical breakthroughs.

Alternatively, an AI like Open AI’s ChatGPT can write different kinds of text, like poems or code or, now, pictures, based on a given prompt which can be delivered via keyboard or verbally.

But, here’s the thing, while Open AI is kick-ass at writing a poem, it’s worse than useless at folding proteins. And while AlphaFold is leading the world in protein research, it hasn’t got a clue what iambic pentameter is.

These are both examples of Narrow AI, which excels in its designated task but cannot perform other unrelated tasks. You could make the case that, whilst these tools are very sophisticated and very fast, they aren’t really intelligent. They most definitely are not sentient beings in their own right.

Introducing AGI: The Dream of Machine Intelligence

Now, let’s talk about AGI – or ‘Artificial General Intelligence’ – often referred to as “Strong AI” or “Human-Level AI.” This is the stuff that science fiction films are full of, where machines can autonomously learn, reason and adapt to new situations without prior programming, just like humans.

That’s Data from Star Trek and JOSHUA from WarGames and it’s The Matrix and, yes, it’s SkyNet and The Terminators.

Examples of AGI:

In the real world, the scientists and engineers who have spent decades building the software and robot hardware that are coming to fruition now, weren’t planning for their work to be used to launch nuclear war on us – they imagined a brighter, utopian future.

One of the ambitions for AGI is the creation of caregivers – robots that can cook meals, clean the house, possibly administer drugs and take care of the very young and the very old.

Another ambition (which you might argue is something of a double-edged sword) is the creation of robot soldiers which can function autonomously in war zones, like human soldiers do, but without the danger to human life.

Key Differences Between AI and AGI

So, it’s not just a matter of more data and faster processors. AI and AGI aren’t in the same league.

Domain-Specific vs. General Purpose

AI is like that one friend who knows everything about vintage vinyl records but is clueless about the latest TikTok trends. That’s specialised knowledge, working within a specific domain. AI is great at specific tasks like image recognition or chess. It’s brilliant in its niche but don’t expect it to multi-task.

AGI, however, is your polymath buddy who can debate philosophy and solve calculus problems, then whip-up a gourmet meal while quoting ‘Star Wars’. It will be a ‘general-purpose intelligence’, applicable to any field, which can switch tasks effortlessly.

AGI will be more human than human!

Learning and Adaptability

AI needs a mountain of data to get good at something. It learns and improves from vast datasets but struggles to adapt if the topic changes dramatically. Also, AI can’t do critical thinking – it can’t evaluate the information it is trained on, so has no way of knowing if the ‘facts’ it has learned are real or made up.

AGI will be more like a savvy learner who picks up a new skill just by observing a couple of times. It will learn from limited data and extrapolate its learning to new, unforeseen situations. It will be adaptable, flexible and always ready for the next challenge.

Autonomy

AI is kick-ass at following strict instructions, but it’s dependent on human-defined goals and parameters. While it can optimise and innovate within these boundaries, it can’t step outside the lines drawn by its creators.

AGI will be able to set its own goals and make independent decisions. It will evaluate situations, prioritise objectives and act based on understanding, reasoning and experience, much like a human would. AGI will be autonomous, navigating complex environments without constant human oversight.

So, AGI doesn’t exist (yet), but AI does and, in a very limited and very controllable way, is making a real impact on our world. AGI is still a dream, but one that many researchers are actively pursuing right now. Opinions differ over when they will achieve AGI or whether it is even possible.
So, the AI that everyone is talking about is not the one that is going to hunt down Sarah Connor and start World War 3, okay? It would much prefer a good game of chess.

The War Against the Machines Began with ‘Generative AI’

The contentious AI tools are called ‘Generative AI’

Unlike traditional AI tools that process and analyse data to pin-point a specific result (such as facial recognition algorithms looking for a particular person on a CCTV feed, or a Search Engine looking for a specific keyword), Generative AI takes it all one step further by processing that vast amount of data, then generating something entirely new from it.

(Yes, yes, I know that the word ‘new’ might be contentious in this context, but bear with me)

Generative AI tools like Midjourney, DALL-E, Leonardo and Stable Diffusion (among many others) can create convincing and attractive images – inspired by the thousands, nay millions, of similar images they have found online.  Meanwhile, the aforementioned Chat GPT, along with a range of other ‘chatbots’, such as Perpexity, Claude, Gemini, Mistral, etc … can create written content in the style of the millions of authors whose work is online, or create something that is a synthesis of many different voices.

Speaking of voices, there are several platforms out there – such as Eleven Labs – which can convincingly simulate voices.

These tools empower those who have no artistic skills to suddenly make great paintings, drawings or photos – in the same way that chatbots like ChatGPT empower people who have never felt able to write to suddenly compose everything from a social media post to a novel.

This is brilliant for self-expression – those who have felt excluded from the creative world either from a lack of confidence or a lack of access, can now describe what they want and see it magically appear before their eyes.

Why is any of this a problem?

The Pros and Cons of Generative AI Writing Tools

Generative AI tools utilise machine learning to generate different creative text formats, from marketing copy to blog posts, even poems or code. While this technology offers exciting possibilities, it also comes with limitations. Let’s delve into the good, the bad, and the grammatically incorrect of Generative AI writing tools.

The Pros:

• Enhanced Efficiency: Writer’s block? No problem! Gen AI tools can take a simple prompt or idea and generate a range of ideas in a fraction of the time it would take a human writer. Meanwhile, this has freed-up valuable time for writers to focus on more creative aspects like research, editing and prepping the next piece.
• Overcoming Writer’s Block and Idea Generation: Staring at a blank page can be daunting. Generative AI can help break the ice by suggesting new ideas, brainstorming topics, and providing different writing angles. Not all of the ideas will be great – but there’s something to improve upon and rework.
• Grammar and Style Improvement: Many Generative AI tools offer built-in grammar and style checks. They can identify and correct errors in punctuation, sentence structure, and even suggest improvements to sentence flow and clarity. Be warned, though, even if you tell them to use English spelling (ise instead of ize) and punctuation (no Oxford Commas) it will probably go ahead and write in American English anyway. You’ll, therefore, need to use a different kind of AI, like Grammarly, to help keep your language consistent and accurate to your region.

The Cons:

• Lack of Originality and Creativity: While AI can be a great brainstorming partner, it currently struggles with true originality. The content it generates is based on the data it has been trained on, which can lead to obvious, repetitive and derivative writing.
• Factual Accuracy Concerns: Generative AI tools rely on the information they’re trained on, and that information may not always be accurate or up-to-date. This can lead to the creation of factually incorrect content, especially in complex or niche topics. Writers need to be vigilant in fact-checking any AI-generated content.
• Hallucinations: Gen AIs can sometimes “hallucinate”. This means it will produce content that appears plausible but is entirely made up. This can throw doubts onto the ‘research’ the AI has done – which is why you need to double check its sources, just to be sure it hasn’t invented them. The danger in this is that the AI, itself, can’t tell the difference between what it has found and what it has created.
• Over-reliance and Ethical Considerations: Becoming overly reliant on AI can hinder a writer’s own development. The best use of these tools is as an assistant, not a replacement for critical thinking, research and editing skills. Additionally, there are ethical considerations surrounding the use of AI-generated content, such as plagiarism, lack of consent and transparency.

The Pros and Cons of Generative AI Image Creation

Let’s begin with the benefits that AI Imagery offers:

The Pros:

• Cost-Effectiveness:

AI imagery provides a cost-effective alternative to hiring professional photographers, illustrators, artists and graphic designers – which is especially important for businesses, especially during economic downturns or for startups with limited budgets. For example, full access to Midjourney – which is pretty much the gold standard image generator – is available for as little as $10 a month. That’s just £8 (plus VAT) in real money.

• Speed of Creation:

Unlike commissioned artwork and photography, which requires time for communication and execution, AI can generate multiple image results in a matter of minutes, significantly speeding up the feedback and the amendments process in content creation.

• Customisation:

These tools respond to keywords to create tailored content that meets the challenge set by the user’s prompts. Through careful prompt engineering (which basically means choosing the right words) you can create a range of options that can be edited and customised using various settings and tools, until you get exactly what you were looking for. Parts of the image can be re-imagined, allowing a lot of direct control over the output.

And now, let’s flip Harvey Dent’s coin and look at the other side…

The Cons:

• Ethical and IP Issues: Let’s address the big ole Ai-lephant in the room, first. Generative AI tools are ‘trained’ using information scraped from the internet – designs, artworks and photographs created by real, living, humans. Their work has often been ‘scraped’ without their knowledge or specific permission. They certainly haven’t been paid for contributing their work to the knowledge base of the AI. The Generative AIs then create mash-ups of existing works without crediting the original artists or copyright owners. This can be seen as an infringement of copyrights and IP (Intellectual Property) rights.
• Employment Concerns: AI tools pose a threat to jobs in creative industries, potentially replacing human artists and designers. That was a major contributing factor to why Hollywood was on strike for much of 2023. USA Today covered the causes of the strike in this thought-provoking piece.
• Inconsistent Quality: AI doesn’t understand biology, nor the rules of physics, so it can create images which exhibit anomalies like people with too many hands or hands with too many fingers, while you don’t want to think too hard about the structural engineering of some of the buildings it designs.
• Need for Human Oversight: Despite the impressive capabilities these AI tools demonstrate, they are simply not capable of working unsupervised. In just the way that the prose written by chatbots very quickly falls into cliche and repetition, the images the text-to-image generators produce require human intervention before they can be used professionally. Images generated by AI must be vetted for issues of taste, inclusiveness, visual accuracy and derivativeness.
  

Well, all of these issues – both for and against – are also true of AI-generated video.

AI-Generated Video

Previous attempts to create videos from text prompts has resulted in surreal, painterly, obviously animated reels that were short and largely devoid of meaning and consistency. Take a look at this video from the always-overly-enthusiastic Matt Wolfe and you’ll see how limited the technology was, just a year ago.

Anyway, this all changed on February 15th, 2024. Open AI (the people behind the ChatGPT text-bot and the DALL-E text-to-image generator) unveiled Sora.

As I write, this technology still isn’t widely available, but that hasn’t softened the seismic impact it has already had. Just one week after the Sora demo-reel dropped, one-man-movie-industry, Tyler Perry, cancelled a years-in-the-planning expansion of his movie studio in Atlanta. He was about to invest $800 million in building a dozen sound-stages, now – as he tells Hollywood Reporter – he doesn’t feel he needs to, because “his productions might not have to travel to locations or build sets with the assistance of the [new] technology”.

Hollywood Embraces AI

Hollywood is already adopting the technology, and this will only increase as more tech giants unveil more AI tools. There isn’t a week that goes by without a slew of new developments. Indeed, as I was writing this section, I got a notification that Adobe is building Generative AI tools into its Premiere Pro video production platform, having already included them in its still image editor, Photoshop.

So, you can see the dilemma. Actors who have had their likeness scanned can be ‘animated’ using AI and made to say and do anything, without the original actor being consulted or even paid. This is such a concern that they even wrote it into the Ryan Gosling / Emily Blunt starrer, The Fall Guy. You wanna frame someone for a crime they didn’t commit? Use Deep Fake AI to create a video of them doing it and – bingo – prosecution rests, m’lud.

Similarly, artists, photographers and filmmakers are all concerned that they will be replaced by AI which can ape their subject matter and style, without crediting or paying them.

If these tools can make everyone into an artist, what happens to those whose livelihood relies on them selling their artistry?

Hollywood Versus AI

The spectre of AI hovered over the Hollywood strikes of 2023. The Writers Guild of America (WGA) and the Screen Actors Guild‐American Federation of Television and Radio Artists (SAG-AFTRA) went on strike together – the first joint strike in over 60 years. The central issue was the use and regulation of AI. Members of the WGA expressed concerns that producers might use AI to write scripts or complete unfinished screenplays5. They urged production houses to agree to safeguards around its usage.

The actors’ union, SAG-AFTRA, joined the WGA because they feared that AI could replace human creativity, with actors’ images being used to create characters without any humans involved. The unions wanted ironclad guarantees from studios and streamers that robots wouldn’t replace performers and writers. The technology to create without creators was emerging, and star actors feared they would lose control of their lucrative likenesses.

As we’ve discussed – the tools aren’t sufficiently well-developed to work without human guidance and editorial supervision. But that will change – and soon!

And the Winner Is…

After months of deadlock, the Hollywood writers eventually secured a significant victory and negotiated a contract that allows AI tools but sets guardrails to ensure they remain under human control. Studios cannot use AI to write or edit scripts entirely, protecting writers’ roles.

Meanwhile, The Screen Actors Guild (SAG-AFTRA) – whose beef was also with the streaming platforms for not compensating actors properly when featuring their work, successfully agreed an “unprecedented wage escalation, significantly improved streaming compensation, and the first-ever crucial protections around the use of artificial intelligence technology.”

This deal specifies that actors must give permission for AI version of their likeness to be created, and the payment they receive has to equate to the amount of work they would have done had they been there in person.

So, the unions and the creatives may have won that battle – but, at the same time, the studios and the streamers are actively hiring AI experts who will oversee the introduction of AI tools at every level of film and TV production.

So, the war rages on!

The Pros and Cons of Using AI in Film and TV Production

These days, every successful film is adapted from existing IP. Every movie script employs the same character types and the same beats as every other film. The big movie stars play the same role in every film, just in a different location and with a different character name. Therefore, you would be forgiven for thinking that robots could easily take over writing and producing those films – because originality isn’t really very important.

And you’d have a point – that’s clearly what the studios are hoping for. To achieve the same level of cookie-cutter predictability which the audience clearly craves, without the inconvenience of having to employ expensive people.

But, here’s the problem with that – the AIs are trained on what already exists. They don’t create, they mash up and reproduce. That would be fine in the short term. But, in the medium to long term – with no new ideas, no new approaches and no new interpretations entering the production flow from human creators – the films made by AI would be even more similar than Hollywood films already are until, eventually, the system would only be capable of telling one story.

This is called ‘Model Collapse’. The TLDR is that the online information the Generative AI tools are trained on is, itself, being created by AIs.

There are a range of issues associated with this but, for our purposes, the one we are concerned with is the lack of new ideas being fed into the system. As more and more of the text and imagery the AI algorithms scrape is created by AI, the more limited and repetitive the imagery will become.

AI is here. AGI is on the way. Pandora’s Box is open and we can’t close it. So, we can bitch about it, but we can’t ignore it. Those people who refuse to adapt and become familiar with at least some of the new tools, will be left behind by those who use them every day.

A Positive Use of AI

But – beyond surrendering all control to the mighty algorithms – there are functions that AI can effectively and ethically perform.

• Production Research: AI can analyse vast amounts of content to identify trends, audience preferences, and potential hits, helping studios make informed decisions about which projects to greenlight.
• Automation of Time-Consuming Tasks: AI can automate administrative tasks, such as creating comprehensive schedules based on cast and crew availability.
• Restoration and Preservation: AI can restore old films and TV shows by cleaning up footage and repairing damaged frames, preserving cultural heritage for future generations. It can also help with upscaling old films for commercial re-issue (as recently demonstrated by James Cameron on his re-issues of Aliens, The Abyss and True Lies).
• Stunt work: As demonstrated in The Fall Guy, the new technology can easily replace a stuntman’s face with an actor’s – meaning that Tom Cruise will still be able to look like he’s hanging off the side of a building or jumping off a mountain – without having to get out of bed!
• Sustainability: AI can optimise production schedules, manage resources efficiently, and reduce waste, contributing to more eco-friendly film and TV production practices.
• Accessibility Improvements: AI can generate subtitles, audio descriptions, and translations, making content more accessible to a global audience, including those with hearing or visual impairments.
• Personalisation: AI can analyse viewer preferences and generate personalised content recommendations, enhancing the viewing experience and helping audiences discover new shows and movies they’ll love.
• Fan Engagement:

AI can create interactive experiences, such as chatbots and virtual reality content, allowing fans to engage more deeply with their favourite shows and movies, enhancing overall viewer satisfaction.

In conclusion

So, the fighters are ready, the seconds are out of the ring and the bell is about to ring on what is already turning into a pretty heated debate. In the red corner, you’ve got folks cheering for AI, excited about its potential to boost creativity, streamline processes and make digital services more accessible. Over in the blue corner, there are serious concerns about job losses, ethical dilemmas, legal entanglements and concerns about the quality and sustainability of AI-generated content.

The fact is: AI is here. AGI is on the way. Pandora’s Box is open (classical reference, look it up) and we can’t close it. So, we can bitch about it, but we can’t ignore it. Those people who refuse to adapt and become familiar with at least some of the new tools, will be left behind by those who use them every day.

There isn’t anything inherently apocalyptic about AI; if we abuse the tools, or cause harm with the tools that will be entirely down to us. Just remember; any tool can be turned into a weapon in the wrong hands, just ask Jason Bourne.

Also, this tech has been out in the wild for less than two years, so we’re still in the early days. It’s too soon to predict exactly how it will shape our world, but one thing’s for sure—we need to navigate this terrain thoughtfully and keep the conversation going.

And don’t forget, when interacting with an AI chatbot, always say ‘please’ and ‘thank you’. Y’know, just in case!

Further Reading

DeepMind: https://deepmind.google/discover/blog/a-glimpse-of-the-next-generation-of-alphafold/
Forbes: https://www.forbes.com/sites/bernardmarr/2023/09/19/what-is-generative-ai-a-super-simple-explanation-anyone-can-understand/

The Wall Street Journal: https://www.wsj.com/tech/ai/the-smartest-way-to-use-ai-at-work-ce921ff4

The Guardian: https://www.theguardian.com/commentisfree/2023/oct/20/the-best-thing-i-learned-at-school-cant-be-replaced-by-ai-it-requires-hours-of-repetition-and-good-grip

MIT Technology Review: https://www.technologyreview.com/topic/artificial-intelligence/

The Atlantic:  https://www.theatlantic.com/category/ai-artificial-intelligence/

CNET: https://www.cnet.com/tech/theres-ai-and-then-theres-agi-what-you-need-to-know-to-tell-the-difference/

Westlink: https://westlink.com/blog/agi-vs-ai-understanding-the-difference/

Hollywood Reporter: https://www.independent.co.uk/arts-entertainment/films/news/hollywood-ai-writers-strike-tech-b2376457.html

Variety: https://variety.com/2023/film/news/hollywood-rocky-2023-strikes-ai-box-office-1235846165/