” Don’t you know the world is built with blood? And genocide! And exploitation!”

When I think about AI, I often think about Bo Burnham’s song, How The World Works:

The way that many of the frontier labs have acquired and used training data to create their AI systems seems really questionable. Meta, for example, appears to have pirated tens of millions of books (80 terabytes of text) from libraries like Anna’s Archive. The degree to which media generation models like Midjourney (images) or Suno (music) can replicate famous artists and studios is suspicious, to put it mildly. Much of this training was done without permission and against the wishes of the people and organisations who created the content.

AI is also going to impact many people negatively. To make it worse, many of the people who will suffer – and are already suffering – are those whose work was used without their permission to train the models that will make their skills less valuable. Digital artists, for example.

I hate it. I feel dirty about it.

But I also believe that the toothpaste is out of the tube, and whatever I think or feel about it, it’s not going back in.

As Bo Burnham explains in his song, it has always been thus. It’s how the world works. I don’t like it, but I also have to accept it. Instead of railing against the territory I have to adapt.

And so it goes with AI. It’s going to have an enormous impact on the future.

We have to move forwards in one way or another, and nothing is perfect. Whether or not I benefit from the sins of others who came before, there isn’t much benefit to dwelling on the issue. I can support regulation and enforcement of that regulation (and I do! Much more than most!) but I ultimately have to acknowledge reality and make the most of the territory.

You can ‘t short the apocalypse

There’s another dimension to this.

If you’re pessimistic about AI – if you think there’s a 10% or 20% chance of genuinely catastrophic outcomes – what exactly are you going to do about it?

Again, you can support and advocate for sensible regulation and its enforcement, which I do. But at the individual level, we have to adapt to the territory.

You can’t invest in a portfolio that is guaranteed to perform well if AI destroys the economy. You can’t prepare for a world where human labour has no value. You can’t hedge against scenarios where the basic structures of civilisation unravel.

The only scenarios you can actually plan for are the ones where you have agency. And those are the scenarios where AI is a tool that, in totality, increases our collective capabilities rather than destroys them.

This isn’t optimism. It’s pragmatism.

AI is transformative

With that out of the way: the new generation of AI tools are astounding.

(For the most part, when I talk about “AI” I’m talking about large language models (LLMs) like OpenAI’s ChatGPT, Anthropic’s Claude, Google’s Gemini, and others. But this also includes image diffusion models like Midjourney, and text-to-speech models. I’m playing loose with the term, but I’m mainly talking about LLMs.)

I use AI all the time. I use it to learn things. I use it to interrogate my thinking. I use it to guide me through tasks around the house I’d otherwise be afraid to do. It has expanded the range of my capabilities.

I use it to gain confidence about medical issues so I can ask better questions of professionals. I’ve used it to get perspective on heated conversations with loved ones. I use it to understand new topics, like music theory, where no human alive would patiently answer my repeated questions. I use it to make purchasing decisions I’d otherwise second-guess for weeks.

Three years ago, I couldn’t code at all. Now I can. I’ve built tools I use daily and automated processes I previously did manually. I’ve started a software company!

None of this would have been possible without AI. I didn’t have the technical knowledge or years to spend learning. But I had problems I wanted to solve and domain expertise to know what solutions would be useful.

AI removed barriers that previously stopped me.

This is transformative in ways that are hard to overstate. It’s not that AI made me the greatest developer in the world (or the country, or my city, or even my postcode). But it made me one, and is helping to make me a better one. It made me capable enough to build things that work and add value. And it’s made me more capable across dozens of other domains where I was previously limited.

If I can do this, others can too. And that means we’re heading into a world where capability curves are shifting in fundamental ways.

Intelligent idiots

Having said that: AI can also be frustratingly unreliable.

Lots of people who are AI-sceptical have had the experience of asking an AI a question and getting a terrible response. That’s often a function of using free and lower quality models, and/or anchoring to their experience from a year or three ago.

It’s still the case that even state of the art models aren’t always reliable or good at everything.

I think a lot of people don’t appreciate that there is a skill component to using AI. The same as any other tool. You have to understand what it’s good at, what it’s bad at, how to use it, and you have to understand which tool to use for the job.

Another analogue is having an assistant or intern. If you want them to perform a task, you need to communicate your expectations and what you want clearly. You might get into a rhythm where things are easier than that, but it can take time, and involves building a shared understanding and level of trust. The same with AI. You need to work out what you can expect to pass onto it and how to do it.

The Best Available Human (BAH)

AI is often criticised because it can’t do a job that’s as good as an expert.

However, that isn’t necessarily the standard we should always apply. Often, AI doesn’t need to be better than the best human expert. It needs to be better than the best available human to you.

If I’m trying to understand ADHD and my options are “figure it out myself from Google searches” or “use an AI to help me navigate research and ask clarifying questions,” the AI doesn’t need to be better than a specialist. It needs to be better than my own confused Googling. And it is.

If I need to decide between two laptop models and I don’t want to just go to a store and trust the salesperson who (a) might not know much and (b) probably has incentives very different to my own, my options are “spend hours trying to decode technical specifications I don’t fully understand” or “paste the specs into an AI and have it explain the trade-offs in plain language,” the AI doesn’t need to match an IT expert’s knowledge. It needs to be better than me going through reviews and forum posts. And it is.

If I’m having a heated conversation with someone I love and I need perspective, the AI doesn’t need to be better than a trained therapist if I wasn’t going to go to a therapist in the first place. It needs to be better than stewing in my own biases without any external input. And it is.

For most people, for most tasks, the best available human is no one. There’s no expert on call. There’s no specialist who will take the time. There’s no one who knows your specific context. There’s just you, trying to figure it out alone.

That’s the real comparison. Not “AI vs expert” but “AI-assisted you vs unassisted you.”

AI is not magic. But nor is it useless. It’s a powerful tool with significant limitations. The capabilities of new state of the art models continue to increase, but they are already powerful and useful.

AI can be amazing and frustrating. It can be transformative and overhyped. It can be extremely valuable, and also very risky. All of these things, all at the same time.

The path forward

AI’s training data was acquired in ethically questionable ways. It has harmed, and will harm, people who did nothing to deserve it. The benefits are likely to be more narrowly distributed than I’d like, while the costs will probably be shared more broadly. There are legitimate concerns about labour markets, social disruption, and inequality.

All of this is true.

And also: AI increases capabilities in profound ways. It makes things possible that weren’t possible before. It can make goods and services more abundant and accessible.

Both things are true. The opportunities and the risks are real.

My conclusion: embrace it, but with nuance.

Understand the limitations. AI is a tool, not magic. Using it effectively requires judgment, domain expertise, and willingness to verify outputs.

Understand the risks. Support regulation and enforcement where appropriate. Be clear-eyed about who benefits and who bears the costs.

But also: use it. Learn how it works. Figure out how to increase your capabilities. Identify problems you care about and explore whether AI helps you solve them.

The model we should be aiming for isn’t “AI replaces humans and we implement UBI”. It’s “AI augments humans and we figure out how to increase everyone’s capabilities and make the pie much, much bigger.”

The toothpaste is out of the tube. The only scenarios you can plan for are the ones where you have agency, whether you exercise it or not. And in those scenarios, AI is a tool that can increase your capabilities in ways that weren’t possible before.

In my view, that’s not optimism but pragmatism.

And so: I embrace AI as much as I can.