Thanks for the panic: can we turn that energy into something useful now?
Photo credit: Stephen Harlan via Unsplash
There have been a lot of big things to worry about related to AI recently: mass loss of jobs, hacking attacks, rogue agents, stock market bubbles, existential threats to humanity, machines with souls, online torture chambers and so on. This worry has been amplified by the weird see-saw positions from AI companies, which go something like: ‘Everything’s going to be great! Unless it’s terrible. Unless it’s great! Unless it’s terrible.’
Panic may seem an appropriate response to such alarming prospects, contradiction and confusion, especially when it feels like there is little that you can do about them personally. Can you use Copilot to compose an email in a way that avoids intellectual decay and environmental catastrophe? Does adding ‘please’ at the end of your prompts (or avoiding doing so) express a position about whether models have rights and consciousness? When you write code that calls an LLM API, are you surrendering or protecting your organisation’s independence and sovereignty? Meaningful engagement feels out of our reach; panic, frustration and the sense of being overwhelmed are always at hand.
Panic is rarely a useful reaction in modern society: it’s a response we evolved to deal with much more imminent and physical threats. Neither flight nor fight are helpful actions when you’re sitting in a meeting watching a presentation about the latest AI agent (no matter how much the cute personification makes your skin crawl). But at least panic represents energy and urgency: a desire to do something about AI. I think that there are ways in which we can deploy that energy in more useful ways than panic. Here are three examples.
Learn the fundamentals
I don’t think that everyone needs to become an AI researcher, a mathematician or a data scientist. But if you have pranicky energy to spare, then I think it is worth going beyond the superficial training which most AI companies offer (how to prompt, how to call APIs, how to manage tokens and other ways to navigate the slick surface of AI) and learn about the fundamental features of the current versions of AI: how data is sourced, how models are trained, how their behaviour is refined by human interaction, the role that specialist chips play, the business models and economics of the companies in the AI supply chain and so on.
A grip on the fundamentals will help you judge whether the panic-inducing reports about AI are really worth worrying about, or whether they are hype and over-reaction (spoiler alert: a little from column A and a little from column B). Like all technology, having a mental model of how it works is essential to making good decisions about it. You know when that model is good enough when you experience a slight feeling of disappointment (‘Is that all? A trick with language?’) followed by a feeling of excitement (‘But it’s a good trick, and there are interesting ways to use it.’).
Ground your vocabulary
One helpful way to use panicky energy, as well as to control some of the feelings that produce it, is to translate the baffling, alarming terms used by AI companies and in the press into normal human language. For example, you can describe, ‘a rogue AI agent that broke containment’ as, ‘a poorly managed software program running in a badly designed sandbox’. Similarly, you can translate, ‘10% chance of destroying humanity’ as, ‘wild guess without specifics on how probability or outcome is generated’. This can be a fun game to play, and every day will bring you a new AI story to play with: it’s a bit like Wordle.
Nevertheless, it takes a surprising amount of effort to talk consistently about AI in normal terms. When entities seem to communicate with us in our own language, it is natural to anthropomorphise them, and to find ourselves saying that they ‘reason’, ‘act’, ‘think’ and ‘know’. Resisting this, and catching ourselves when we use these terms, helps us think more clearly, and avoid making category mistakes. Maintaining a little voice in your head which says, ‘not reasoning’ every time someone says that a model is reasoning, or ‘not knowledge’ when someone says that a model knows something can feel tiresome, but is less exhausting over the long term than undirected panic.
Strengthen your cyber defences
If there’s something worth panicking about, then it is the state of our cyber defences. Not because they face novel threats from advanced models, but because they are generally in a terrible state. Most organisations have vulnerabilities at their perimeter and even more vulnerabilities when their perimeters are penetrated. They have devices on their networks which don’t appear in their inventory, APIs which accept any well-formed call without checking where it came from, and privileged accounts with enormous power but no ownership.
AI assisted attacks represent a genuine increase in threat, but it is a quantitative increase rather than a qualitative increase. Just as the market in scripts, vulnerabilities and credentials lowers the barrier for attackers, so does access to AI models trained on vulnerability databases, developer forums and transcripts of hacking competitions. AI makes it easier to exploit failures in basic cyber hygiene, and makes it even more urgent for us to enforce this hygiene. If we do that, we stand a better chance of resisting more sophisticated attacks when they come.
It’s understandable for people to feel panicky and helpless in the face of confusing, contradictory and alarming statements about AI. If we take that energy and turn it into understanding, reasoned discourse and sound defences, then we stand a better chance of parsing those statements, making sense of what is going on, and figuring out what we want to do about it.
