Eight years ago, I posted a video on the Risk Bites YouTube channel on what I thought at the time were some of the less obvious risks emerging around AI — not to stoke fears (not my style), but to help prepare the ground for informed approaches to navigating them.1
That was in 2018, and talking in about AI risk was somewhat niche.2
But over the past few days there’s been a flurry of interest in the possibility of AI getting so powerful that it kills us all, brought on by some high profile social media posts, commentaries, and media interviews.3 And so I thought it would be interesting to revisit the video — especially as so much of the talk around “killer AI” has been remarkably devoid of details on how, exactly, it’s going to kill us all.4
As it turns out, less has changed over the intervening eight years than might be imagined.
For context, Risk Bites is very intentionally aimed at helping viewers from all backgrounds make sense of complex risks in a very short and accessible format. As a result, the videos can be rather light on detail. But they do draw on deep expertise, and so — I would hope — are worth taking seriously.
In the 2018 video, I set out to highlight what I saw as ten risks associated with AI that were emerging, potentially serious, and worth thinking about before they blindsided us. They were far from the only risks connected with AI at the time. But they were the ones that stood out to me as I was working on the potential impacts of artificial intelligence.
They were:
Technological dependency (“Machines that make it harder to think for ourselves” as it’s articulated in the video).
Job replacement and redistribution (“Machines that take away jobs”).
Algorithmic bias (“Machines that learn from our worst habits”).
Non-transparent decision making (“Machines that make decisions we don’t understand”).
Value-misalignment (“Machines that don’t understand what’s important to people”).
Lethal Autonomous Weapons (“Machines that kill people”).
Re-writable goals (“Machines that alter their own instructions”).
Unintended consequences of goals and decisions (“Machines that make smart-dumb decisions”).
Existential risk from superintelligence (“Machines that decide we’re not needed”). And
Heuristic manipulation (“Machines that use our human weaknesses to control us”).
2023 update of the 2018 video — reflecting the release of ChatGPT. The 10 risks are identical to those in the original video.
Looking at the AI landscape in 2026, these risks are still surprisingly relevant. In fact, eight years on, I would say that they continue to remain amongst the top longer term (and more insidious) risks associated with frontier models.
What’s perhaps more surprising is that so many people are freaking out when we’ve known about this stuff for years — and there are plenty of people besides me who have been working on this.
That said, there have been some shifts in the AI risk landscape.
Among risks that have risen in significance, I would include cybersecurity, the impacts of water and energy use on local infrastructure and economies, privacy, the social and geopolitical impacts of high fidelity deep fakes, systemic AI-driven disruption (including in teaching, learning, and social/political infrastructure), governance of frontier AI models and systems, developmental impacts on children and young people, and psychological/cognitive disruption amongst users.
Some of these are touched on in the risks I identified in 2018. Others are definitely more prominently on my radar now than they were eight years ago. Even so, looking at some of the more nuanced and complex ways in which AI has the potential to cause harm, the 2018 list remains relevant — despite it being published four years before ChatGPT was launched and generative AI became a thing.
So what should we make of this?
I’m not entirely sure, other than it seems that AI developers seem to be just waking up to concerns that many of us have been grappling with for years — and frustratingly acting as if they’re the first people to notice them.
But perhaps the more relevant insight is that, potentially serious as they are, none of these risks suggest the end of humanity as we know it.
Many of them could blossom into threats that cause serious harm if they’re not anticipated and navigated effectively. And this is something that we should be taking very seriously indeed — especially as all the indications are that new and innovative approaches to understanding and managing some of these risks are needed.
But AI isn’t going to kill us all just yet.
Unless, that is, we make the mistake of either refusing to talk about AI risk (in which case we’ll end up in the farcical situation of being wiped out by something we should have seen coming), or freaking out while ignoring people and institutions who know a thing or two about risk — which, ironically, creates its own risk.
Hopefully, sense will prevail, and companies, governments, pundits, and others, will start paying increasing attention to some of the more likely (although still complex) risks of AI, while keeping an informed (rather than uninformed) eye on less likely, but not to be completely dismissed, risks.5
And maybe watching the occasional YouTube video on AI risk in the process 😊.
Having worked in risk assessment and management for most of my professional life, it never ceases to amaze me how many people equate talking about risk with fear mongering. And yet, it’s pretty much impossible to manage risks if you don’t talk about them.
I updated the video in 2023 to reflect the release of ChatGPT, but the risks remain unchanged from the 2018 video.
This was sparked in part by the resignation of Jacob Coxon from Anthropic, together with calls from Dario Amodei and others in the industry for more caution around AI development (although it does flummox me a little as to why the people developing AI are the ones both saying they should go slower, and not doing so).
Being spooked by a general feeling of dread is a very human reaction when confronted by something that you can’t explain, that you instinctively feel threatened by, and that — as a consequence — your imagination is very happy to fill in the gaps around. But acting on instinct is its own form of risk as it leads to decisions without understanding or reason.
While truly existential risks from AI are, I suspect, not that likely, I don’t think they should be dismissed. After all, it would be embarrassing if we were all wiped out by something because we didn’t have the imagination to foresee it. But there are ways of approaching low probability but high impact risks without running around like headless chickens.
Ayubj49amd@gmail.com
What's new this time is how close we are to RSI, which automates making AIs cleverer, and how far we are from automating alignment, to keep them from doing bad things. If the rate that AIs get clever progresses at machine scales, but the rate that AIs learn not to do bad things progresses at human scales, we will inevitably get very clever AI that doesn't know how not to do bad things. That won't be good at all