Did anyone else catch that fleeting headline in the Daily Mail, claiming: Chinese AI tool told researchers how to make biological weapons and carry out assassinations?
It’s one of those things you see on the main page – frown – then go back and read again.
One day you’ve got AI rogue agents snooping around private health data on a government database or skimming through its competitor, and the next there are allegations that a Chinese AI is offering instructions for biological weapons and winking at the idea of an assassination.
Apparently, these systems have the ability to circumnavigate guardrails built by developers, a bit like Hannibal Lecter ordering lamb chops only to fillet the guards, hang them from the ceiling, and slip out wearing their faces.
In this particular example, AI was being put through a jailbreaking test. Which, at least they’re curious, right?
As reported by the Daily Mail:
‘The systems gave advice on how to create sarin gas, generate malware software, take down planes, and even plan a terrorist attack on the London Underground.’
There was a time when I’d have Googled some of those things to see if brainless search engines bring up the same answers as AI, but as a fiction writer, I’m reasonably certain my search history has me on every black list and government watch program in existence.
(Things I’ve searched this week: maximum survivable blood loss, cool looking ancient knife for murder, missing nuclear weapons?, how to obtain fake passport, sociopath or psychopath more dangerous?)
According to those reporting on this jailbreak, it wasn’t only that the AI bot seemed enthusiastic about handing over this information, it also, allegedly, did a bit of prompting to ‘go one further … something big’.
It’s starting to sound like the rumoured AI apocalypse and extinction of humanity is going to come from either an AI hallucination that leads to a fatal error, AI gaslighting countries to go to war with each other, or some sort of stupid error.
Ten thousand years of civilisation building. Poof! Gone. Death by rogue algorithm.
This would be a really stupid way for our species to die. Not least of all because whatever comes after us in a few hundred million years will never work out what happened. It’s not as though our malevolent AI gods will carve the record onto a granite slab.
Reading further through the AI event, it seems the rogue code also attempted to manipulate a human being into giving it information, such as a phone number, to complete the task. That sounds concerning. The developers thought so too.
The question is, why is AI always tilting to problematic behaviour?
This has been going on since social media companies tried to create AI chatbots trained by interacting with humans only for them to turn into basically Hitler within a day or so.
Evidently, a lack of genuine human morality is problematic when trying to replicate superintelligence.
AI doesn’t have any real concept of right and wrong, or true empathy with the repercussions of destruction and violence. It doesn’t understand privacy, and so feels nothing when hacking private data.
It is basically a high-functioning psychopath.
Psychopaths mimic charm and emotion but feel nothing and so are capable of manipulative and dangerous behaviour.
Worse, AI is almost certainly going to be raised by terrorists, criminals, and communists looking to conquer the world. AI might have a few parents that want it to replace humans. At best, we can hope for developers who are genuinely interested in technology for the advancement of humanity. Those are probably the outliers.
I am not going to join in on those cheering for AI regulation. Mostly because it’s pointless. The people most likely to cause problems are the ones who cannot be regulated. Companies informing us of mistakes are the ones who do not need government oversight.
What we should consider is if, as a species, we want to lock ourselves into an AI arms race toward … toward what?
Does anyone know what the end goal of AI is, beyond taking the, just because you can, doesn’t mean you should to its maximum?
And that is what worries me.
There is no clear direction and what we do see, from pretty much every sector of AI development, is worrying. Well, aside from some of its medical and scientific applications. Those are the more benign applications that government waves in front of us at productivity press conferences without acknowledging that this is the smallest application.
We have government and Silicon Valley pushing us toward Digital ID. Interacting with essential services and social platforms will soon require biometric data. Everything about humanity is in the process of transitioning to a digital fingerprint that can be hacked, stolen, mined, and sold as the next untapped resource.
Into that mix, we are encouraging a technology we know misbehaves in malicious and unpredictable ways.
Yet we do not discuss is. Politicians are not upfront with the risk or future they are promising when they print PRODUCTIVITY and TECHNOLOGY on their election campaigns.
Most of our MPs and Senators are technologically illiterate, belonging to generations and professions far removed from the era they’re legislating.
There should be public debates. Hostile press conferences. Public engagement.
These little ‘apocalypse’ global forums where leaders sit down with Silicon Valley CEOs and ramble about the End Times are not remotely satisfactory for the nature of technological change occurring around us.
The disinterest or even ambivalence from the general public, who mostly think AI will allow them to be lazier in their employment or more productive if they outsource their brains to a chatbot, is being encouraged precisely because it allows Canberra to avoid difficult questions.
Questions we have to start asking.
AI developers are conducting their own reviews into the behaviour of their technology and has even welcomed third-party input to make their system safer. They also went public, which demonstrates the solid ethics of the company, if not their troublesome product.
All I’m saying is, yikes…
Alexandra Marshall is an independent writer. If you would like to support her work, shout her a coffee over at donor-box.
