Canadian writer Cory Doctorow has described AI as a virtual asbestos that corporations are shovelling into the walls of our society. Once the mistake is admitted, he says, it will take decades to dig the rotten stuff back out again.
But will the mistake be admitted? Or do the terrible faults of AI actually suit some of the powerful groups who run the world?
Naturally I paid attention to some of the recent stories in which AI corporate overlords warned that their inventions might kill us all within a decade. I had to laugh. First, because it’s not the AI as such but the corporate overlords who are most likely to kill us, even if AI is one of the tools they use to speed the process. I mean, the gigantic corporations that rule the world are already killing the planet, so what’s new?
The second reason I laughed – a bitter laugh – was that AI is already being used to kill people. It was AI (through that vile corporation Palantir) that made the decision to blow up a girls’ school in Iran last February. The White House wanted a thousand targets to hit in the first 24 hours of its unprovoked attack and AI delivered the target list. A Pentagon review admits a mistake was made, but there isn’t anybody to blame because the US doesn’t maintain very many people to review such decisions by AI. This echoes the way Israel used its data-matching and AI systems to compile death-lists for Gaza. Algorithms matched names with alleged Hamas links and the bombs and missiles followed. To begin with people checked the target lists the algorithms compiled, but after a while it was decided there was no longer any need.
You can see the beauty of all this from certain points of view. If you don’t really care about mistakes or collateral damage then the speed at which AI compiles lists for automatic targeting is a great advantage. The second great advantage is the impunity the system provides. There is literally nobody to blame. AI did it. No human oversight.
This is exactly what the world’s nastiest governments and militaries are seeking: a fully automated system where all humans have to do is frame the instructions and let the AI and autonomous weapons systems do the rest. Nobody has to take the rap. You can minimise situations like Australia faced when its soldiers were accused of killing innocent people in cold blood in Afghanistan. And like Israel faced when evidence emerged that IDF soldiers deliberately murdered a little girl who was crying for help. And like the USA faced when Wikileaks revealed the “Collateral Murder” videos showing helicopter gun crews eagerly slaughtering civilians in Iraq. Admittedly, such crimes are almost never exposed and even more rarely punished. How much easier will it be to simply say it was the robot that stuffed up?
The impunity AI offers may be a comfort to admirals and generals when the victims are merely schoolgirls and various unarmed innocent civilians, but removing humans from the oversight loop may also bring some unanticipated dangers to those firing the missiles and dropping the bombs. For example, a report is circulating about an alleged incident where a US analyst asked an AI chatbot to help with targeting. The chatbot said a Chinese ship was carrying nuclear weapons parts to Iran. The interceptors were already their way before a human reviewed the chatbot’s decision and found it was a “hallucination”. The chatbot had simply made up the story and presented it as a formal report. One can imagine the incident that might have followed if nobody had noticed the AI-generated report was rubbish.
AI’s impunity and anonymity is also attractive to people other than bomb-droppers and assassins. The Albanese Government, for example, is very happy with the algorithmic system it uses to decide the level of care needed by old people. Doctors and nurses are cut out of the picture by this system, retaining only a clerical role in filling in a 500-question matrix. As the ABC revealed on Four Corners, most of the questions on the matrix are ignored by the algorithm, the chain of logic the algorithm uses is deeply flawed and appeals are difficult to mount and opaquely handled. The Government is happy enough, however, presumably because the system delivers financial savings and the victims don’t have anybody they can blame for unfair decisions. As an aside, this system also sounds like the one the Government is using to calculate the personal contributions that aged people pay for nursing home care. It’s a shame the Government doesn’t seem as interested in finding out how the private operators administering these aged care packages can justify the massive slices they take from the funds allocated to each client. These operators “clip the ticket” on aged care packages in a myriad ways, drastically reducing the amount that actually gets used to benefit clients. Seems to me the Government could save a lot of money reining in these robbers, instead of sooling dumb AI bots onto the hapless elderly.
The Google Maps effect
Researchers into human behaviour have recently named “the Google Maps effect”. This is where people come to depend so much on satellite navigation aids that they can longer find their way without them. They struggle to read maps or to remember routes that they’ve seen before they set of on their journeys. It’s bad enough when it’s affecting personal navigation, but it seems the effect is also degrading the ability of medical practitioners to understand the meaning of scans they review. Pathologists who are “assisted” by AI have been found to be reversing correct diagnoses under pressure from faulty AI, breast cancer screeners are increasing false-positive recalls and gastroenterologists are losing the ability to spot problematic polyps without AI help.
Commercial pilots are become so reliant on automated cruise systems that they are losing their manual recovery abilities. And software coders are finding that, even though AI can help them write great code, the effort in finding the inevitable bugs in AI-written code is horrendous.
The problem is that AI bots seem so “self-assured” and that when they don’t know answers they simply make up something that fills the gap in their knowledge. Finding the spots where they’ve done that – hidden among much output that is perfectly accurate and true – can be incredibly hard.
Hearteningly, the impunity that governments and big AI corporations are assuming is not necessarily a permanent feature. A couple of years ago Air Canada tried to argue that it wasn’t liable for faulty advice one of its chatbots provided to a customer. The bot told a customer he was entitled to a discount when he wasn’t. Air Canada refused the discount, claiming the bot was “a separate legal entity responsible for its own actions”. A court rejected this crazy claim and forced the airline to pay.
“Just give us all the data in the world and we will make perfect decisions for you.” That’s what the grimy, slimy corporate demons in charge of AI are promising. What they mean is more like, “just give us all the data in the world and we will make your lives insecure and miserable. And there won’t be a thing you will be able to do about it.”
As time passes and more and more of the detrimental effects of this new “asbestos” scourge become apparent, it’s up to all of us to reject the corporate insistence on impunity and to show corporations and governments that we can and will do something about it.
