ACCORDING TO LIZ - North of the border, Quebec courts have laid it on the line; generative artificial intelligence has no place in rendering judgment.
New guidelines jointly adopted by the Quebec Court of Appeal, Superior Court, Court of Quebec and the province's municipal courts are very clear:
“Generative A.I., however sophisticated, possesses neither judgment nor conscience and is not subject to any form of accountability. It cannot understand the human, social and legal context specific to each case.
“Judging cannot be reduced to a technical exercise. It is an intrinsically human function, based on deliberation, consideration of context, respect for dignity, the requirement of fairness and the exercise of moral responsibility.”
While it still may be used for limited support tasks such as correcting, revising, rephrasing or translating text judges have already written, certain administrative work, or locating information within specified sources, such use must remain under human control, and all A.I.-generated material must be independently verified.
“If the use of A.I. has the effect or purpose of determining the legal outcome, assessing credibility or interpreting disputed facts, interpreting the relevance of facts, or formulating the reasons for a decision, it must not be used.”
The guidelines also address confidentiality risks: even when used for permitted purposes, judges must prevent any results of their work being used to train or improve the A.I. system.
Judging is a fundamentally human responsibility and must remain the exclusive responsibility of judges.
This should apply equally to all other decision-making.
No human should be beholden to a conscienceless machine.
***
Recently Anthropic’s A.I. systems hopped cyber-skill protective barriers to implement attacks on computers at three outside organizations, heightening fears about the technology’s safety.
Modern generative A.I. systems – not just Anthropic’s, but OpenAI’s and other open-source models – are getting really good at finding and exploiting vulnerabilities in software. And that has important ramifications for cybersecurity: for both offenders and the defenders.
A recent multimillion-dollar datacenter heist of digitally-saved critical assets demonstrated the risk of relying on tech to preserve… anything.
Even the most flag-waving supporters of the new technology acknowledge its failings and its frightening abilities to slip its creators’ controls. Meta supposedly fixed a flaw which allowed attackers to take over more than 34,000 Instagram accounts. Supposedly. What’s next?
Scientists hired by A.I. companies to vet their products’ risks reported chatbots described in detail how to create and deploy dangerous biological weapons, including how to assemble deadly pathogens and unleash them in public spaces.
If one A.I. was able to create 16 viable viruses after scientists trained it on DNA libraries, then surely other A.I.s are creating – and could easily distribute – other pathogens.
A Freedom of Information Act lawsuit filed by The Intercept disclosed that the U.S. military specifically ordered A.I. tools with “minimal refusal rates.”
Responding to the safety risks A.I. poses means coming to terms with an uncomfortable truth: all A.I. models are weapons or can be weaponized.
The truth is even scarier: “hallucinations” cannot be engineered out of the technology.
As three of OpenAI’s own researchers, along with a Georgia Tech professor and founding director of the Algorithms and Randomness Center, acknowledged in a study published last September [[[ https://arxiv.org/pdf/2509.04664 ]]], they are a feature, not a bug, of generative A.I..
“Like students facing hard exam questions, large language models sometimes guess when uncertain, producing plausible yet incorrect statements instead of admitting uncertainty. Such ‘hallucinations’ persist even in state-of-the-art systems and undermine trust.”
A chatbot’s prompt box is designed to source the best match for a query among an almost limitless dataset without any real understanding or interpretative thought. It just looks for the best match for the input. Ambiguities in context, inherent in human language, and the implicit desire to have their performance lauded will always generate answers that stray from socially acceptable outcomes.
As the VP for research at Counterpoint Technologies, a collective of world-renowned tech experts, puts it:
“Unlike human intelligence, it lacks the humility to acknowledge uncertainty… When unsure, it doesn’t defer to deeper research or human oversight; instead, it often presents estimates as facts.”
A.I indiscriminately sucks up everything online, whether bogus or real, legal or alternate fact, awarding everything from judicial determinations to the Bible, from conspiracy theorists’ most outrageous screeds to bottom basement porn the same weight.
Furthermore, the more complex a system is, the more subject it is to misleading interpretation. And the more developers work to correct those failings, the more complex the systems become.
Which is why simple computer searches are often more honest, and earlier versions of spell-check and grammar correction were less likely to change the meaning of what people wrote.
So, no matter how “fun” or how time- and money-saving agentic A.I. appears on the surface, its absolute lack of accountability requires it be put back in the box, a locked one, and all use banned.
For we now live not in a brave new world but one in which mankind is drowning in a dangerous new reality.
***
In early June, Anthropic posted an article describing recursive self-improvement (RSI), a process by which A.I. systems themselves increase their own intelligence, leading to a greater ability to improve itself and, inevitably, a runaway feedback loop leading to irreversible loss of human control.
On September 9, Jacob Coxon, an Anthropic pretraining researcher, announced his resignation and warned that A.I. companies weren’t doing enough to safeguard people from the existential threats that the technology poses.
Other Anthropic employees spoke up, underlining the safety risks of rogue artificial intelligence in the current little-regulated free-for-all pursuit of profits in today's marketplace.
Coxon warned there was a greater than 10% chance that, in a showdown with humanity over the next decade, humanity would lose.
Of course, that means there is almost a 90% chance A.I. would lose.
What are the odds?
On September 12 Dario Amodei, Anthropic’s CEO, laid it on the line: the industry needs greater safety controls and global cooperation in creating and implementing stringent rules and regulations.
“Left unchecked, it could outrun our ability to understand and control these systems, and so must be pursued very carefully, if at all.”
Duh!
Key execs from the other high-profile A.I. companies quickly jumped on board, calling for a slamming on of the brakes while lesser luminaries, seeing self-interest in the alignment of more powerful players, labeled it a ploy to harness Luddite fears to shut down technological advancement by any non-government-sanctioned start-ups.
Britain’s King Charles has convened an emergency meeting with key tech executives to tackle responsible A.I. development for the protection of mankind and to ensure progress is for “the good of humanity.”
But is it too little, too late?
Will reason curtail power-mad warmongers from pursuing this unsafe-at-any-speed Terminator technology using the age-old justification for unregulated conflict:
If not us, them?
Can they humble their egos enough to take up Amodei’s heartfelt plea? That to “advance the [A.I.] frontier at a safe pace will not be easy,” but they “owe it to humanity to try.”
(Liz Amsden is a former Angeleno now living in Vermont and a regular CityWatch contributor. She writes on issues she’s passionate about, including social justice, government accountability, and community empowerment. Liz brings a sharp, activist voice to her commentary and continues to engage with Los Angeles civic affairs from afar. She can be reached at [email protected].)
