(#15) Templates Control Decision-Making :: Justice 4 EI Misconduct
Vavilov ¶121: Atrium Templates Facilitate 'Reverse-Engineering' Decisions
Atrium Templates (Justice 4 EI Misconduct)
Disclaimer: This site contains the Personal Opinions of Canadian Citizens. All Claims are being Tested in Court. Until then, everything is for Entertainment Purposes only.
(Continued from Error #14: Erroneous Templates ‘Reverse-Engineer’ Decisions)
Authorities
Atrium is analogous to the ‘Chinook’ AI-Template-Generation tool employed by Immigration Canada. There have been many Cases discussing this issue in that context.
This does not appear to have been adjudicated at the Appellate level (FC=23+, FCA=0), but the FC Decisions are consistent: whenever reliance on the tool leads to ignoring key Facts & Pleadings, the Decisions are deemed unreasonable.
2023 FC 775: Safarian v. Canada (C&I);
2023 FC 805: Khosravi v. Canada (C&I);
2023 FC 1521: Arani v. Canada (C&I);
That said, none of the Chinook cases appear to have addressed the fundamental questions I listed above. They largely pertain to the fairness of involving AI in the ADM’s Decision-making process. Since AI provides unprecedented efficiencies at unparalleled cost savings, it is inevitable that it will increasingly participate in government processes – maybe even replace humans one day. (Therefore, I do appreciate the Judicial reticence to ‘ban AI’ on Fairness grounds – especially by establishing Appellate Precedence…)
For that reason, I am not making any Submissions about the fairness of employing AI & ML as a technological means to ‘aid’, ‘support’, or otherwise ‘improve’ processes. These tools are all coded, trained & managed by real people – at least for now. In all the ‘Key Ps’ (People, Paper & Processes [+Programs]), human interaction is still required. (Although ‘paper’ is quickly morphing into ‘electronic records’ split across disparate data sources.) Regardless of whether the templates are auto or manually inserted, the boilerplate language itself was still written by a human. Whether the Decision is purely written or only ‘assisted’ by AI/ML tools, an ADM still must ‘Approve’ or ‘Publish’ it.
This is why the lower Courts have consistently found that relying on these tools to the degree that crucial facts, evidence & pleadings are ignored – left unaddressed in reasons – is unreasonable. This Factum contains a long list of strongly germane evidence & arguments left ignored or unaddressed by ADMs. Ergo, this Decision is Unreasonable.
And lastly, the [partial] list of questions about Atrium Templates listed above all concern the human touchpoints into the system. None of them require findings about the ‘Fairness of using AI’ as a rule. They all relate to the human side of managing & operating ‘intelligent’ programs. They equally apply, regardless of whether a TM or Bot primarily authored the Decision. At least in this context (erroneous & legally conflicting templates), any ‘algorithmic unfairness’ still requires many human ‘points-of-failure’.
(NB: Simply replace the terms ‘AI/ML’ with ‘personal assistant who Xeroxes, cuts & superglues’ to prove this point. The fundamental ‘unfairness’ in this situation is directly linked to the [in]actions of assigned ADMs – and various CEIC & SST leadership & support personnel. Isolating & removing the ‘intelligent technology’ factor does not change the fundamental problem – the unfairness is not ‘caused by’ the ‘Program’ [any ‘AI enhancements’] – they only exacerbate policy choices already present due to the ‘People’ & ‘Processes’.)
Argumentation
There are [at least] four different problems arising from Atrium ‘Template’ usage:
(1) Imposed Constraints, (2) Template Correctness, (3) Improper Application & (4) Reverse-Engineering
Constraints: There is nothing inherently ‘wrong’ with using a database of ‘templates’ to facilitate faster, ‘more consistent’ Decisions. The fairness problem arises when said templates impose ‘backbone’ Legal Tests, Arguments & Citations upon EI ADMs, who are then constrained in their analysis, reasoning & decision-making by content they did not write.
Correctness: An important corollary is the correctness of the ‘boilerplate’ template content itself: what if they contain legal errors, irrelevant citations, incorrect interpretation, or wrong tests?
Application matters too: What if the unique factual circumstances ‘at bar’ are different (distinguished) from what precipitated the original template? Errors. Justice requires that each Case be assessed & Decided on its own merits – templates potentially undermine this core right.
Reverse-Engineering: And lastly, there is a fundamental fairness problem when templates can contradict each other – standardising the Outcomes while using opposite reasons & legal conclusions to ‘get there’ – thereby reverse-engineering ‘desired results’. Using a database of such templates – one that facilitates ‘quick & easy’ ‘search & entry’ – seriously increases the opportunities for error. It permits ADMs to search for their desired Outcome, rather than analytically process the Facts & Argument to arrive at a Just & Reasonable conclusion.
ADMs are responsible for their Decisions: ‘Templatization tools’ reduce ownership by enabling ‘reliance on the system’ and ‘passing the buck’ to whomever wrote the template.
All [4] of these template-related problems listed above potentially impacted my case:
Legal Constraints & Imposition: My GD TM (Ms. Usprich) already determined – before I was even sworn-in – that my Hearing would be limited to only the ‘Misconduct Test’. 1 This summary judgement was not based upon ‘full fact-finding’ as the Hearing had not even started yet. Was this based on the Decision Template she was using? (Did she freely ‘choose it’ or was it ‘chosen for her’?)
Template Correctness: This factum contains several examples of legal errors contained in content that is reused elsewhere – indicating they are templates. 2
Specific Application: The [over]use of specific Case Law to Deny EI Benefits was based upon templates that appear in hundreds of other C19-MM Cases. Whole paragraphs with ‘canned’ citations & footnotes exist across many of these Cases: this is clearly Template-driven. 3 (cf. Problem #3: Case Law)
Reverse-Engineering Results: The selective application of KVP – justified using complex templates – proves that ‘reverse-engineering’ is happening.
Application
Using tools that: insert prewritten content (arguments, citations & footnotes) containing legal errors into Decisions – and pre-selecting the Legal Tests TMs must apply – influences the ultimate Outcome & violates ‘Procedural Fairness’ per FCA §18.1(4)(b).
In many cases, it also led to key facts & pleadings remaining unaddressed in Decisions.
And unless there is full public disclosure about how this program works – and proper opportunity to subject Template-Management decisions to Judicial Review – this feature in Atrium also violates the ‘internal records’ unreasonableness factor in Vavilov. (¶95)
Decision
We respectfully ask this Court to Quash this Decision as being Unreasonable. I have demonstrated real-world examples where legally erroneous Reasons & incorrect Citations resulted from using this ‘template’ tool. And the reasons publicly provided to justify their existence are not rationally connected to their actual use today.
Using any internal system that fundamentally alters Decision-making processes in a non-transparent way is unreasonable. As is operating a tool with so many unanswered Justice & Fairness questions, when the result is reinforcing legal errors & misdirection that results in missing key Facts & Pleadings.
Sources:
[24/10] SST (ESDC) – How SST Works re. SCC Vavilov Decision ('Reasonableness')
[18/07] SST (ESDC) – Report: Tribunal Activities & Accomplishments (2016-2017)
https://Justice4EIMisconduct.com/assets/7-Templates/SST-Report-2017.pdf
[23/02] SST (ESDC) – Evaluation of ‘Plain Language’ Decision-Writing
https://Justice4EIMisconduct.com/assets/7-Templates/SST-Decision-Writing.pdf
[24/01] SST (ESDC) – Evaluation of How Easy it is to Read SST Decisions
[24/05] SST (ESDC) – Style Guide: Social Security Tribunal of Canada Decisions
https://Justice4EIMisconduct.com/assets/7-Templates/SST-Style-Guide.pdf
I discuss this Jurisdictional error in Problem #2: Rule of Law. (¶83), (cf. ¶233b)
For example: TM Usprich’s legally erroneous holding (supposedly based on ‘Lemire’) that ‘implementing policies’ automagically makes them legally-binding ‘conditions of employment’.
This templatized error was cited to Deny 14x Claimants EI Benefits. (¶101f, FN-112f)
Citing different cases where other, unrelated Claimants self-admitted to Misconduct to justify instant Misconduct findings without requisite Fact-Finding and circularly using that [‘Misconduct’] to cite other Jurisprudence ‘justifying’ it. (Intentional refusal to Fact-Find citing ‘vires’: Jurisdiction.)