Dame Victoria is wise to warn against current risks: hallucination, automation bias and threats to judicial independence. But might her analysis be anchored more in the present generation of AI, rather than its future?
The shortcomings she identifies are mostly current engineering challenges for AI rather than intrinsic, permanent flaws. Might legal AI be used to review draft judgments: identifying overlooked authorities, exposing inconsistencies and potential bias, testing reasoning against precedent and acting as an additional checking mechanism? Properly governed, might it strengthen judicial independence?
Is the greater long-term risk, not that judges rely too much on AI, but that they fail to employ a technology capable of exposing human error while leaving judicial responsibility exactly where it belongs - with the judge?
This practitioner back in 1962 (sic) bumbled his way into an intensive criminal advocacy practice lasting forty seven years via a sub-Dickensian office with wet photocopying which made you think twice and then think again about copying ANYTHING.
BUT: though struggling daily to this day in what I laughingly call “retirement” I am a humble “survivor” of the West Mids PCC’s Ethics Panel where like my colleagues we wrestle with the ever more complex sophistries of algorithms in policing practice. Change we must embrace but surely in part the way to tackle the dangers Dame Victoria rightly identifies is to have multidisciplinary entities such as ours, so as to (try and) ensure that before any “blessing” being bestowed on a new AI application’s impact on the public there has been a meaningful human appraisal of the merits. It is hard though rewarding work.
Surely it is a model worth emulating across the entire system?
'They are trained on immense datasets...' - datasets drawing on existing copyright-protected works? For example see, Society of Authors 'Artificial Intelligence': https://societyofauthors.org/where-we-stand/artificial-intelligence/
Dame Victoria is wise to warn against current risks: hallucination, automation bias and threats to judicial independence. But might her analysis be anchored more in the present generation of AI, rather than its future?
The shortcomings she identifies are mostly current engineering challenges for AI rather than intrinsic, permanent flaws. Might legal AI be used to review draft judgments: identifying overlooked authorities, exposing inconsistencies and potential bias, testing reasoning against precedent and acting as an additional checking mechanism? Properly governed, might it strengthen judicial independence?
Is the greater long-term risk, not that judges rely too much on AI, but that they fail to employ a technology capable of exposing human error while leaving judicial responsibility exactly where it belongs - with the judge?
This practitioner back in 1962 (sic) bumbled his way into an intensive criminal advocacy practice lasting forty seven years via a sub-Dickensian office with wet photocopying which made you think twice and then think again about copying ANYTHING.
BUT: though struggling daily to this day in what I laughingly call “retirement” I am a humble “survivor” of the West Mids PCC’s Ethics Panel where like my colleagues we wrestle with the ever more complex sophistries of algorithms in policing practice. Change we must embrace but surely in part the way to tackle the dangers Dame Victoria rightly identifies is to have multidisciplinary entities such as ours, so as to (try and) ensure that before any “blessing” being bestowed on a new AI application’s impact on the public there has been a meaningful human appraisal of the merits. It is hard though rewarding work.
Surely it is a model worth emulating across the entire system?
Malcolm Fowler
🙌