Fortsetzung der Diskussion von Any Stockpickers out there?:
Why not? A well trained AI has much of the information of a human on a certain subject. I would even consider it negligent not to have different AI models checking the same project or code. Of course, always trained and supervised by a human.
The simple answer is that thereâs no liability when it comes to AI, yet (if ever!).
Why? I would even consider it negligent not to have different AI models training and supervising. 
Making a management decision is different from assessing AI safety.
Even in 1979 IBM would have been foolish not to have their management decisions being supported by modern computer assessments.
Isnât this in and off itself a management decision? Safety needs accountability and liability, hence criminal proceedings when for example a building collapses, a train derails or a bar burns on new yearâs eve.
Bravo to IBM, these bona fide boomers (not merely someone who can string three words together) seem grounded andâŠhuman.
Few month ago I listend to a Economist Podcast about AI. Programming with AI, security leaks etc was adressed. They interviewed one guy from a IT security service and he claimed impressive numbers how fast AI can detect security leaks.
Unfortunately I cant figure out anymore which it was
I disagree. Assessment itself is not a management decision. The decision is to confirm that the assessment has been done to the best of oneâs knowledge and with all care and knowledge of the matter. Not including AI in such an assessment nowadays to me seems negligent.
To be clear: I think we are not that far off from each other. You might agree with me that AI should be part of the assessment process if it can add value. And I would certainly agree with you that the final rubber stamp and liability has to be with the human.
Weâre agreeing more or less 100%, my worry and potential knee-jerk reaction is when the human is taken out of the loop for whatever - usually grubby - reason.
I hate that weâre going towards a place where more and more important services with wide-ranging reach get replaced by AI and automated without due process to save bucks. This is not the 90s Microsoft support where the âsupportâ barely spoke English, Mrs Muller will be fine if âher printer is stuckâ, sheâll just have to dot matrix print her ASCII flowers and show Mr Schmied another day.
Apologies, I am ultra conservative when it comes to process.
P.S. I recall, irrelevant, my cousin noting that his big bugbear when he worked in the US was that there nearly no way to get into contact with a Homo sapiens about anything
This presents a chicken and egg situation.
Before you use AI, organisations would require an assessment and so AI cannot be used in this assessment as it would not yet have been approved.
Thatâs all valid and good.
Whatâs your take on this chicken and egg situation then? I could imagine pointing out to the auditors and/or management that AI should have been included in the assessment but that it was not possible due to missing approval. Then itâs up to them to decide how to proceed.
I would support integrating AI into the justice/executive branch of our government (Mythos level models). It could be the first level instance judgment/approval/rejection. Currently we have a big backlog for approvals/judgments. Imagine if you could submit approval documents for which u get responsens within hours. Complex cases (which are minority) would still go through humans and with of course an appeal process.
Could you give a real life exemple how it would work?
For example the ones on this list. Any process that has low rejection/acceptance rate (90+%) and has low cost for errors is a good fit
Wouldnât it be similar to a self declaration with check list and according to the criterias filled or not it shows you if it is accepted or not?
But i get your point. I sometimes just ask myself if it is just âdigital optimisationâ or âreal AIâ.
EU Product Liability Directive 2024/2853
Iâm sure it can speed up approval of minor things. Like soundproofing wall panels in Valaisan night clubs etc.
No ai governance expert is gonna make themselves redundant. Duh
