The cited paper is a classic example of naive writing about algorithms. Leaving aside all the questions about the nature of rules and the social contexts that shape their interpretation and use, how long would it be before device manufacturers were gaming the system in their use of language to describe their products? This is a proposal to assess the similarity of texts not of devices...
In the aviation and military world “similar” items to already certified items still need to be certified (or qualified in their jargon). Such qualification can be by test, analysis (e.g. simulation), read across to previously qualified parts, etc. This proposal seems to be suggesting adopting the read across approach but this should only be used if the changes are relatively minor. Creating a new item using different materials, different functionality and different reliability and maintenance regimes do not count. As RD has already said, manufacturers will start gaming the system to get their products through. Which is precisely what they do now when trying to get aerospace parts qualified - they try to qualify by read across because it is the quickest and cheapest method. The only thing keeping them on track is the regulatory authorities but I have little confidence our health regulators are anywhere near the level of competence, integrity and professionalism of our air safety regulators.
Sounds like a recipe for total disaster!! Decoy Ducks (Fowl play) and Trojan horses spring to mind. Will the AI bot recognise the thing that looks like a duck, quacks like a duck but is really a horse? Only humans with critical thinking brains and innate intuition can see it's not a duck and may cause more harm than good. This will end up costing more in medical care than we would have paid physical human researchers to weed out the decoys in the first place.
The over reliance on AI really worries me. We use it at work and I've named it artificial stupidity, as the simplest mistakes are made (and this is data input, really easy stuff)....and government is known to skimp. They always want the cheapest system, not the best or most efficient. Disaster looms.
The cited paper is a classic example of naive writing about algorithms. Leaving aside all the questions about the nature of rules and the social contexts that shape their interpretation and use, how long would it be before device manufacturers were gaming the system in their use of language to describe their products? This is a proposal to assess the similarity of texts not of devices...
In the aviation and military world “similar” items to already certified items still need to be certified (or qualified in their jargon). Such qualification can be by test, analysis (e.g. simulation), read across to previously qualified parts, etc. This proposal seems to be suggesting adopting the read across approach but this should only be used if the changes are relatively minor. Creating a new item using different materials, different functionality and different reliability and maintenance regimes do not count. As RD has already said, manufacturers will start gaming the system to get their products through. Which is precisely what they do now when trying to get aerospace parts qualified - they try to qualify by read across because it is the quickest and cheapest method. The only thing keeping them on track is the regulatory authorities but I have little confidence our health regulators are anywhere near the level of competence, integrity and professionalism of our air safety regulators.
Sounds like a recipe for total disaster!! Decoy Ducks (Fowl play) and Trojan horses spring to mind. Will the AI bot recognise the thing that looks like a duck, quacks like a duck but is really a horse? Only humans with critical thinking brains and innate intuition can see it's not a duck and may cause more harm than good. This will end up costing more in medical care than we would have paid physical human researchers to weed out the decoys in the first place.
Went to the premiere of Aseem Malhotra’s movie ‘Do no pharm’ at the Odeon Leicester Square London on Monday.
Quite an interesting take on profit, trial data and the failure of the regulatory bodies to regulate.
https://nopharmfilm.com/
Movie will be shown in the senate on Capitol Hill today.
The over reliance on AI really worries me. We use it at work and I've named it artificial stupidity, as the simplest mistakes are made (and this is data input, really easy stuff)....and government is known to skimp. They always want the cheapest system, not the best or most efficient. Disaster looms.