I would like to propose a desirable structural AI Certification-chain whose objective is that to guarantee a basilar security and reliability of the product as this hit the market.
I envision the following hierarchical structure:
- EU Certification body: to charter operational institutes (Data agencies and Application Software).
- Data agencies (private or public), possibly segmented by field of interest (e.g. medicine, home care, school learning, autonomous driving, etc.). To maintain/mine huge quantity of data on a specific domain; certified to train AI applications.
- AI original application software producers (programmers).
- AI educational institutes and schools to educate students and benefit the future of Industry.
In the absence of a similar structure we run the risk to let super-powerful multinationals supplant designated EU Institutions as legitimate authorities, what has already happened with Google-Play Store!
- Inicie sesión para publicar comentarios
- Etiquetas
- Certification Organization Structure
Comentarios
Interesting and challenging. I came across similar ideas in the past months. What strikes me as questionable is the creation of agencies "certified to train AI applications". As the added value of AI often arises from machine learning and testing on proprietary data sets, do you imagine standard data sets for learning or would the agencies merely supervise the learning based on data from different sources?
- Inicie sesión para publicar comentarios
Jaroslav, the handling and verification of data is a profession in itself and requires expertise. The practice is not new as widely used by a specialist branch of marketing agencies; it is known with the name of "Information Database Agencies".
It consists of various statistical sources as: supermarket customer behaviour, Electric Power consumption by zip code, vehicle size held by town-area, purchaded toiletries and several other kinds of info. Purity and reliability of data is of the utmost importance as marketing decisions are taken upon those data; you will appreciate that the reliability of those agencies set the "price" to buy those data.
In the case of data mining for AI learning (and deep learning) it is very important that the software "learns" from reliable sources of information.
If you like, it may be compared to a child educated by cultured parents rather than the unschooled alternative.
- Inicie sesión para publicar comentarios
En respuesta a Jaroslav, the handling and por Elio PENNISI
Thank you, Elio. However, you are only talking about marketing date, which would be curated by these agencies. However, there is more sensitive data out there, such as healthcare records or public sector data, which arguably provides much higher value in combination with advanced AI algorithms. Would this data be curated in the same way? Correct me if I'm wrong, but I don't see a uniform approach to all AI algorithms and to all data as feasible.
- Inicie sesión para publicar comentarios
En respuesta a Thank you, Elio. However, you por Jaroslav BARAN
Jaroslav, I only made examples. In my conception, the healthcare Administration (agency, hospital, ....) would apply to the EU Authority for certification on the basis that "this Helthcare Admin. hold and continually updates patient's data that could be fed into AI systems. This Admin./hospital guarantees that those data are free of processing errors and would be released according to this (....) format and devoid of anagraphical data. The person responsible of this department is Dr. Alfa Zetha,"
The EU Authority would reply conceding certification for ....exclusively Healthcare AI Systems' use.
By the same token, other sectors would undergo the same procedure, considering that certification may be released for utilization on multiple domains. In this way the AI application producer and the whole system (Sw+hardware) would be guaranteed that "the machine" is technically and ethically reliable ....and certified to be released to the market.
- Inicie sesión para publicar comentarios
Regulation like this will apparently be necessary not only to hedge monopolist power, but also to ensure desirable levels of quality, to fight misuse, and to monitor (not control) development.
It would, furthermore, be supportive to the realization of essential elements of applied AI , responsibilty, accountability, and transparency.
- Inicie sesión para publicar comentarios