Dear Alexandra.

Those are excellent questions and you are not alone asking them. I’ll try to give my quite short feedback directly under  your question, although I think that these questions need much wider scrutinity and discussion.

• • What type of an AI should we develop?

I’m not sure what do you mean by AI type, do you mean that technology there is behind or does the next question define the choices we have?

I’d like to make a point that in this point of history we do not have sufficient laws and standards to define the type of the AI we should build. (Local legislation is not known here). You can use any technology that is available for you, just try your very best to see that the product doesn’t harm anyone or break any laws in the intended area or country of use.

• Is machine learning build as a utility-oriented AI or as a Human-Centered AI?

Machine learning is the “culprit” of the recent forward leap of the AI development. Machine learning itself isn’t good or bad thing, nor does it build as utility- or human-centered AI. You can use it in utility centered applications, like recognizing patterns in the pictures that you take with your cell phone resulting more awesome memorabilias, or use it in the very human centered applications like speech recognition. 

• • How do we make machine learning systems accesible to all, if humans do not understand the language of AI?

Not all humans needs to understand how AI is programmed. The main thing is, that the developer of the particular AI has understood what he or she has been doing.

European Union is on its way to define the roles and laws around the AI and I’m sure these requirements will follow to the European standard of an AI. At that point we have an example which to follow when developing AI systems. One outcome of that is, that everyone should be able to use those AI-based applications and machines.

However, we shouldn’t just wait and see what there is to come on the field of legislation and standardization. We could start to build speech recognition and natural language packages for each language we have in the European Union. In that way more and more Europeans could use the AI services and utilities. Normal user would like to type or command with the voice that AI system which he or she is using with the most familiar language for him or her. That would greatly level the field of an AI, and make it available to everyone.

There is also one thing we should remember, that AI most probably will not reside in the small gadgets, but is located in the cloud service in some machine room far away from the user. Your information and usage of the AI will be processed and logged there. Where this information can be used and how much every user know about this process needs to be very open and simple to understand for everyone that is subjected for it. As you can see in here we need more of the legislation and regulation.

I hope that these thoughts were any help for you…?

One thing that you and i can do is to learn the basics of an AI. The Helsinki University has launched a free course in the Internet called the elements of an AI so that everyone can de-mystify AI and learn what it is all about. https://www.elementsofai.com/

You can find it at least on English, Finish and Swedish languages. I hope that you find it interesting.

With best wishes:  Kai Salmela  AI Specialist

Prizztech Robocoast DIH   Finland.

Thank you for the good questions;

- What type of an AI should we develop?

Counter question; What defines an AI?

AI does not really come in types, it is after all only computer code. AI would be more appropriate to be defined by its deployment, i.e. where and how we use it. A bit like people with different jobs having different titles - carpenter, programmer, manager etc. A business intelligence (BI) system could very well be said to be an AI.

- Is machine learning build as a utility-oriented AI or as a Human-Centered AI?

Once more it depends on the purpose of the coding and how it is applied/deployed (see previous answer). Both types exist.

. How do we make machine learning systems accesible to all, if humans do not understand the language of AI?

Counter question(s): What makes a generic computer program into an AI? Does AI have a specific language?

Today many people already use different types of applications of computer code without any knowledge or real understanding of what it does or how it utilizes data supplied, mobile phones, pads and other computers. So, accessibility to its functionality is often generic and often only needs limited understanding. Most Human-Computer Interaction (HCI) uses normal language to communicate with a user. Being able to actually code an 'AI program' is different, but just like with the day-to-day use of a mobile phone application not necessarily a necessity for usage. 

I think the biggest issue is the perception that the AI is something special and unique. If we look at the development of software over time the capabilities have slowly evolved from what we can call stand-alone program to integrated computer software systems and then with the development of internet the "cloud". But the software has always been able to handle input and make it's own deductions based on the programming itself. The fear of many today is that the computers now will create its own code and that there is no human control of exactly what it does - i.e. creates its own "black box".

With the continuing growth of Internet of Things (IoT), our society is already in a situation that whether we call it AI's or just program(s) or systems, we lack control over how it learns, works and interacts based on code and/or data it compiles. I think information is key as Kai Salmela mentioned below and I commend the Finnish initiative as a starting point for this. But, as an old saying goes; You can lead a horse to water, but you can't make it drink...

Morgan M. Broman, Researcher, Lawyer, System Developer...et al! :)

The rapid technological advance in non-AI domains pushes for more dialogue between AI experts and engineers, citizen science and educators.

From what I know, there is little to none interest in engaging seniors in AI initiatives, which is quite a pitty given the EU population ageing.

When we talk abut being human-centered, all demographic groups shall be included. While there is a lot of institutions adressing the young European involvment in AI, I could not find any initiative to involve seniors. Their relation to AI differs from the the one of other groups. 

I am supportive to a social-centered AI rather than human-centered AI. 

We at ADEMED Romania face this problem: there specific needs and opportunities fror elederly, in relation to AI and not only in respect of health care (which is somehow low-profile use of AI) but in developemnt of new AI features and its proper development planning.

I think in EU we trust to much the documents and we simpley forgot more than one quarter of Union's population: senior population (which includes a tremendous expertise, in the retired section of this population).