5 New Principles for Justice in the age of AI (and other networked technologies)?

@estragon sums it up:

AI is like Mr. Hyde to doctor Jekyll.

Is your companion who does difficult tasks for you, but with something in exchange.

It is the angel and the devil on your shoulder.

It is American arrogance and the desire for Chinese conquest.

It’s a dream sprinkled with terrible flashes.

It is the fierce and overbearing soul that will lead us to be less human before even realizing that we are too human to manage it.

They put into words what I have been struggling to articulate as the reason behind having organised the workshop on AI and Justice in Brussels on 19/11. This workshop was part of an ongoing conversation on our community platform ( overview here).

Its purpose is to make sense of what we can do to take agency over the development of our futures in the age of what is often framed as inevitable developments towards one of two outcomes: The undoing of humanity vs a path to save us from ourselves.

What have I understood so far? It all boils down to the following:

We need new general principles of justice as a foundation for all policymaking in an age of networked technologies.

  1. The ethics of substantial freedom. This includes citizens’ inalienable rights of ownership, control, access and possession of information about themselves and their communities.

  2. The Capability Approach as defined by Amartya Sen.

  3. Data is not a commodity. In the digital society it is the equivalent of the air we breathe.

  4. Optimisation is not neutral. The choice to optimise at all has to be assessed based on its potential impact on the substantial freedoms of the individual.

  5. “The individual”, “the citizen” and the “Human” are understood to mean individual members of the species Homo Sapiens.

If we wish to ensure rule of law in Europe, we need a principle of balance of investment when it comes to AI and other networked digital technologies.

This means that for every investment of public resources into promotion, education, research, development, or monetisation of these new technologies - we need to ensure equal injection of resources towards providing the means for the individual citizen to gain ownership, control, access and possession of information about themselves and their communities.

These resources have to be directly accessible and accountable to the individual citizen and their community.

A moonshot mission:

If we want to ensure that AI/New Internet Technologies serves european citizens and businesses Europe must invest in building, maintaining and developing public cloud infrastructure.

This is a first draft and open to suggestions. Any thoughts?

8 Likes

Nice. Are you proposing it for re:publica?

Sure if it offers an opportunity to build on/develop this further. Ideally, people would take these principles and conclusions into discussions and then feed what they learn back into the conversation.

2 Likes

I’d emphatically agree, except that I’d scrap “cloud infrastructure” and put in “open source IT infrastructure”.

The cloud is really just somebody else’s computer. There may be efficiencies of scale, obtained in exchange for a massive lack of control. Network speeds inside Europe (25 - 100 Mbit/s via ADSL and VDSL in most places now) is already sufficient for self-hosting, for taking back full control. Of course that needs a massive investment both into open source software (to make it nice and simple to use) and into skills. For the latter, I’d create an obligatory subject in school, called “server and data management”. I mean, why not. Controlling and safeguarding ones own data is a skill that will be useful even in 100 years if (!) Europe decides to go that way rather than continue on the current road of letting everyone else but not the citizen take control of citizen data …

To flesh this out a bit, the type of public and open source IT infrastructure I’d like to see for an efficient and enjoyable data-driven society would include the following. I don’t focus on AI technology, as that’s just another (and slightly more powerful) type of algorithm, not fundamentally changing what computers can be applied for.

  • Company starter kit. Basically all the tools we are building or still missing to run a small company like Edgeryders, as an integrated open source toolkit. Including an open source version of FreeAgent, which again must include cashflow projection etc. …

  • Federated messaging and calling. Instant messaging is a nightmare by now, as we lack any recognized interoperable standard (like HTML, e-mail etc.). So we have a lot of competing systems, and everyone needs up to 10 software packages for instant messaging and phone calls these days. They do the same, more or less, and it’s a nightmare to use, worse than the public phone system that is being replaced with these tools. A Europe-wide standard for mandatory interoperability between these systems would be a great start. And of course an open source, security audited reference implementation. Like e-mail servers or Riot / Matrix, this system should run decentralized on multiple servers, and people would then have that server running in their own home typically, as part of their Internet router hardware.

  • OpenStreetMap as the official map. OpenStreetMap is great because, with map software like OsmAnd, it became quite simple to contribute to a “wiki map” of the world. I do that sometimes in combination with cycling trips in the forest. It’s not perfect yet: the user experience must become better, and governments should be mandated to contribute all their own geospatial data there.

  • Consent-based advertising. Currently, advertising is very similar to the worst case of how women experience online dating: lots of attention seeking and often disgusting messages, and other uninvited and unwelcome advances. Instead, just like more modern (OkCupid style) online dating, advertising should be based on audience consent. Practically: I may accept and even welcome a product advertisement message from a company which I like, and for that I want to browse through company profiles at times and select those I like. Another case where a Europe-wide digital infrastructure would be a big win for citizens.

2 Likes
  1. Don’t we already have an open source cloud infrastructure through, e.g. OpenStack?

  2. What is the risk that Thierry Breton, former CEO of beloved household names such as ATOS, France Telecom and Bull, will wield his newly granted power over EU industrial policy to invest EU tax payer money in the high-performance computing capacity (i.e. European cloud services) of beloved household names such as Atos?

1 Like

We need new general principles of justice as a foundation for all policymaking in an age of networked technologies.

  1. The ethics of substantial freedom. This includes citizens’ inalienable rights of ownership, control, access and possession of information about themselves and their communities.

Is there a theoretical framework for this? I connect it with Philip Agre and perhaps some of the philosophical frameworks developed in the EU FIDIS program around 2005. But I also recognise some huge public-administrative challenges in e.g. Sweden with changing the ownership structure for personal data/information in this way (i.e. information about private persons has been used to “optimize” public services, and must, in order to fill this role, belong to the government).

  1. The Capability Approach as defined by Amartya Sen.

Yes!

  1. Data is not a commodity. In the digital society it is the equivalent of the air we breathe.

I wonder about the difference with copyright (and trademark) law here:

In copyright law, the holder of the right can unilaterally dictate to users of the work the terms of usage. Under data protection law, however, it is the user of the data that unilaterally dictates the terms of usage to the holder of the right. Is either of these models envisaged or something different? For the air, surely it is the user of the air (a polluting industry, for instance) that dictates to the holder of the right to air (the population, perhaps?) the terms of usage. Somehow.

  1. Optimisation is not neutral. The choice to optimise at all has to be assessed based on its potential impact on the substantial freedoms of the individual.

!!! Yes. There was a very good “understreckare” in Svenska dagbladet about this one and a half decade ago, or so, that questioned the concept of “effectivization”. What does it mean to make something “more efficient” (one of the primary goals of digitalization, in both EU and Swedish mainstream politics)?

  1. “The individual”, “the citizen” and the “Human” are understood to mean individual members of the species Homo Sapiens.

I would like more work on the Extended Mind Hypothesis (http://cogprints.org/320/1/extended.html ) and the integration of data, A.I. or electronic devices as extensions to our cognitive abilities somehow. But I should read Hilary Putnam, I suppose, to understand this better.

2 Likes

Indeed, we have. My point is rather that cloud infrastructure is not very useful in general – or should not be, as with proper decentralization and local control of data, there are only a few use cases for high-performance data centers. In political organization, we have federation (5 levels here in Germany). In data management, we only have centralization (“data centers”). It does not have to be that way …

Substantial, right?

What kind of industrial policy, tech policy and data policy would prevent the emergence of quasi-monopolies like ATOS in the first place?

1 Like

What kind of industrial policy, tech policy and data policy would prevent the emergence of quasi-monopolies like ATOS in the first place?

It’s a good question. Industrial policy ties into tech, innovation, data and security policy because it’s ultimately about preserving the material welfare of (at least parts of) the local population. In the EU it’s a huge challenge that all the local populations are not the same - e.g., benefitting ATOS will almost certainly provide a material benefit to someone in France, but it’s not clear that this benefit will disseminate to other parts of Europe in equal measure.

We can interpret “benefit” or “material welfare” as in the Capabilities Approach by Sen (as proposed by @nadia above) if that’s convenient.

This opinion piece by Piotr Arak starts out a bit shaky, but I think the latter half explains pretty well the problem of a green economy from a Polish perspective: https://euobserver.com/opinion/146932 I think we could add the “ATOS-problem” to the complexity, e.g., Arak raises the issue of the “materials industry problems” (implying, I think correctly, that subsidizing the industry for eco-materials is unlikely to increase the capabilities of Polish rural communities) - and the ATOS problem is the same.

If the goal is simply to avoid quasi-monopolies like ATOS (and I’m not sure in which industry segment they’d be particularly dominant?), then perhaps ortholiberalism coupled with some Tim Wuean concept of lowered market entry barriers would be sufficient. If the goal is, on the other hand, to spread capabilities around Europe, we’d have a more difficult task at hand.

2 Likes

… meanwhile, looks like China is having “robot judges” adjudicate “millions of cases”. It does put a new spin on this stuff.

hi everyone especially @J_Noga @teirdes @katjab

As a result of the workshop report, we have been contacted by the office of a Member of the European Parliament with the Greens/EFA group with a request for input on a text on the consumer protection aspects of regulation of AI. Is this worthwhile getting people to contribute to? Deadline is sunday

Context: https://alexandrageese.eu/three-proposals-for-artificial-intelligence-in-europe/

1 Like

Hi Nadia,

If she asked you directly, could be nice to contribute and build a connection with her (irrespective of final impact)? I guess she asked you to comment on this text right? After a quick glance, a few points that she might include in the final report:

  1. a fundamental rights impact assessment before system deployment (and even after for high-risk systems);
  2. (strict) liability for those businesses in the value chain that profit from deployment
  3. AI systems funded by public money should be open-source by default (stipulate this in tender).
1 Like

“Hi Nadia,
great! Thanks so much for taking the time to give feedback! For your points: 1. yes will include. 2. there is a special liability report on AI in the Parliament so we cannot touch this 3. I believe we have this already (see P11 - point 7.)”

Interesting! Sorry I missed this.

1 Like

Great, thanks for conveying Nadia. Overlooked that open source was already in. Indeed, there are a few different interlinking reports, all commenting on AI white paper which is itself still a kind of preliminary document.

1 Like