Faccio mea culpa… Gli informatici come me spesso credono, in virtù della loro metodologia di approccio ai problemi, di sapere quale sia il processo migliore per la soluzionedel problema, di qualsiasi problema.

Quando poi le cose non hanno l’esito sperato, c’e’ sempre una colpa di qualcuno o un problema imprevisto ed inevitabile.

Con l’età poi si capisce che in genere i problemi avevano degli aspetti taciti, non formalizzati, ecc. che, se fossero stati presi nella giusta considerazione all’inizio, avrebbero mitigato la sicumera, che la funzione obiettivo che si cercava di indirizzare non era quella appropriata, ecc.

Source: Wired

Opinion: AI For Good Is Often Bad

After speaking at an MIT conference on emerging AI technology earlier this year, I entered a lobby full of industry vendors and noticed an open doorway leading to tall grass and shrubbery recreating a slice of the African plains. I had stumbled onto TrailGuard AI, Intel’s flagship AI for Good project, which the chip company describes as an artificial intelligence solution to the crime of wildlife poaching. Walking through the faux flora and sounds of the savannah, I emerged in front of a digital screen displaying a choppy video of my trek. The AI system had detected my movements and captured digital photos of my face, framed by a rectangle with the label “poacher” highlighted in red.
WIRED OPINION

Mark Latonero (@latonero) is a fellow at the Harvard Kennedy School’s Carr Center for Human Rights Policy and a research lead at Data & Society.

I was handed a printout with my blurry image next to a picture of an elephant, along with text explaining that the TrailGuard AI camera alerts rangers to capture poachers before one of the 35,000 elephants each year are killed. Despite these good intentions, I couldn’t help but wonder: What if this happened to me in the wild? Would local authorities come to arrest me now that I had been labeled a criminal? How would I prove my innocence against the AI? Was the false positive a result of a tool like facial recognition, notoriously bad with darker skin tones, or was it something else about me? Is everyone a poacher in the eyes of Intel’s computer vision?

Intel isn’t alone. Within the last few years, a number of tech companies, from Google to Huawei, have launched their own programs under the AI for Good banner. They deploy technologies like machine-learning algorithms to address critical issues like crime, poverty, hunger, and disease. In May, French president Emmanuel Macron invited about 60 leaders of AI-driven companies, like Facebook’s Mark Zuckerberg, to a Tech for Good Summit in Paris. The same month, the United Nations in Geneva hosted its third annual AI for Global Good Summit sponsored by XPrize. (Disclosure: I have spoken at it twice.) A recent McKinsey report on AI for Social Good provides an analysis of 160 current cases claiming to use AI to address the world’s most pressing and intractable problems.

While AI for good programs often warrant genuine excitement, they should also invite increased scrutiny. Good intentions are not enough when it comes to deploying AI for those in greatest need. In fact, the fanfare around these projects smacks of tech solutionism, which can mask root causes and the risks of experimenting with AI on vulnerable people without appropriate safeguards.

Tech companies that set out to develop a tool for the common good, not only their self-interest, soon face a dilemma: They lack the expertise in the intractable social and humanitarian issues facing much of the world. That’s why companies like Intel have partnered with National Geographic and the Leonardo DiCaprio Foundation on wildlife trafficking. And why Facebook partnered with the Red Cross to find missing people after disasters. IBM’s social-good program alone boasts 19 partnerships with NGOs and government agencies. Partnerships are smart. The last thing society needs is for engineers in enclaves like Silicon Valley to deploy AI tools for global problems they know little about.
Advertisement

The deeper issue is that no massive social problem can be reduced to the solution offered by the smartest corporate technologists partnering with the most venerable international organizations. When I reached out to the head of Intel’s AI for Good program for comment, I was told that the “poacher” label I received at the TrailGuard installation was in error—the public demonstration didn’t match the reality. The real AI system, Intel assured me, only detects humans or vehicles in the vicinity of endangered elephants and leaves it to the park rangers to identify them as poachers. Despite this nuance, the AI camera still won’t detect the likely causes of poaching: corruption, disregarding the rule of law, poverty, smuggling, and the recalcitrant demand for ivory. Those who still cling to technological solutionism are operating under the false assumption that because a company’s AI application might work in one narrow area, it will work on a broad political and social problem that has vexed society for ages.

Sometimes, a company’s pro-bono projects collide with their commercial interests. Earlier this year Palantir and the World Food Programme announced a $45M partnership to use data analytics to improve food delivery in humanitarian crises. A backlash quickly ensued, led by civil society organizations concerned over issues like data privacy and surveillance, which stem from Palantir’s contracts with the military. Despite Palantir’s project helping the humanitarian organization Mercy Corps aid refugees in Jordan, protesters and even some Palantir employees have demanded the company stop helping the Immigration and Customs Enforcement detain migrants and separate families at the US border.

Even when a company’s intentions seem coherent, the reality is that for many AI applications, the current state of the art is pretty bad when applied to global populations. Researchers have found that facial recognition software, in particular, is often biased against people of color, especially those who are women. This has led to calls for a global moratorium on facial recognition and cities like San Francisco to effectively ban it. AI systems built on limited training data create inaccurate predictive models that lead to unfair outcomes. AI for good projects often amount to pilot beta testing with unproven technologies. It’s unacceptable to experiment in the real world on vulnerable people, especially without their meaningful consent. And the AI field has yet to figure out who is culpable when these systems fail and people are hurt as a result.

This is not to say tech companies should not work to serve the common good. With AI poised to impact much of our lives, they have more of a responsibility to do so. To start, companies and their partners need to move from good intentions to accountable actions that mitigate risk. They should be transparent about both benefits and harms these AI tools may have in the long run. Their publicity around the tools should reflect the reality, not the hype. To Intel’s credit, the company promised to fix that demo to avoid future confusion. It should involve local people closest to the problem in the design process and conduct independent human rights assessments to determine if a project should move forward. Overall, companies should approach any complex global problem with the humility in knowing that an AI tool won’t solve it.

via Opinion: AI For Good Is Often Bad | WIRED.

Share now →

5G is fundamental for the economic development of the country. Yes, but why?

Because of its lower latency?

Because it will allow us to talk to the fridge and to know if the yogurt has expired?

To enable the circulation of self-driving cars or to allow remote surgical operations?

Let’s try to shed some light, let’s understand – beyond the hype – why the real “killer application” of the new mobile generation, for users, will probably not be any of those advertised so far.

5G, some technical elements

Let’s remember that the promise of the distinguishing features of 5G relates to a lower energy consumption for the same bits transmitted (which is ecologically commendable) coupled with higher “bandwidth” (throughput) with lower latency; all with greater capillarity and lower emissions.

To understand throughput with a simple analogy, think of a fireman’s hose and a straw: the throughput is the amount of water that comes out of the pipe per time unit. It depends on many factors such as the section of the hose, the water pressure and, last but not least, how much water per second the aqueduct is able to actually deliver to the hose.

Latency is a measure of how long it takes for the water molecule entering the pipe to exit on the other side.

Table 1 – Throughput (” bandwidth “) of the various standards (indicative values)

(*) without congestion

Table 2 – Latency (milliseconds) of the various standards (indicative values)

It is often said that 5G will have a latency of 1ms. This is a limit, under laboratory conditions that will not be experienced in practice.

We must also ask ourselves “latency to get where?”. In order for our bit inserted in the network in Rome to get out in Sao Paulo, Brazil, even at the speed of light and with few switching devices in between, requires over 100ms (Sao Paulo is far away!). So actually the latency will go from aprox 150ms with 4G to about 120 ms with 5G. It may seem a small reduction, but it’s actually a big gain.

If we consider that most of the contents and services that we usually access are not located on the opposite side of the world but in a datacenter near to us, we can understand that when accessing blog.quintarelli.it the latency will be nearly halved. If then we put a myriad of servers, with replicated content, in each city, then the latency could fall even beyond the values indicated in the table.

Low latency, for which business reason ?

What type of business would need such a low level of latency to justify such a large and widespread server infrastructure? Certainly not streaming video – it is not dramatic to wait 30milliseconds more for a movie to start – but for example interactive games that require fast reaction times (the reaction times of sportsmen vary between 100ms of a world record-breaking 100meter and 250ms of a “normal” professional). So a professional gamer connected to a 5G access would have an advantage over other professionals using 4G, a difference that could earn him the first step of the podium. (Of course, not all of us are professional gamers).

All this, provided that the rest of the network is carefully dimensioned and has no other bottlenecks (congestion) other than the access segment. And this is rarely the case.

While I’m writing, a simple ping command (which provides an estimate of round-trip latency) to a server located at the heart of the Italian network, from my PC reports a minimum value of 46ms and maximum of 68ms, with a variability of about 50%. (long live the smartworking! With so many people normally in the office contending the available network capacity, often the variability is larger).

There is then a further element to consider, that is where to put the application: let’s think about a virtual reality or augmented reality helmet: when the wearer turns his head, in order not to feel a motion sickness, it is important that the sliding of the images seen by the eye accompanies the sliding sensation determined by the vestibular apparatus in our inner ear. Having a low latency can be very useful to mitigate this annoying sensation.

The closer the application is to the viewer, the better.

Bringing applications as close as possible to users is a paradigm that is called edge computing. Some hypothesize to deploy myriads of small datacenters close to the antennas. But what would be the application that requires such low latency, staying within the operator’s business perimeter, in order to justify these investments? (that is to justify a sufficient number of users that pay something more than the traditional price in order to benefit from this latency reduction?).

The question is not trivial, especially if you consider that the most extreme level of edge computing is to put the computer in the user’s home. If the application was delivered from a home console connected with a cable (or an ultrawideband wireless link) to the helmet, it would be (with a few exceptions) even better than if the application was delivered from a server near the network border (near the radio access portion of the network). It’d be even better if the application was directly in the helmet.

So what is 5G really for?

Some TV commercials promote the image of 5G-enabled drones delivering cups of “frappuccinos” to our door, children entertaining themselves with dinosaur holograms popping up in the garden, advanced telemedicine, so much so that a surgeon can perform a surgery while waiting for a wedding, cars driving alone in the city and spectators watching basketball games circling their point of view around the players on the field.

But are these the applications of 5G? Searching online you will find interesting ideas for less extreme use than TV advertising. I comment some of them:

• “5G will make life easier for robots, that is, the world called the Internet of Things. Things, from traffic lights to the fridge at home, always connected to the network. The goal of 5G is basically to propose in mobility all those essential features present in fixed networks, but with a latency even lower than cable connections”.
  Traffic lights and refrigerators in mobility. But then are we sure that they need all this large bandwidth and that 20ms less are decisive to decide to switch from green to yellow or to tell us that the yogurt is running out?
• “…will also allow musicians, singers and artists who are in different places to interact in a single show”.

  A reduced latency, accompanied by low-latency interfacing circuits of musical instruments, will probably allow people to play musical instruments together remotely. The task is not trivial and also depends to some extent on the type of music played, but this is a use case that seems likely, even if it is not yet the killer application of 5G.

• “This connection will be important in the medical field for the new connected ambulances that will be able to give a lot of information to the hospital even before the patient arrives. Planned uses are in telemedicine and rehabilitative robotics with the possibility to use IoT devices and remote control”.

  Of course we are talking about all those ambulances that will have on board diagnostic imaging systems (CT, MRI) that must transmit many Gigabytes in a few seconds, yesssss….. because certainly to provide saturation informations and ECG requires just a few bytes and during the lockdown we all got used to make video conferences… Likewise for physiotherapeutic rehabilitation: clearly some believe that the electromechanical machines (that today are found only in specialized centers), if used at home, need much more than the 50Mbps that a current LTE can give and you can not do without a latency as low as hundred meter professional runner latency to be able to control the patient is bending a knee.

• “In the field of security and video surveillance will be used high resolution camera with 5G support, installable in stations and crowded places”.

  4K video requires about 16Mbps, 8K video (7680 × 4320) will realistically require about 40-50Mbps. Nothing a decent landline access can’t support. Of course, you can’t watch it if you’re not at home.

• “It will be also usefulin tourism and journalism, the latter thanks to 5G will be able to provide more timely images and news in every part of the city”.

  Which, as we know, is the core problem of journalism today…

• “5G also brings benefits to self-driving cars, as they will be able to communicate in real time with the road infrastructure and obtain important information for road safety and security”.

  One wonders how autonomous cars are doing today. Do we really need that latency reduction to signal positions and obstacles? Even assuming that this is indeed the case, that a very reduced latency is absolutely necessary, then wouldn’t it make more sense to make direct connections between cars, without passing through a network that, however, adds some latency?

  As you know, I am very skeptical about the concrete possibility of having cars autonomously whizzing through the cities because of safety problems related to an uneven and above all unmonitored environment. A differrent story are highways that are homogeneous and very well supervised environments. We will probably see autonomous driving on the highway, probably with reserved lanes and more than intervehicular communication we may need a lot of mobile bandwidth to be able to work or entertain ourselves during our travels.

• “Even smart homes will benefit from 5G, all objects in the home will be able to communicate with each other, receive information from the outside and be controlled remotely from a single device”.

  What is it that can’t be done today with fixed network access and wifi? (hoping that the device that remotely controls our home belongs to us…) Are we sure that the complexity of Narrowband-IoT (NB IoT) and the 5G version will prevail over wifi and bluetooth? It seems to me that the probability that this will happen in the short to medium term is quite low, unless some black swan gives the necessary impulse.

• “An ultra-performing network, will be fundamental for the transition to the Internet of things, i.e. to ensure the development of applications and services for the Smart City based on sensors (for example for traffic control, waste collection, urban lighting, logistics)”.

  In 20 milliseconds, many cars pass through street intersections and a lot of waste is thrown. If a lamp burns, then, we really know first!

• “The new, very fast 5G mobile networks will also help improve safety on the roads. In addition to allowing more data to be transferred in the same unit of time, they have a much lower latency and a reduced error rate: if one data packet in a thousand is “lost” with 4G, with 5G you get up to one in a million. It is these last two aspects that will make the electronic security systems of cars more effective, which will “dialogue” with each other in real time and with the certainty that the information arrives”.

  About the latency I mentioned above; about the loss of packets, as people who deal with networks know, network protocols (TCP) have control mechanisms to ensure applications that all transmitted packets arrive. This aspect, therefore, is already insurable to current technologies.

• “The productive world will be revolutionized through the full digitalization of production facilities – the so-called Industry 4.0 – and the development of precision agriculture”.

  Indeed in the business market there could be an interest in 5G for IIoT (Industrial IoT), unlike the residential market. The 5G will allow a higher communication density up to one sensor per square meter, that is one million sensors per square kilometer compared to the 50 thousand allowed by the best NB IoT technologies. We must also bear in mind that in the industrial field wired network technologies are very expensive (PROFINET cables and switches cost one order of magnitude more than the non-industrial equivalents) and introduce rigidity in installations that could be more flexibly realized and reconfigured using wireless connections. (examples 1, 2, 3, 4)

Be careful! I’m not saying that 5G doesn’t help, but that the “killer applications” that are proposed today (which often include “life saving” arguments, because on those you can’t squeeze investments), with the notable exception of the business market and IIoT, are generally trivial and almost certainly destined not to materialize.

Figure 1 – Piazza Maggi, Milan

This does not mean that a new network infrastructure should not be built, on the contrary! The first motivation for a new infrastructure is that it must precede the demand and enable it: the new infrastructure will serve a demand that is not there today. Einaudi, a very famous italian economist and politician, said that markets expresses demands, not needs, meaning they express immediate requirements and not long-term needs.

Also from this point of view the idea of pooling investments to co-invest in a 5G network infrastructure seems reasonable.

The networks they are a changin’

The “mobile” networks (which are not mobile: it is people who are mobile, not the networks) work by emitting signals that fade as you move away from the antenna, so that phones and antennas have to transmit with increasing power to be able to communicate, like two people talking to each other by moving away from each other. Or, another solution, is to place many more antennas that will be much closer to the user so that they will be able to communicate with lower power; the lower the power, the the larger the number of antennas that are needed. This is what has happened with mobile telephony: at every stage of evolution from GSM to UMTS (or 3G) to HSDPA (or 3.5G) to LTE (or 4G) emissions decrease and antenna density increases.

Consequently, the wireless access segment (the part from the user to the antenna) tends to become shorter: from the many kilometers of GSM (with low bandwidth) to a few tens of meters with Wi-Fi and 5G (which provide a lot of bandwidth).

With 5G, in Italy we will have hundreds of thousands of small antennas, with very low emissions (also because we have the lowest emission constraints in the world). In the future we will find them at the base of many buildings and they will provide us ubiquitously performances similar to Wi-Fi, wherever we go.

To each of these antennas we will need to feed some bandwidth with a network connection hookup (usually fixed). For this reason it is of obvious the importance of the presence of fixed networks, whether they are via the “old cable TV” as in some European countries or telephone networks with optical fibers. But it is good to overcome also this conceptual differentiation. The network is always and only one: it serves to carry data of any type. It does not matter if it was born for the telephone (copper pair) or for TV (cable networks) or if it is made with optical fibers.

Such a high density of antennas means that each antenna will be used by fewer people. If an antenna covers a radius of two kilometers, all users in a small town will connect to it. If an antenna covers a twenty meter radius, only the few people in that building will connect to it.

The transmission capacity available from an antenna is shared with the people who connect to it. In the case of the small town, it will be shared among many people; in the second case it will be shared among very few people so that each individual in the second scenario will have much more bandwidth available than their peers in the first scenario. Increasing the capillarity of the antennas helps to reduce emissions and increase the performance available to users.

Lower emissions thanks to 5G

It seems a counter-intuitive thing: lower electromagnetic emissions with more antennas; higher capacity with lower emissions. An example helps to clarify: imagine a room with forty people and two of them talking with a megaphone. The noise pollution will be very high and the overall capacity of the room will be just one conversation. If everyone whispers, the noise pollution will be minimal and the overall capacity will be twenty simultaneous conversations.

Ultimately, the network is not what we are used to think it is: a single object managed by an operator. Instead, it is a mix of different types of routes, with different technologies, much likey a road network.

From the point of view of the user, in the future there will not be a great distinction between fixed network and wireless network. The fixed network will have a very high capillarity and at its edge there will be an antenna. If located inside a house, it will still be a Wi-Fi managed independently by the most sophisticated users, if it will be located outside of the house it will instead be a 5G network managed by an operator.

A mental image of networks

Let’s picture circles to imagine service boundaries of an operator.

Today there is a circle that reaches the users’ home, the fixed network, to which a Wifi access point is connected; it can be right next to each side of the circumference, depending on whether it is provided and managed by the operator (inside the circle) or installed directly by the user (outside the circle). Outside this circle there are the user’s devices, from PCs to smart refrigerators (!?) to TVs connected to disks to store documents, photos and movies (and managing their complexity).

Then there is another circle, the “mobile network” one, which radius is smaller. It’s the network that powers the cellular radio stations located right inside the edge of the circle, inside the perimeter of the operator, to which users connect with their mobile devices.

Over time this second circle has expanded and with the 5G it will come close to the border of the first circle, so close that many users will find it more convenient to directly connect their devices, entrust the operator with the custody of their documents, photos and movies (without having to manage the complexity) reducing its management burden, often increasing the level of security and being able to have everywhere what would otherwise only be available at home / office.

These two circles, the fixed network and the mobile network have grown closer over time and will continue to do so with 5G; there is a large overlap in the surface area of the two circles. The more widespread the “fixed” network is developed, the more it will also contribute to the infrastructure for 5G.

The telephonist’s drama

Let’s go back to the use cases: many of those told, especially those that are presented as useful to save lives, suffer from what we could call “the telephonist’s drama”. I call it this way because i often perceive it when talking to many friends, very competent people, who work for telco operators who, being born as telephone operators, keep that imprinting in their DNA and seem to me they still are a bit conditioned by that way of thinking.

An example of a typical “telephone” way of thinking is that of value added services.

Let’s imagine a system to improve driving safety, based on a myriad of temperature and humidity sensors scattered along a mountain road so that a car that arrives that behind the blind curve can be informed there is a high risk of icing.

This is a typical use case in 5G narratives.

Let’s ask ourselves: did Vodafone or Telefonica put those sensors? If Vodafone put them, will they be accessible to cars connected to Telefonica’s network? And to those connected to Iliad’s ?

If we were in a monopoly situation, the problem would not arise: the same single operator puts the sensors and connects the cars.

But in a competitive situation, how can it be solved ? All mobile operators sign service interoperability agreements (and payment settling) whereby one operator gives the other access to the information of the sensors that he has placed in street X, a second operator gives access to the information for street Y to the first operator and then periodically they make a balance to see who owes the other how much money, after compensation.

You can bet that, while they are busy negotiating multiple bilateral service roaming agreements, someone will come along who will install sensors that will communicate via the Internet, taking away these possible services from the operators and making them over the top. And remunerating them with advertising, credit card or freemium models.

Does it recall something ?

In order to avoid a similar scenario, the operators should probably make a consortium and create a joint over the top service provider, presenting themselves on the market as a single integrated entity.

I doubt it will ever happen. Genetic imprinting is hard to die. Consider that in the 2020’s 5G WhitePaper issued by the NGMN Alliance (Alliance for Next Generation Mobile Networks, an association of operators, vendors, manufacturers and research institutes operating in the mobile phone industry) the word “Internet” appears only once (just once!), in the glossary, in the definition of “IIoT: Industrial Internet of things”.

What fate for 5G networks?

I have to make a major premise: the operators’ business cases to sustain the realization of widely spread 5G networks, with myriads of access points, are anything but obvious. It is likely that we will see a gradual roll-out, starting where the market can reward technology, i.e. in the business market, especially in industry.

As far as the residential market is concerned, the destiny of 5G use, for a growing user base, could lead to a reduction in the myriad of WiFi access points, now bought everywhere and by anyone, installed by anyone and uninspected for quantity and quality of their emissions. For many users 5G could greatly reduce their need of a fixed hookup to their homes, replacing the last meters of WiFi access with the last tens of meters of 5G.

TodayAs today, the way we use the network at home or on the move, are different: at home we have our files, backups, and our devices that (in general) we cannot access when we are on the move. A scenario like the one described could allow the members of a family to benefit from all their data in mobility, anywhere, as if they were at home. Perhaps this could be a killer app (that might be of interest to operators, though in a clash with OTT and device manufacturers who already provide these services, although with a looser billing relationship with the customer ).

However, there is a fundamental consideration that must be made: history teaches us that at home as at work, on the internet, people want to do more and more things in less and less time. So the traffic increases in a sensitive and inexorable way.

It is true that in nature there are no exponential curves that grow forever (sorry Mr. Kurzweil!), that all exponential curves sooner or later become logistic curves and flatten. There is certainly an asymptote of how much bandwidth a user can consume. My impression is that we are still very far from reaching it. This seems to be confirmed also by the recent actual usage data trends: the bandwidth used in Italy (and in the world at large) is still growing by ca. 50% per year with a comparable number of users.

As Alfonso Fuggetta said with a lucky expression, one cherry pulls the other, i.e. we start with simple things and then we make more and more, more and more complex, more and more requirements and more and more bandwidth. And at the same time our expectation of ever shorter reaction times increases.

Ultimately, for users, the killer app is still human impatience.

Share now →

Post muto.

Il Ministero ha allestito 140 sale Centro Intercettazioni Telecomunicazioni (CIT) con rete dedicata e cablaggio e dotazione dedicata di Pc portatili. In ogni sala CIT è stato inoltre installato il server ministeriale e realizzato il software per la gestione dell’archivio digitale multimediale e per l’archivio documentale; 60 milioni di euro sono gli investimenti già spesi per le infrastrutture tecnologiche, per le opere murarie e per gli acquisti necessari; 700 i server e i rack dedicati alle sole intercettazioni; oltre 1100 i PC dedicati e destinati alle sale d’ascolto; circa 3.500 le persone coinvolte nella formazione specifica (personale amministrativo, magistrati e polizia giudiziaria).

via Intercettazioni, in vigore la nuova disciplina – Bonafede: investiti 60mln.

Share now →

English version below.

Questi sono dei passaporti sanitari italiani (“fedi di sanità”)  in vigore durante la peste del 1722 (grazie wikipedia, ebay).

All’inizio della pandemia avevo letto una intervista ad un famosissimo virologo americano (che non ritrovo) in cui dava per scontato che ad un certo punto avremmo avuto dei braccialetti attestanti la nostra immunità.

La Electronic frontier foundation ha scritto un pezzo molto acuminato contro l’idea di passaporti di immunità digitali. In esso mischia considerazioni a base tecnologica con considerazioni a base medica. E’ chiaro che un prerequisito affinchè un simile passaporto abbia un senso è che i test medici siano accurati e che l’immunità sia duratura e prevedibile. Non credo francamente che la EFF sia nella posizione per affermare che questo non sarà il caso. Anche la questione etica in sè delle “fedi di sanità” e delle discriminazioni che ne potrebbero derivare, a ben vedere prescindono dalla questione tecnologia; non cambia se è un braccialetto o uno smartphone. Ne scrivo più sotto.

Le critiche legate all’aspetto tecnologico mi paiono molto legate alla situazione negli USA, non tanto a quella italiana:

1. le persone si potrebbero abituare ad esibire “prove di stato” digitali che potrebbero riguardare altri aspetti delle persone che i gatekeeper considerano rilevanti – però mi pare che sia già così quando paghiamo la metropolitana con lo smartphone o esibiamo il biglietto per entrare al cinema o il titolo di viaggio in treno. Possiamo fare tutto in cartaceo ma l’utente può scegliere di farlo in digitale.
2. tutte quelle informazioni potrebbero essere accumulate in un DB – si, ma solo se chi effettua quei controlli è lo stesso soggetto. nell’esempio precedente se il gestore della metropolitana, del cinema e della ferrovia appaltassero i sistemi di verifica allo stesso soggetto con possibilità di accedere ai dati (cosa limitata dal GDPR)
3. essere obbligati a fornire il proprio telefono, sia bloccato che sbloccato, ad una autorità di controllo potrebbe essere una fonte di abusi – non so quante persone sappiano se e quando ci si può opporre a consegnare il proprio dispositivo alla polizia. In Italia ciò è possibile solo in caso di perquisizione, sequestro o arresto in flagranza. (quindi in linea generale non si può) E comunque non si è tenuti a consegnare i propri PIN/password di cifratura.
4. immagazzinare informazioni mediche nei dispositivi potrebbe essere foriero di leak di dati – a ben vedere anche accumulare i dati nei sistemi degli ospedali e dei laboratori di analisi.

Mi limito ad osservare che è possibile fare dei sistemi sicuri, che non svelino informazioni se non su decisione dell’utente. Covidcreds è un sistema interessante che si basa sul paradigma della SSI-Self Sovereign Identity (la blockchain è una scelta implementativa con vantaggi e svantaggi).

Dal punto di vista delle considerazioni etiche, la cosa è forse meno ovvia di quanto potrebbe apparire; Wikipedia riporta due punti di vista opposti:

Human Rights Watch sostiene che, richiedere certificati di immunità per lavoro o viaggio, potrebbe costringere le persone a fare esami o rischiare di perdere il lavoro, creare un incentivo perverso per le persone a infettarsi intenzionalmente per acquisire certificati di immunità e rischiare di creare un mercato nero di certificati di immunità falsi o altrimenti falsificati. Limitando le attività sociali, civiche ed economiche, i passaporti di immunità possono “aggravare le disparità esistenti di genere, razza, etnia e nazionalità”.

D’altra parte, si sostiene che [in caso di restrizioni] sarebbe sproporzionato privare le persone immuni delle loro libertà fondamentali. Questo costituirebbe invece di fatto un caso di punizione collettiva. Di conseguenza, Govind Persad e Ezekiel J. Emanuel sottolineano che un passaporto di immunità seguirebbe il “principio dell'”alternativa meno restrittiva” e potrebbe anche andare a beneficio della società: l’aumento della sicurezza e dell’attività economica consentito dalle licenze di immunità andrebbe a beneficio di chi non ha la licenza. Ad esempio, l’assunzione preferenziale di persone immunitarie in case di cura o come operatori sanitari a domicilio potrebbe ridurre la diffusione del virus in quelle strutture e proteggere meglio le persone più vulnerabili a COVID-19. Gli amici, i parenti e il clero che sono immuni potrebbero visitare i pazienti negli ospedali e nelle case di cura.

Tutto questo, per commentare il fatto che, senza clamore, le fedi di sanità sono tornate tra noi.

Il ministro della Salute Roberto Speranza ha firmato un’ordinanza che prevede tamponi obbligatori per chiunque entri in Italia da Spagna, Grecia, Malta e Croazia o, in alternativa, un test sierologico negativo fatto non più di 72 ore prima dell’ingresso nel nostro Paese. Detto in altri termini: o fai il test quando arrivi (e stai chiuso in casa fino all’esito), oppure fai un test prima ed arrivi con una fede di immunità.

Non ci abbiamo fatto molto caso, probabilmente perchè si tratta di spostamenti internazionali. Ma se dovesse essere richiesto anche per poter andare da Milano ad Ancona ? O da Torino a San Siro ? E  quando arriveranno i vaccini, vorremo limitare agli immuni di andare al cinema o allo stadio ? Cosa ne penseranno gli esercenti dei cinema davanti alla possibilità di riprendere le attività dopo oltre un anno di paralisi ? dovranno aspettare moti altri mesi sino a quando avremo raggiunto una immunità di gregge ?

Credo che tra pochi mesi (settimane) il tema diverrà di attualità.

These are Italian health passports (“fedi di sanità“) in force during the plague of 1722 (thanks wikipedia, ebay).

At the beginning of the pandemic I read an interview with a very famous American virologist (which I can’t find) in which he took it for granted that at some point we would have wristbands attesting our immunity.

The Electronic frontier foundation wrote a very sharp piece against the idea of digital immunity passports. In it, they mix technology based considerations with medical considerations. It is clear that a prerequisite for such a passport to make sense is that medical tests are accurate and that immunity is long-lasting and predictable. I frankly do not believe that EFF is in a position to say that this will not be the case. Also the ethical issue itself about the  “fedi di sanità” and the discrimination that could result from them, in the end, does not change whether it is a wristband or a smartphone. I’ll write about it below.

The criticisms related to the technological aspect seem to me very much linked to the situation in the USA, not so much to the Italian one:

1. People could get used to showing digital “proofs of status” that could be about other aspects of the people that gatekeepers consider relevant – but it seems to me that this is already the case when we pay for the subway with our smartphone or show the ticket to the cinema or the train ticket. We can do everything on paper but the user can choose to do it digitally.
2. all that information could be accumulated in a DB – yes, but only if the company performing those checks is the same. In the example above if the company managing the data for the subway, cinema and railway would be the same, with access to the data (something limited by GDPR privacy regulation)
3. being forced to provide your phone, whether locked or unlocked, to a law enforcement authority could be a source of abuse – I don’t know how many people know if and when you can object to handing over your device to the police. In Italy this is only possible in the event of a search, seizure or arrest in flagrante delicto. (so in general you cannot) And in any case you are not required to hand over your PINs/encryption passwords.
4. storing medical information in devices could be a cause of data leak – but data storage in the systems of hospitals and testing laboratories could also.

I would just point out that it is possible to make secure systems that do not disclose information except after a decision by the user. Covidcreds is an interesting system based on the SSI-Self Sovereign Identity paradigm (the blockchain is an implementation choice with advantages and disadvantages).

From the point of view of ethical considerations, this is perhaps less obvious than it might appear; Wikipedia reports two opposite points of view:

Human Rights Watch argues that applying for immunity certificates for work or travel could force people to take tests or risk losing their jobs, create a perverse incentive for people to intentionally become infected to acquire immunity certificates, and risk creating a black market for fake or otherwise forged immunity certificates. By limiting social, civic and economic activities, immunity passports can “exacerbate existing disparities in gender, race, ethnicity and nationality”.

On the other hand, it is argued that [in case of restrictions] it would be disproportionate to deprive immune persons of their fundamental freedoms. This would in fact constitute a case of collective punishment. Consequently, Govind Persad and Ezekiel J. Emanuel stress that an immunity passport would follow the ‘less restrictive alternative’ principle and could also benefit society: the increased security and economic activity allowed by immunity licences would benefit the unlicensed. For example, the preferential employment of immune persons in nursing homes or as home healthcare workers could reduce the spread of the virus in those facilities and better protect those most vulnerable to COVID-19. Friends, relatives and clergy who are immune could visit patients in hospitals and nursing homes.

All the above, to comment on the fact that, without clamor, the fedi di sanità are effectively back among us.

The Italian Minister of Health, Roberto Speranza, has signed an ordinance that provides mandatory swabs for anyone entering Italy from Spain, Greece, Malta and Croatia or, alternatively, a negative serological test done no more than 72 hours before entering our country. In other words: either you take the test when you arrive (and stay indoors until the result), or you take a test beforehand and arrive with a fede di sanità.

We didn’t pay much attention to this, probably because it’s international travel. But what if you are also required a fede di sanità to go from Milan to Ancona ? Or from Turin to San Siro stadium ? And when the vaccines arrive, do we want to limit the immune to go to the cinema or the stadium ? What will the cinema owners think about the possibility of resuming activities after more than a year of paralysis ? will they have to wait for months more until we have achieved herd immunity ?

I believe that in a few months (weeks) the subject will become topical.

Share now →

[Placeholder]

Share now →

Circa quattro mesi fa avevo sentito la necessità di replicare ad un articolo di Franco Debenedetti dal titolo “Il caso TIM e il bastone tra le reti”. Questa era la replica.

Adesso colgo lo spunto offerto da un suo nuovo articolo sull’Huffington post per dire il mio punto di vista sulla vicenda.

Ecco qui:

La situazione è in un’oggettiva impasse. Nel frattempo che si decide, Open Fiber non può smettere di cablare e a Telecom Italia aspettare fa male.

Open Fiber non può smettere perché, seppur in larga misura per cause esogene, è in ritardo e deve recuperare, altrimenti rischia di dover pagare penali. A TIM aspettare fa male perché man mano che la rete di Open Fiber si estende, la sua rete in rame (che è sempre stata citata dalla stampa come asset di ultima garanzia del debito) si deprezza.

Questo grafico mostra (in Milioni di euro) l’andamento dei ricavi e dei margini (EBITDA) del gruppo Telecom negli ultimi 11 anni in Italia e nel Resto del mondo (ROW). Va detto che il perimetro è cambiato, ovvero sono stati venduti degli asset per ridurre il debito che, come si vede, è calato. Si vede anche che il contributo dal Resto del mondo negli ultimi anni è sostanzialmente stabile.

Quest’altro grafico mostra l’andamento solo per l’Italia di ricavi, margini debito e numero di linee vendute a clienti al dettaglio (non fornite all’ingrosso ad operatori concorrenti; sulla scala di destra milioni di linee)

Premettendo che magari sbaglio e vi invito pertanto a verificare i numeri sul sito di Telecom, a giudicare da questo andamento, se dipendesse da me, anche io cercherei di andare a matrimonio con una giovinetta che promette di essere maggiormente a prova di futuro.

Il fatto è che questa evoluzione non è un fulmine a ciel sereno e IMHO l’operazione andava fatta molti anni fa. Non è avvenuto anche perché il capitale è passato di mano n volte, ciascuna volta per pochi anni. Se l’azionista avesse avuto un respiro di più lungo termine probabilmente non si sarebbe opposto alla separazione servizio/rete, anzi.

Secondo me, una rimonopolizzazione, mettendo assieme TIM ed Open Fiber, difficilmente potrebbe essere autorizzata dall’autorità Antitrust (Europea) senza che ci sia una separazione tra servizi e rete, tra operatore wholesale ed operatore retail. C’è da scommettere che più di un concorrente vi ricorrerebbe opponendosi ad un monopolio integrato e mettere d’accordo tutti gli operatori per evitare un ricorso mi pare francamente poco probabile, a questo punto.

Anche si optasse per una separazione wholesale/retail, una operazione del genere, comunque, non è fattibile nell’ordine dei mesi.

Rispetto ad un intervento statale, alla fine, torniamo alla domanda: per fare cosa ? Per motivazioni finanziarie o per dotare il paese di una infrastruttura strategica ? Quale ? Quale obiettivo infrastrutturale avrebbe il neo monopolista pubblico ? E con quali altri azionisti allinearsi per assicurarsi di tradurre l’interesse del Paese in obiettivi aziendali ? Per proteggersi da takeover ostili basta la Golden Share. Per decidere quale grande piano infrastrutturale fare devono essere d’accordo gli azionisti.

Ma quale è il grande piano infrastrutturale che giustifica tutto questo ? Questa è la domanda da cui partire, IMHO.

Share now →

This time I side with Trump.

Twitter reminded us that we are all guests on their platform and we follow their rules. They have decided that the enough is enough and from now on they can put warnings, next to twits to inform us that they contain falsehoods. They also decided that although Trump has repeatedly violated the terms of service, he can continue to do so because he is him, and not a generic Mrs. Anna who can quietly disappear from the social network in silence.

This is not a failure by Twitter, but by the laws. The law allows a person to lie and to insult and it makes little difference if she does it at the bar in front of 5 patrons or on a social network in front of 80 million followers. If this falsehood or insult should result in damages, these will be ascertained and possibly sanctioned, in a process’ time, in civil or criminal proceedings. Freedom of expression ends up bordering on the freedom to lie.

By law, platforms enjoy exemption from liability if they merely carry the information and make it available, without altering it. If they alter it, they assume responsibility for it. (with different nuances in Europe and the USA). This follows as a principle from the same rules for which a telephone or postal operator is not responsible for the content transported.

In fact, Twitter has not altered Trump’s message, it has changed the way it is presented. Where does Twitter get its authority to decide who should and should not be subjected to such treatment ? To decide what is true and what is not ?

In an ideal world Twitter should not intervene in any way, if the laws would suffice to promptly and effectively sanction those messages that cause socially undesirable effects.

One problem is that our institutions are analogical, they live in a material dimension that is slow and badly suited to the syncopated times of social networks. The different speed is a problem.

When, in “understanding media”, McLuhann taught us that the medium is itself a message, there were wide audience gaps between the media. Pubs and publishing were (and are) evidently in different categories. And, while ensuring freedom of expression, they had different rules. Pub customers are not required to declare their identity while the publisher is required to declare his identity. In one case the right of rectification is not required while it is in the other. The bartender is not required to be vigilant, which is required of the publisher, etc.

But today this clear separation no longer exists in social networks; now there is a continuum. A person can have a pub or a publisher audience but the rules are the same, and that’s not good. We need a gradual set of rules, with balanced rights and duties, not linked to the medium but linked to the single account (or group of accounts).

Freedom of expression (and with it the right to lie, i.e. to have no verifiable objective evidence) must be protected, but the scale imposes different rules, to be provided for gradually (“regulatory ladder”).

Scale and speed are at the root of our ineffectiveness in regulating social networks.

We are wrongly following traditional models of accountability, as Trump did in response to Twitter in order to impose editorial responsibility by decree, and we do not understand that we have to regulate scale and speed in a graduated way.

In order to regulate speed, we need to introduce a gradual friction: it does not undermine freedom of expression if a message can be redistributed, beyond a certain scale, in hours rather than in real time. Eventually, limiting the mass diffusion in real time to those subjects who fall into a certain category, deriving different rights and duties compared to a generic bystander of the social network.

Regulating scale and speed makes it possible to introduce appropriate weights and counterweights and also allows for the introduction of frictions to make enforcement practicable.

Today, institutions, which do not keep pace with the explosion of social media, delegate to platforms the control functions that are implemented thanks to artificial intelligence (a computer method that structurally produces errors). The platforms, with their own rules, become a new power next to and above the State, with no possibility of appeal, as Trump is discovering at his own expense.

We have learned of Trump, given his notoriety. We will never know, however, how many times a Mrs. Anna is censored as a result of these systems or how many Mrs. Annas have been blocked.

Regulating scale and speed, structurally providing for mechanisms of remedy (“redress by design”) can help to restore the sovereignty of judgment in the institutional seats, ensuring the necessary protection of the parties.

Share now →

Premessa: non sono un biologo, nè epidemiologo, nè virologo, quindi le mie sono argomentazioni da bar.

Qualche volta anche qualcuno al bar ci piglia, il più delle volte, no.

Un po’ di considerazioni sparse, come premesse:

Il posto al mondo dove hanno la minore diffusione del contagio è il giappone. Nessun lockdown, nessuna misura “innovativa”. Una mania per la pulizia e contatti sociali non intimi (niente baci e abbracci, strette di mano, appena uno ha un minimo cenno di raffreddore si mette la mascherina, dispenser di igienizzanti in ogni dove, se uno usa un fazzoletto e se lo mette in tasca ti guardaanao schifato, niente scarpe in casa…) Alcuni di questi tratti sono comuni con altri paesi asiatici, in misura minore. Se possono sembrare poche cose, mi dice un epidemiologo che fanno una bella differenza.

Ecco una prima proposta: per la prossima, o per la prossima ondata, più diventiamo giapponesi e meglio è.

Ci sono diversi fattori che contribuiranno (spero) a farci riprendere: dall’aumento della temperatura per la primavera, alla disponibilità di qualche farmaco (ce ne sono diversi che apparentemente si stanno giudicando utili) per abbreviare di molto la presenza in terapia intesiva, di fatto moltiplicando la disponibilità di possibilità di trattamento e la gestione di pazienti in reparto (o a casa). Arriveranno metodiche di testing diverse, per consentire di fare test su grandi numeri (oggi siamo a meno di 20k test/die, dobbiamo fare almeno 10x) ed identificare i sani e gli immuni.

Il virus colpisce perlopiù persone fragili, anziani e malati (i decessi hanno una media di 3,4 patologie) e perlopiù maschi (3/4).

Condivido l’idea del mio amico Udi Shapiro del Weizmann Institute per uscirne appena arriviamo al plateau della diffusione grazie al lockdown (e sperando ci sia un test sierologico disponibile): anzichè tenere tutti in lockdown ed isolati, vi restino solo le persone a rischio e facciamo in modo che non restino isolati e che ricevano supporto da immuni, evitando di contagiarli. In pratica, creiamo uno scudo tra le persone a rischio e le persone non a rischio mediante delle persone immuni. Potremmo chiamarlo “shielding”.

Questo richiederà un sistema di certificazione e di coordinamento. Potrebbe essere un braccialetto come dice qualche epidemiologo, o forse meglio un’app, gestendo opportunamente i necessari aspetti di anonimato con meccanismi a doppio cieco, che può diventare uno strumento di coordinamento e di supporto per le persone che ricevono aiuto e per quelle che lo forniscono. E magari usiamo quest’app come “antenna” per registrare i contatti a breve raggio (usando il bluetooth e sempre con adeguate garanzie di anonimizzazione, come fa quest’app), per un eventuale rimbalzo in autunno, se epidemiologi e virologi riterranno che possa essere una cosa utile.

Adesso sono un po’ di fretta, mi riservo di aggiornare qs. post anche sulla base dei vs. commenti

Share now →

Ricevo un gustoso articolo nella mailing di Franco Debenedetti dal titolo “Il caso TIM e il bastone tra le reti”.

Franco scrive benissimo, ma l’articolo mi pare veramente molto denso di imprecisioni (Per la verità, di alcune di queste ne avevo già parlato con lui molti mesi fa).

Dulcis in fundo, quando si parla di fare una rete unica, si parla di “scorporo della rete”, intendendo che tra servizi e infrastruttura (che hanno due profili di profittabilità, di investimenti e di tempi di ritorno diversi) si pensa a Telecom come “ciò che resta dopo lo scorporo della rete”.

E se invece ragionassimo al contrario ? Se pensassimo che si scorporano i servizi e pensassimo a Telecom come ciò che resta dopo lo scorporo dei servizi ? Avrebbe un grande futuro, da grande compagnia con un monopolio naturale e margini stabiliti dall’autorità, in grado di remunerare gli investimenti (su rete e data centre della SDN).

Questo anche perchè in realtà c’è un disallineamento (ovvio) tra obiettivi aziendali ed obiettivi del paese. Il paese ha come obiettivo avere una rete in fibra in ogni angolo del paese (si vede quanto serve adesso, con questa epidemia). L’azienda ha l’obiettivo di fare profitto. Sarà una mancanza mia, ma non ho ancora visto UN amministratore delegato di un incumbent che avesse nei propri obbiettivi la massimizzazione della copertura in fibra, ma per tutti ci sono dividendi e prezzo dell’azione.

Mi viene in mente quel convegno del 2012 organizzato dal Financial Times in cui gli AD di 11 telco europee (incumbent) chiedevano deregolamentazione per “Meeting the investment and growth challenges ahead”.
11 Amministratori delegati che nell’anno precedente avevano distribuito dividendi per oltre 20 miliardi di euro. Tre miliardi quell’anno, tre miliardi l’anno dopo, sai in 9 anni quanta rete in fibra si sarebbe fatta in Italia ?

PS. era molto tempo che non rispondevo più ad articoli sul tema, ma mi pare che il dibattito sulla rete sia ripreso e, forse complice un po’ la stanchezza dopo un decennio e più di discussioni, mi pare che qualche concetto fondamentale si sia un po’ annacquato…

Share now →

Suppose we’ve trained a system to make medical diagnoses.

The system, perfectly trained, has specific recall and accuracy. (recall: of all the diseases that have been submitted, what percentage did I recognize? precision: of all those that I identified as diseases, in what percentage were they really?)

Accuracy and recall are a short blanket. it does not depend on a “calibration” of the machine, it is not an inherent error. it is a matter of statistical properties.

We know the system, perfectly trained, will make incorrect predictions.

Of course, even aspecialist doctor makes incorrect predictions. For this reason his predictions are according to science and conscience, follow a decision-making path and sometimes he might be called to motivate his decisions.

Now suppose we provide the diagnostic machine to the doctor.

Machines, in fact, can scale in speed and quantity to levels that a human cannot reach. But when they make a prediction, they cannot explain why.

Let’s now suppose that the machine makes a prediction that differs from the doctor’s opinion.

What will the doctor do?

In my opinion, the incentive for the doctor is to confirm the prediction of the machine, even if it contradicts his own opinion.

One could argue that we could just legally recognize the probabilistic nature of  the machine’s outcomes, thus providing doctors with a way to cover themselves if they contradict the machine. But would it really be an incentive for an in depth analysis by the doctor ? I don’t think so. it would just be less troublesome and quicker to agree with the machine.

Now suppose there is a patient with a serious acute illness.

• Suppose the machine makes the correct prediction, the doctor makes a mistake and his decision is the final one. The patient gets worse and the doctor is called to justify himself. Why has the doctor stubbornly taken his position? Even the machine indicated that there was a pathology!
• Suppose the machine makes the wrong prediction and the doctor sees it right but conforms to the prediction of the machine. The patient gets worse and the doctor is called to justify himself. Why did the doctor confirm the prediction of the machine? Didn’t he recognize the symptoms? He can always argue that in science and conscience he didn’t consider them sufficient to contradict the machine.

In both cases, the doctor’s position is more defensible if it conforms to the prediction of the machine, even if he thinks otherwise, even if the doctor knows that the machine produces false predictions. In any case, he can lift his hands and say “also the machine confirmed this”.

The obvious conclusions are that doctors are redundant, that a certain level of risk (due to recall/precision of AIs) is inevitable and we should live with it.

Now let’s suppose we knowingly alter some of the predictions of the machine (keeping track of them) and inform the doctor that some of the predictions that are communicated to him, are actually the opposite of what the machine really determined (of course he doesn’t know which ones). We call these predictions “poisoned”. (and those that are not, “sincere”).

The machine makes a prediction, but the doctor is told a poisoned prediction. The machine believes that there is a pathology but the doctor is told that the machine believes that there is no pathology.

What can the doctor do now? When he believes that the prediction of the machine is wrong, will he confirm it ? And what if the machine’s diagnose was poisoned ? He must tell what he truly thinks, in science and coscience.

At this point we have a situation in which

• the prediction of the machine is sincere and that of the doctor is in agreement. in this case the diagnosis is confirmed.
• the prediction of the machine is sincere and that of the doctor is discordant.  the case is highlighted and another opinion is required.
• the prediction of the machine is poisoned and that of the doctor is discordant. in this case the diagnosis of the doctor is confirmed.
• the prediction of the machine is poisoned and that of the doctor is in agreement. the case is highlighted and another opinion is required.

Such “another opinion” could be by a doctor or it could be a second machine, with different algorithm and training, absolutely non poisoned.

This is a simple example of how the concept of redress by design could be implemented with prediction poisoning.

A single  doctor is a “one opinion system”. We invented the idea of science and conscience and all audits on correct behaviors, requiring explanations, to build incentives to avoid negligence and guilt.

A doctor with a machine is a “two opinions system” where (I believe) the machine’s opinion, as argued above, would tend to prevail. Lab technicians can already confirm this tendency. Furthermore, inspections and audits might not be effective on systems that cannot explain their motivations (like humans do).

The redress idea exposed above is that when the two opinions (machine and doctor) are in disagreement, a third one needs to be involved to resolve the dispute

The act of “poisoning” only serves the purpose of keeping doctors attentive, so that they avoid just conforming to the machine thereby being able to spot some incorrect predictions. It is a mean to improve effectiveness in  the detection of incorrect predictions.

Someone could object that such a procedure would eliminate the benefit of speed and scale increase allowed by the application of AI. As a matter of fact, since the physician knows that his predictions are subject to an AI-assisted examination procedure, he can operate at a higher speed because his activity is augmented with an AI based error control system. In this case a desirable result is obtained: the functions of the doctor are augmented by the AI and not replaced.

Arguably, such a procedure will be costly. In some states some variations could be envisioned, for example by involving patients in the decision. Let’s say the machine’s prediction is 0.65 confident. Before going for a third opinion, the doctor and the patient could have a conversation about the costs of listening or not listening to the machine and about any reasons the doctor thinks the ML might be wrong and let the patient decide after having received all due informations.

One final comment about the “poisoning” word I’ve chosen to express this concept. It may appear somewhat ungraceful (even more when used in an example about healthcare) but I think the term is quite appropriate as the term is already used with a comparable meaning in network security (ARP poisoning, DNS cache poisoning) .

Share now →

-(A sunday morning provocation)

Artificial Intelligence is a name born in 1956. It has gained a lot of attention thanks to its resonance with humans, and favoured the devlopment of an imaginative Hollywood production stream.

The name “artificial intelligence” has an implicit bias that does not allow for a cognitive perception adherent to reality.

On the contrary, the name favours the suggestion of the possibility of machines to develop some form of consciouness, emotions, acquire a “personality” similar to humans’ and, ultimately overcome human limitations and developing a self superior to humans.

You’ve seen the movies, you know the narrative… But they are only devices that extract correlations from data and use those correlations to make predictions and a load of very useful things. And as having calculators doing square roots, they can do it at a scale and speed far exceeding human’s performances. By throwing in some logic and randomness they can also exhibit some interesting and original behaviors. Yet machines have no clue of what reality is. At best, they mimic a model of the reality, so they are two steps away from reality.

After a conference on AI at the Pontifical Academy Of Science in Rome, discussing with some friends (among them Aimee van Wynsberghe), we argued that the first and foremost AI bias is its name. It induces analogies that have limited adhrence to reality and it generates infinite speculations (some of them causing excessive expectations and fears).

Because of this misconception, we proposed we should drop the usage of the term “Artificial Intelligence” and adopt a more appropriate and scoped-limited terminology for these technologies which better describe what these technologies are:  Systematic Approaches to Learning Algorithms and Machine Inferences.

Now we have redefined the name, will we still support the idea that SALAMI will develop some form of consciouness ?

Will SALAMI have emotions ?

Can SALAMI acquire a “personality” similar to humans’ ?

Will SALAMI ultimately overcome human limitations and develop a self superior to humans ?

Can you possibly fall in love with a SALAMI ?

Can we suddenly perceive a sense of how all these far flung (unrealistic) predictions look somewhat ridiculous ?

Share now →

[draft]

We recently had a debate in Italy about anonimity on social networks and the possibility/difficulty/impossibility to prosecute possible online crimes.

The whole debate started when a politician proposed to impose registration to social networks with personal IDs. This is a recurrent proposal. In every legislature, at least since the early 2000 some member of the parliament makes a similar proposal, there’s a public debate (now on twitter..) and then everything vanishes after a couple of days.

The proposal is both technically impractical and politically unsettling, given it exposes commenting persons to possible retaliations for their opinions determining a chilling effect on free speech. On the other hand, there’s already a chilling effect due to the huge amount of bot accounts that harass online conversions, or reputation damaging lies. (let alone other crimes like sexual abuse or hatred).

IMHO, the discussion fails to grasp an important point: we _already_ are reacting putting political pressure and legal oblilgation on intermediaries to police online speech. (I wrote about this in a book Parole e Potere (Words and power) cowritten with Oreste Pollicino, full professor of Constitutional law at Bocconi University and Giovanni Pitruzzella, Advocate General at the European Court of Justice).

Facebook has recently reported data about their policing. The figures are staggering, as they amount to many billions of interventions in 2019.

And they do it mostly proactively (e.g. not upon notification from users, they filter in near realtime upon content upload). For example, in the category “Child & nudity”, 99,5% of removals were proactive and just 0,03% of contents have been reinstated after a remediation request by users. The same goes for “Violence and graphics”: 98,9% proactive removals, 0,007% redress; “hate speech” 67,8%proactive, 0,05% redress; fake accounts 5,4Bn removals, 99,8% proactive, no data on redress.

Perhaps Twitter is not doing as much as Facebook is doing, maybe because of the level of required investment, but the “effectiveness” of Facebook actions is driving up pressure on everyone and eventually Twitter is going to end up with similar approach.

De facto, there are already private companies that, using Artificial Intelligence, decide who gets to see what, and who can say what, with very limited (or almost non existent) redress possibility, transparency and accountability.

This _already_ is the case.

And censorship will extend over time as Artificial Intelligence proliferates.

There are two measures we should look at, when working with probabilistic systems like machine learning: precision (of all those identified as wrongdoers, how many really were wrongdoers ?) and recall (of all the cases examined, how many of the wrongdoers has been identified ?)

AI works, for what a policy maker is concerned, as recall can be very high. Like it or not (I don’t), this will lead to an increase of requests of content policing by online intermediaries.

But we should care of precision, avoiding false positive to the maximum extent. Because for the specific person that has been denied one of her rights because of an incorrect prediction of a non defective system, her damage is unitary. A 100% violation of her rights. This is why, IMHO, we should be really concerned about redress, combined with a protection of anonimity to preserve user’s rights and prevent chilling effects.

I think the censorship power should not reside in private hands (unmonitored).

Borrowing some words from the US vs. Columbia steel antitust decision of 1948 (we are not talking about steel, but about communication control..):

We have here the problem of bigness.

Its lesson should by now have been burned into our memory by Brandeis. The Curse of Bigness shows how size can become a menace — both industrial and social.

It can be an industrial menace because it creates gross inequalities against existing or putative competitors.

…
In final analysis, size in steel is the measure of the power of a handful of men over our economy.

That power can be utilized with lightning speed. It can be benign, or it can be dangerous.

The philosophy of the Sherman Act is that it should not exist.

For all power tends to develop into a government in itself. Power that controls the economy should be in the hands of elected representatives of the people, not in the hands of an industrial oligarchy.

Industrial power should be decentralized.

It should be scattered into many hands, so that the fortunes of the people will not be dependent on the whim or caprice, the political prejudices, the emotional stability of a few self-appointed men.

The fact that they are not vicious men, but respectable and social-minded, is irrelevant.

That is the philosophy and the command of the Sherman Act. It is founded on a theory of hostility to the concentration in private hands of power so great that only a government of the people should have it.”

In this case, power should not reside in elected officials, but in justice systems.

This is why I proposed in the past legislature in Italy a bill (italian) on protected anonimity (as defined in a resolution that I proposed and was approved with unanimity, based on the final Declaration of Internet rights [Art. 10] by a Chamber of deputies study commission) and am pushing the concept of redress by design which has also been included in the policy and investment recommendations to the European Commission by the High level expert group on Artificial Intelligence of which I am a member (AI HLEG).

I already wrote about the idea of redress by design in this blog. The way how it was worded in the AI HLEG Policy recommendations 2019 is “establishing – from the design phase – mechanisms to ensure alternative systems and procedures with an adequate level of human oversight (human in the loop, on the loop or in command approach) to be able to effectively detect, audit, and rectify incorrect decisions taken by a “perfectly” functioning system, for those situations where the AI system’s decisions significantly affects individuals.”

The AI HLEG Guidelines explicitly state: Oversight may be achieved through governance mechanisms such as a human-in-the-loop (HITL), human-on-the-loop (HOTL), or human-in-command (HIC) approach. This can include the decision not to use an AI system in a particular situation, to establish levels of human discretion during the use of the system, or to ensure the ability to override a decision made by a system. Moreover, it must be ensured that public enforcers have the ability to exercise oversight in line with their mandate.

My proposal of protected anonimity in conjunction with a redress request for an invalid censorship decision is aimed to have courts taking a final decision, with al guarantees of a due process. This can be achieved, on a procedure triggered by the user,  by associating the specific censored content (not all of the users’ online activity) to a token that can be reconciled by the court to the user’s identity, thanks to the cooperation of a trusted third party (anywhere in Europe, relying on the existing eIDAS trust framework), that furthermore has the obligation to store the association offline, in order to prevent privacy violations and to permanently delete such association after a specific amount of time (in line to the expiration terms for the alleged violation’s prosecutability).

It may sound complicated, but it can be made as simple as logging into a wifi; IMHO it relieves the intermediary of possible responsibility consequences, it ensures all citizen’s safeguards are in place, it increases the reliability of the system by relying on a europewide exixting network of fiduciaries, it dramatically mitigates the risk of unjust decisions and, last but not lest, it can create a service industry for European companies coherent with European values.

Share now →