Contact Tracing & Privacy: come risolvere il Trilemma

Per un Italiano che vive a 10,000Km di distanza, non è piacevole assistere al conflitto interno tra stampa, cittadini, governo e imprenditori, causato dalla (giusta) scelta di utilizzare “Immuni” come app di Contact Tracing. Oltre 4 milioni di risultati per una semplice ricerca “Immuni Privacy” su Google suonano come un’immensa onda di energia diretta nella direzione sbagliata, come dimostra il comportamento mostrato in un numero infinito di occasioni precedenti (app di incontri, condivisioni di foto, consenso all’accesso ai dati per giochi su Facebook, ecc.).

Tralasciando il piccolo sfogo iniziale, ho deciso di scrivere questo post per descrivere la nuova tecnologia chiamata “privacy-preserving computation”: si tratta di tecniche matematiche, studiate da anni ma rese utilizzabili nella vita reale solo recentemente, che permettono a un’applicazione di utilizzare i dati degli utenti senza però “vedere” i dati stessi. Proviamo a vedere come funziona:

Il problema: “Contact Tracing Trilemma”

Forzando leggermente l’esempio, ci si pone il problema di come risolvere il ‘Contact Tracing Trilemma’, cioè il compromesso che le tecnologie correntemente utilizzate impongono: la maggior parte dei modelli proposti non permette di avere sia il tracciamento della posizione, sia la tutela della privacy, sia l’analisi dei dati in tempo reale, ma consente di ottenere due di questi elementi in contemporanea (al massimo).

Seguendo i modelli ‘tradizionali’ di gestione dei dati, garantire la privacy vorrebbe dire limitare fortemente la condivisione della posizione o l’analisi dei dati in tempo reale; tracciare la posizione e garantire la privacy senza permettere l’analisi dei dati in tempo reale, renderebbe vano qualunque progetto. Una soluzione parziale per garantire un buon compromesso tra privacy e funzionalità potrebbe essere quella di avere più app allo stesso tempo (per esempio, una app diversa per ogni Regione) al fine di diminuire il possesso dei dati di una sola applicazione: tuttavia, come rivela la ricerca svolta dall’Italiano Luca Ferretti del Big Data Institute dell’Università di Oxford, almeno il 60% di una certa popolazione deve utilizzare la stessa app, affinché si arrivi ad uno stato di “epidemic control”.

Decentralized Privacy-Preserving Computation

Alice vuole sapere l’età media di tutti gli studenti presenti al suo corso privato di matematica. Alice ha due possibilità: 1) chiedere l’età ad ogni studente o 2) chiamare Roberto, il quale, chiedendo l’età ad ogni studente, riferisce solamente la media ad Alice.

Caso 1) Alice conosce l’età di ogni studente

Caso 2) Roberto conosce l’età di ogni studente, Alice conosce solamente la media. Roberto promette di non rivelare i dati a nessuno.

In entrambi i casi, gli studenti devono fidarsi del fatto che Alice e/o Roberto non utilizzeranno i loro dati per altri scopi in futuro: sarà però estremamente difficile per uno studente verificare che il dato non sia stato rivelato, diciamo, 5 anni dopo.

La Decentralized Privacy-Preserving Computation permette di risolvere questo problema: i dati vengono inviati ad una serie di scatole nere interconnesse e programmate in maniera tale da eseguire calcoli sui dati che vengono inseriti al loro interno, senza mai rivelare i dati stessi a colui/colei che richiede il risultato: il ‘Cloud sicuro, privato, e distribuito (decentralizzato)’ creato da Oasis Labs, la startup presso la quale lavoro a San Francisco, è un esempio di questa tecnologia.

Riprendendo l’esempio precedente, gli studenti potrebbero indipendentemente inviare il loro dato (la loro età) alla piattaforma, avendo la garanzia che nessuno (nemmeno Oasis Labs) possa avere accesso al dato ma solo al risultato del calcolo.

Contact Tracing Trilemma — Luca Cosentino

Seppur non del tutto corretto, l’esempio mostrato spiega come avviene il calcolo sui dati privati, o Privacy Preserving Computation. Cosa vuol dire “Decentralized”?

Se nessuno ha accesso al dato, bisogna trovare un modo per verificare che il risultato ottenuto sia corretto e che comportamenti maligni vengano bloccati: attraverso la distribuzione/decentralizzazione, il meccanismo di verifica è reso altamente disponibile, in modo che nessun malintenzionato possa nascondere le sue azioni illecite. In sostanza, la distribuzione elimina la necessità di fidarsi di ogni singolo ‘validatore del calcolo’, garantendo la confidenzialità del dato e la corretta esecuzione del calcolo.

In termini leggermente più tecnici, le chiavi per decriptare i dati non sono in possesso di individui ma solo di algoritmi autorizzati dagli utenti stessi solo per il fine stabilito: in aggiunta, tutto quello che avviene sulla piattaforma viene registrato in un database indelebile.

Come questo potrebbe aiutare con la situazione attuale dovuta al COVID?

Se sostituiamo l’età dell’esempio riportato sopra con il dato che traccia la nostra posizione, è abbastanza intuitivo capire come questa tecnica permette di risolvere il Trilemma mostrato sopra:

sarebbe quindi possibile garantire la privacy senza sacrificare la geolocalizzazione e la possibilità di analizzare i dati in tempo reale. L’utente avrebbe il completo controllo del proprio dato, la possibilità di decidere a chi darne accesso e di verificare l’effettivo utilizzo.

In aggiunta, gli utenti potrebbero decidere di dare accesso selettivo ai propri dati (sempre mantenendo il dato nascosto) a Governo, ricercatori/ricercatrici, medici, ecc, i quali potrebbero utilizzare questi dati per creare modelli predittivi.

Entrando più in dettaglio, una possibile implementazione del modello sarebbe la seguente:

• Quando Alice e Roberto si incontrano (in un certo raggio di distanza), i loro telefoni scambiano un ‘token’ tramite Bluetooth
• Roberto va a testarsi in un ospedale e risulta positivo; un codice criptato ma univoco viene associato al risultato del test e inviato dall’ospedale a un server (in realtà i server sono due per non permettere a nessuno di riconoscere l’utente e associare lo stato dell’infezione)
• Il telefono di Alice controlla se i token delle persone che ha incontrato esistono su questo server che contiene la lista dei token ‘infetti’
• Trovando quindi il token di Roberto, il telefono di Alice avverte Alice stessa che potrebbe essere stata esposta al virus

Il sistema Epione, sviluppato attraverso una collaborazione tra l’Università di Berkeley, la National University di Singapore, e Oasis Labs, propone il modello appena descritto.

Un ulteriore vantaggio del modello Epione rispetto ad alcuni dei modelli esistenti è la prevenzione contro le dichiarazioni di falsi-positivi: un utente malintenzionato potrebbe infatti falsamente dichiarare di essere stato testato positivo. Ciò diffonderebbe false informazioni e spaventerebbe altri utenti e ridurrebbe la fiducia nel sistema.

Conclusione

Spero che questo post, per quanto pieno di inesattezze tecniche volute al fine di semplificare il messaggio, contribuirà all’evoluzione del prodotto . La speranza più grande è che la continua lotta tra le parti finisca nel migliore dei modi per favorire una ripartenza veloce e sostenibile.

Un grande in bocca al lupo a Bending Spoons, il vostro talento ed il vostro successo sono un bene prezioso per il nostro Paese.

Luca Cosentino
luca.cosentino@berkeley.edu

Data Scientist to Product Manager? It’s not a common path, they say

Why data science makes great product managers.

You spent a few years crunching data, analyzing information, going deep on customers’ behavior online, offline, omnichannel, mobile, in store, and every possible intersection of those. You are constantly under pressure, deadlines seem to run fast and in your opposite direction. Your Python and R are full of packages and libraries, and your folders contain plenty of v1, v2_lc, …, v11_final_lc (thanks Google for solving that, btw).
Your shoulder hurts for the many pats you were given for your valuable contribution to the team, you have just won a $20 movie card with free pop-corn or even an Amazon voucher.
You finally realize that, while you have acquired a unique skillset, you have learned how to navigate complexity, and you have trained your thought process, you have never been in control.

You start your research, talk to people, and read a bunch of articles, until you realize that your background is incredibly appropriate for a career you had never thought about before: Product Management.

In fact, Data Science makes great Product Managers.

Let’s see why:

Data Scientists and Product Manager use data to inform their decisions.

Data Scientists’ bread and butter is data; they analyze large quantities of information, synthesize it in a few key points, and suggest decisions based on their findings.

Product Managers are obsessed with shipping features their users love. Users’ needs and feature development are based on surveys, research, tests, data.

Data Scientists and Product Managers present their findings to their stakeholders and seek consensus.

Data Scientists present their findings to their audience; regardless whether they work for internal or external clients, they always have to sell their story to their audience and, guess what, they use data to do so.

Product Managers, by definition, coordinate multiple stakeholders at the same time and won’t be successful unless they are able to convince everyone. Data is PM’s friend.

Data Scientists and Product Managers work cross-functionally.

Data Scientists work with and for a number of teams within the organization; they may supply information to clients, client managers, product managers, finance team, and strategy group.

Product Managers work with designers, engineers, market researchers, finance, and product leaders. Everyone generally knows what everyone else is doing and who is the point of contact for specific topics.

Data Scientists and Product Managers see success as a team achievement.

Data Scientists may do most of their work independently. They are often given a task and they execute on it. Successful Data Scientists, however, need the big picture and this often requires creating relationships with the broad team. No matter what, every achievement will never be the result of their sole work.

Product Managers may have the best intuition or create the best mockup; however, success depends on execution and execution requires the entire team to be involved in the process.

Data Scientists and Product Managers have to prioritize.

Data Scientists are overwhelmed with an infinite amount of data; their analyses can take multiple directions and they can always go one level deeper. Hence they are constantly forced to prioritize.

Product Managers are overwhelmed with options; the user-centric approach always leads to a number of different options and the only way to succeed is to focus on what transforms into the impact they are seeking.

Data Scientists and Product Managers need to know the market they work on.

Data Scientists may know every analysis technique but they always have to explain their findings in the context of the market they work on while keeping their eyes open to spot important details.

Product Managers need to know everything about users, competitors, potential entrants, customer journey, and industry-specific dynamics. They have to put themselves in users’ shoes, while maintaining a fresh and unbiased perspective.

Data Science is a phenomenal school for aspiring Product Managers.

Bring this argument to the table next time someone you are seeking advice from tells you “I don’t know anyone who transitioned from Data Science to Product Management, it’s not a common path”.

Machine Learning: where to start from

The particular application of Machine Learning to business is becoming more and more popular. I guess what really convinced me to spend time on Machine Learning is exactly this fascinating intersection between technology, algorithms, statistics, and business: very rarely, an innovation has found ground in business as quickly as ML.

This article is meant to be a one stop for beginners in the field: instead of reinventing the wheel, I tried to collect resources I found useful on my path so far.

Some preliminary readings:

How to get started

Top ML Algorithms — must know

Stay informed: website and newsletters

Wanna start something yourself?

In Conclusion…

There is no doubt that ML is the present and the future of many industries. In the next months, I will be analyzing the impact that ML can have on traditional businesses: specifically, I’d like to investigate how traditional companies can compete with more modern entities that are born leveraging ML and data from day 0 (ie, Amazon vs traditional retailers). If you’d be interested in participating/supporting, please do not hesitate to get in touch!

Stay Simple — the 3Bs framework

The relationship report-manager is never obvious. 1:1 meetings generally create more questions than the one they answer, and it’s rarely clear what the right level of depth is.

At Procter&Gamble, I learned something really useful to conduct productive meetings, avoid micromanagement, and win as a team: the 3B’s framework.

My team decided that every weekly report-manager or skip level meeting, at every level of the organization, had to be structured in 3 points:

What’s Big:
does it really make any difference to focus on small things that are not going to make any difference? Even more important, are they worth the time of 2 (or more) people? Focus on something impactful, something big!

What’s Broken:
is there any problem, something that should be fixed? Let’s discuss it with someone above you; you will (most likely) be able to repair faster.

What’s Breakthrough:
how could we do something completely new, that can really improve our competitive positioning in the market?

I advise to use this scheme in both business and personal life; is pretty simple and makes you save a lot of time and effort; rarely we read about something we can action tomorrow.

Meeting Customers’ Expectations

How can you exceed customer expectations? First you need to know what their expectations are. When it comes to expecting certain performance and functionality from your product or service, you need to make sure you can deliver on the “speeds and feeds” that you sold your customers. However, there are more basic expectations that are not related to your product or service that need to be met, or even exceeded. More deals are lost because, in spite of how great your product may be, the customer’s basic needs were not met.

So, what are these customer expectations? What do customers really want? Here are 7 things that customers want.

1. To be respected — Don’t treat me like a fool. Don’t act like I’m stupid. Don’t discount what I have to say. I want to be treated with respect.

2. Fair price for a quality product — When it comes right down to it, doesn’t it make plain sense that customers should receive a quality product or service for a fair price? Of course it does. But why does this become so difficult at times. I paid good money for your product and it’s not performing the way you advertised it.

3. Fairness — I also want to be treated fairly. That doesn’t mean I’ll always get what I want. Sometimes what I want is not possible, feasible or reasonable.

4. Empathy — When I need help in purchasing an item or when I call a company because of a problem with their product, I want them to show a little empathy, or understanding. I want them to put themselves in my situation so they can understand my situation — walk in my shoes.

5. Information — I can’t make an intelligent decision unless I have the facts and information in order to do so.

6. Someone to understand my needs — This is a close kin to empathy. It’s a step further, though, since it means that I want someone to own my problem. If I ask a sales person for help, I don’t want to be dumped on someone else.

7. Someone I can trust — When you say you’ll get back to me, you’d better get back to me. When you say you’ll handle my request, it better be handled. When you say your product will do something, it better do it. Otherwise, I will lose trust in you. When you start building trust with your customers, then you are also building respect.

The importance of Mentorship and how to get the most out of it

Mentorship has always been discussed as a controversial topic. Even though everyone acknowledges its importance, very few understand its real essence and take the right actions.

The business world has become and is becoming more and more confusing. The traditional paths are not paying off as they were in the past and new jobs are entering the market every day. In such a fast context, the importance of mentorship remains the same; it just changes in the way it works. Here are my five advices on Mentorship:

1) Mentorship is not only asking for advice. It’s about establishing a long lasting relationship.

Too often people reach out to establish a contact and get a one-off advice, for example, when it comes to a new job offer. Despite the fact that it may still be useful, it is not the best way to get the most out of it. Having a Mentor means having someone to have regular conversations with, someone who knows you time after time and who is in the position of contextualizing the advice you are asking for.

2) Your motivation is what Mentors’ are looking for.

This slightly depends on the type of people you are speaking with. In my experience it’s not difficult to find mentors, even when you are not directly related to that person. People are happy to give (back, if possible) and their satisfaction is to see determination, motivation and results in the person they are helping. Be angry and grateful, this is all Mentors look for.

3) You shouldn’t see having a Mentor as a fast-track.

You have to be honest; if you are asking a senior person to become your Mentor, you have to believe that his mentorship will benefit you and you have to be sure you will learn. Of course it’s your chance to be visible and make others’ know you better, but your conviction should come first. Don’t be sneaky, it won’t bring you anywhere.

4) Ask, ask and ask.

Business evolves fast. A lot of today’s most common jobs didn’t exist 10 years ago. In this uncertainty, the only thing you can do is ask to learn and be curious. Your Mentor is a resource of information, but you won’t understand the importance of all of them during the conversation. Wait a bit, the knowledge you acquire will be relevant at some point.

5) Don’t trust everything you are told.

I learnt that it’s not a matter of good or bad advice, but mostly a matter of how that advice applies to you. It’s very likely that you receive two opposite advices from two mentors — no worries, it’s normal. Instead of taking the advice as it is, try to understand the vision your mentor has and how that advice helps you align to that vision. Then take your decision.

The Brexit Dilemma. Who’s the prisoner?

51.9% Leave doesn’t tell the truth. Neither do words from European and UK leaders. The irrational Brexit puts both Europe and UK in a prisoner’s position. Stay or Leave, this is the Dilemma. Even after the Referendum.

I moved to the UK, to London in particular, a bit over a year ago. I left Italy because I thought an international experience would enrich me both personally and professionally. I felt that I was losing confidence in the future and didn’t reflect myself in how the economical, political and cultural instabilities have transformed Italians, a population that I will always admire for its strong values and ability to adapt.

Being in the UK today represents something that I won’t probably experience again for the rest of my life. The most stable country in Europe, whose GBP has grown consistently from 1982 onwards (2009 and 2010 excluded), whose unemployment rate is one of the lowest in Europe, whose currency was exchanged at 40% more than the Euro, whose rating was the highest in the region, suddenly calls for a thoughtless referendum and votes to Leave the EU. It sounds irrational.

Suppose two friends, Europe and the UK, are suspected of committing a crime and are interrogated in separate rooms. They are rational human beings so they both want to minimise their penalty. What is called the prisoner’s dilemma offers multiple scenarios: if only one of them betrays the other, who confesses will be free and who keeps silent will be jailed for 10 years. In case both of them keep silent, they will serve only 1 year in prison. If both confess, both Europe and the UK will be given 5 years of prison. In case of asymmetric information, in absence of communication between the entities, this scenario will always lead to a non efficient situation in which both parties choose to protect themselves at the expense of the other participant. As a result of following a purely logical thought process to help oneself, both participants find themselves in a worse state than if they had cooperated with each other in the decision-making process.

Luckily, Europe and UK are not in two different rooms and have the chance to communicate. But unfortunately they are not in a zero-sum game. If one becomes weaker, the other is likely to become weaker as well and it’s not clear who would benefit from this conflict. This unpromising scenario suggests that Brexit is not the only way, as it’s not an efficient one. The well said and formal words that politicians all over Europe have been communicating following the referendum don’t match mathematical and economical principles on the base of which most of business and political decisions are taken. And despite Prime Minister Cameron and London’s Mayor Sadiq Khan both said that people clearly expressed their choice, a majority of 1.9% doesn’t seem big enough to justify such a big decision, in which not motivated by plan and at least partially based on false information: the £350m pledge to fund the NHS has been declared a mistake by Farage, the higher control of the borders is not likely to happen as Britain wants to access the single market. 17.4m people strong “Leave” was largely based on these two points.

It’s no surprise then that the initial requests by Juncker about UK leaving the EU in the next three months has quickly been calmed down by leaders of Germany, France and Italy who decided against a rapid start for the negotiations. Similar case for JP Morgan Jamie Dimon who first announced that the bank would have moved 16 thousand people in case of Brexit and yesterday said that the bank will maintain a huge presence in London, Bournemouth and Scotland. Again, they are simply rationale.

Therefore, after the first crazy days in which markets panicked much more than expected, there is the general feeling that leaders all over the world have sat down to think strategically about the possible next steps. Sadiq Khan has just named Rajesh Agrawal deputy mayor to protect the City after Brexit. The self-made multi-millionaire who grew up in poverty in India before moving to the UK and making his fortune is the first move of Mr Khan to keep his promise to be the “most pro-business mayor yet”. On the other hand, Boris Johnson has decided to not run for Prime Minister and Nigel Farage is stepping down as leader of the UKIP party. The two main supporters of the Leave campaign have dragged UK down into mess and are now expecting others to pick up the pieces.

Despite markets have slightly calmed down, the equilibrium is far to be reached. Another storm is yet to come. The only hope is that troublemaker leaders will adapt a rational behaviour instead of making it their own battle or their chance of visibility. Farage’s last claim “My political ambition has been achieved” suggests he acted for a battle, not for a vision. Neither Europe nor the UK need heroes, they need not to be prisoners. People don’t need to watch videos of those political leaders screaming in Parliament and insulting each other; people don’t need episodes of racism masked as political discussions. Whatever politicians decide, we will always be citizens of Europe. Democracy is clearly the base of our society, but 51.9% of majority under false information doesn’t lead to a rational and democratic equilibrium. I believe Farage and Leave won the Referendum battle but will lose the Brexit war, in the name of rationality, in the name of the equilibrium.