Machine learning

Machine Learning pour les simples humains…

machine-learning

Début Décembre, Nicolas Courtier m’a invitée à présenter devant un parterre d’avocat-e-s et de juristes plus ou moins averti-e-s les bases de l’intelligence artificielle. Ce colloque “Droit et Numérique” avait pour vocation plus large de faire état de la loi et des questionnements sur les disruptions technologiques telles que smart city, big data, machine learning et blockchain. Je ne peux que vous encourage à aller consulter le storify issu de cette journée riche en partage de compétence ou encore de parcourir mon court article sur quelques unes des questions évoquées (in english). Mais je partage également avec vous, cher lecteur assidu, la session que j’ai animée.

Machine learning, kézako. Le machine learning est un sous-catégorie de l’intelligence artificielle, ce qui ne rend pas le sujet moins intéressant. Cette discipline consiste à prédire le comportement d’un système, d’un humain, à partir d’un modèle – plus ou moins précis. Il s’agit donc de s’appuyer sur le passé pour prédire le futur. Oui, enterrons tout de suite notre fantasme d’une machine apprenante, en auto gestion, et qui règlerait les peines du monde à notre place. Parceque le machine learnign se nourrit du passé, précisément, le machine learning nécessite d’avoir une très grande quantité de données, de très bonne qualité. Et oui, le machine learning exige la quantité *et* la qualité. Pour en finir avec les grands principes, afin de déployer une solution à base de machine learning, il vous faut donc, du logiciel qui décrira le modèle de votre système, des data et un data scientist qui affinera le modèle.

Les promesses du machine learning. Elles sont de différents ordres, selon les cas d’usage, mais on peut songer à dupliquer la compétence humaine (comprendre, dupliquer un expert qui a fait de longues études, comme un médecin, un avocat, un travailleur dont la valeur repose sur la connaissance factuelle), on peut également songer à faire mieux que l’humain (en allant chercher des corrélations entre des événements auxquelles le cerveau humain n’aurait pas songé). Dans le domaine des services, c’est la même chose, le machine learning peut améliorer un service (par exemple une recherche par mot clé, ou une recommandation de produits dans un catalogue) ou encore en créer de nouveaux (une gestion intelligente des trolls sur Twitter par exemple).

Les applications gourmandes de machine learning. Potentiellement toutes. Mais les premiers acheteurs de la technologie sont le marketing (la prospection, le contenu sur mesure, le service après vente, le chat automatique…), les services de recommendation/search/match, le monde de la sécurité (pour la prédictions des risques, des fraudes). Tous les espoirs et fantasmes sont permis sur les domaines de la santé (la quêtes de l’immortalité et des diagnostics, meilleurs et au bon moment) et les voitures intelligentes …

Le machine learning, une science qui s’invente. Comme toutes les autres disciplines qui émergent et sont projetées dans notre monde, le machine learning est soumis aux courants hype tels que l’open source, le crowdfunding, le cloud, la création de communauté… La multiplication des initiatives autours du machine learning pour le rendre plus efficace, plus lisible, plus accessible, sont autant d’opportunité d’enrichir cette science.

Bref. Le machine learning est une technologie nouvellement sous les spots de l’actualité, comme blockchain, big data l’ont été ces derniers mois. C’est également une nouvelle opportunité d’agiter les modèles qui décrivent et conduisent notre monde. C’est l’introduction d’un usage encore plus intensif de la données, la fameuse richesse de notre siècle, et c’est également l’acceptation de plus de prédictif, d’à peu près, dans un monde complexe. Une tendance et des usages à observer de près donc…

Et les slides ? Machine learning pour les simples humains, version slide, c’est par ici. Enjoy !

Law and digital disruptions, examples of machine learning and smart city

law

As part of the amazing opportunities I get with my job, I have been invited to a one-day workshop, organized by the AFDIT, the french association of lawyers, specialized in IT and computing systems  (part of the International Federation of Computer Law Association IFCLA). This day aimed to have lawyers discussing the impact of technology on the laws, in public area or business area. The perimeter of the discussion was europe and US, thus some speakers from all around the world came and shared their experience. In order to educate and progress on major 2016 topics, the organizers, Nicolas Courtier and Yves Léon, selected the themes of the day as : smart cities, artificial intelligence and blockchain. Here are some interesting elements that were raised all along the day.

Smart city, what does it means ? We all heard about smart city : it is the promise to improve town management, population mobility, citizen service offer, by connecting all possible pieces of information and building some tailor-made services. That is the vision that some local politicians promoted during the day (Caroline Pozmentier, Stéphane Paoli), together with some French Tech actors. The other way to see it, explained by Art Langer from Columbia University is to position citizen in the middle of the town dynamic. By offering him or her better mobility, frictionless social relation, great work opportunities, better democracy, (which is great news for the humans). That vision suggests a potential coming race, among towns, to become the most attractive town in the world, in order to maintain growth in economics and population. All those improved services will be based on large data collection operations, or interconnection of databases. In order to do so, service providers may be required to have private and public actors collaborating, canvassing the city, the citizen and grab appropriate information. And then came the question of the privacy, which might be one of the most challenging questions in that model where the consumer is a citizen.

Big data privacy challenge in smart city. The relevance of the smart city services are relying on the consolidation of a set of data, for which confidentiality and anonymity are hard to garantee. In addition, this mixing of data set triggers the question of ownership and liability. Who would be owning the data and would be responsible for the the failure of data maintenance ? That question would any way have to be answered with the coming european regulation on privacy. As Massimo Attoresi from explained, this regulation mandates that all actors of a service handling data (collecting, processing, storing or destroying) have to take care of the data, by having clear process for user opt-in, transparency in usage, fairness in collection, data minimization (the less you take, the best it is), storage limitation, integrity and confidentiality and inform the user about potential leaks or incident. How to explain a clear purpose of data retention, when you don’t know which service will come from your data collection ? How can you assess the risks, when you have a dynamic system, with cross-system responsability ? How can you garantee anonimity when so much information, including geo-localized ones are collected ? Interesting questions that smart cities will have to answer…

Smart city opportunities. In case citizen consider smart cities as life improvment, some ways to roll out smart city could come with great benefit for the society. Without ignoring potential threat to citizen privacy coming with smart city, Philippe Mouron drafted for us some positive aspects of it. The idea to integrate citizen into service design could be a great way to improve service relevance. In addition, the collection of data, and the fact that data belong to the citizen may accelerate the movment of open data. Philippe advocated also for a better mixing og legal and tech know how in the lifecycle of devices, in order to make sure, that all do see an interest in “the silence of the chips” (aka, users being in control for stopping data collection and leak towards to servers).

What about machine learning ? We discussed during that day the concept of machine learning. I reminded the audience its basic principles. You know. The fact that machine learnng is a sub-categoty of artificial intelligence, which consists in predicting the future (or the most probable one), based on past data. I listed the required skills and tools to roll out machine learning based services (aka, software, some good pieces of data, a smart scientist fine tuning your model). I reminded the audience the first use cases benefiting the machine learning, which are marketing, search and recommendation, security, health and smart cars. One of the main take away that I asked people to remind was the fact that we were switching from a determinist world (where each line of code is describing a possible situation, and where programs take well know roads), toward a world where we describe our environment with a model, with more or less errors and accuracy. Based, on that I took the opportunity to raise questions that machine learning triggers for me, such as privacy, liability and error management. And I got few answers from the other speakers.

What could be the legal impact of machine learning ? @rubin demonstrated how the machine learning could impact the legal business, replacing some assets of the lawers and potentially introducing a better undertanding of risk and gains around trials. Rubin also reminded that the law was not designed for robots, but for human and insuring fair interactions among humans, including in business situation, leveraging on technology. He gave some intersting perspectives on how to pave the way towards a mastered artificial intelligence deployment, based on few principles. Clear responsability, transparence of artificial intelligence in decision making (specially for the ones suffering the decision), efficient maintenance and regular audit of the artifial intelligence systems involved in services, and lastly, a permanent possibility to challenge the results of services based on artificial intelligence. Those principles based on good will and fair relation were good to hear and could be integrated in any strategy embedding machine learning, now.

My take away from that leal and tech workshop. Yes, definitely, mixing of perspectives and visions are key to have everyone progressing in understanding a transversal topic such as technoogy in society. And. The topic of ethic in software is definitely an additional item to add in our watch list, together with the privacy expectations.