Machine Learning: What Should You Be Aware Of – Legally Speaking
Legal set-up and emerging risks of today’s automatic learning
What is Machine Learning?
The very first definition of Artificial Intelligence (AI) was coined by John McCarthy in 1956 as “the science and engineering of making intelligent machines”.
This term has evolved considerably over the past decades but kept itself close to the original meaning. Nowadays, we can define AI as the simulation of intelligent process and behaviors of the human being.
In this regard, AI can be catalogued in consistence with the simulated processes: rational thinking, learning, physical response, emotional response, etc. However, commercial practice has been using AI, Deep Learning and Machine Learning as if they were identical, even though they are actually not.
Indeed, when we talk about AI, Deep Learning or Machine Learning, we are talking about a genus-species relationship.
Deep Learning is a subtype of Machine Learning, as Machine Learning is a subtype of AI. While AI encompass all human intelligence process simulations, Machine Learning is an application from AI that provides systems with the ability to automatically learn and improve from experience without being explicitly programmed. Deep Learning, on the other hand, is a subset of Machine Learning that has networks capable of learning unsupervised from data that is unstructured or unlabeled.
What makes Machine Learning so popular?
In 2019 February the WIPO published informes relativos a la evolución de la Inteligencia Artificial por parte de la OMPI, according to which Machine Learning was the most popular technique. But, why?
The key is how Machine Learning offers great benefits to all aspects of the industry, from programming till profit improvement.
For instance, in 2017 Google translator reduced its code lines from half a million to only 500 lines using a Machine Learning system. This means that the software developers and programmers didn’t need to manually enter a large and complex code that foresees all possible situations, and at the same time the software was optimized.
At an industry level, Machine Learning helps with the relentless profitability search. Since profitability goes through a continuous analysis of data, decisions making and benefit response, and Machine Learning automate this process, the popularity increase is self-explanatory. Also, that’s why digital twin and predictive simulation are the most popular Machine Learning techniques.
These two techniques are especially useful for manufacturers. They allow the prediction and safe study of decision impacts in the business in the interest of profitability improvement.
Likewise, Machine Learning is more popular than Deep Learning because of the unsupervised nature of the Deep Learning techniques.
Basically, trust between humans and AI systems is starting to grow in manufacturing industry, but unsupervised AI techniques have been revealed the most vulnerable as regards civil rights violation in industries focused on individual data collection and treatment.
Which is the regulatory environment of AI?
The answer is simple, there is no regulatory framework for AI matters. Neither Spain nor any other country, nor even international institutions, haven’t adopted specific regulations in this topic. There’s only search activities carried out and reports issued related to ethical questions about AI systems.
On October 2016, the British House of Commons published a report on Robotics and Artificial Intelligence, describing a set of ethical conducts expected from all software developers in AI matters. The next step came from EU, with the Draft Ethics Guidelines for Trustworthy AIreleased by the European Commission's High-Level Expert Group on Artificial Intelligence on December 2018. Those reports are characterized by recommendations tips and only listed a few binding conducts.
In USA there is no specific policy proposal related to AI in any federal level, from an ethic or regulatory perspective. But it has to be nevertheless recognized that the very fragmented legislative system in the said country must be scrutinized on a case-by-case basis, due to the legislation produced at state level or concerning guidance for specific industries dealing with AI matters.
At present, the most advanced legal system might be Canada. The Treasury Board Secretariat of Canada started a search program concerning the responsible use of AI in government programs and services offer, and in March 2019 it released the Directive on Automated Decision-Making. This represents the very first regulation about the legal regime of AI, establishing even a compliance and liability system for Machine Learning issues in government programs and services.
Aside from that, most of the countries and international organizations have ruled about individual data protection. Examples can be found in the EU General Data Protection Regulation, the Personal Information Protection and Electronic Documents Act in Canada, and the multiple regulations about it the USA..
Which emerging risks should you be aware of with AI?
There’s no need for a meticulous regulation at this moment. Indeed, usually an emerging industry or technology grow up faster unregulated. But excessive lack of regulation implies legal uncertainty, and this increase legal costs for software developers. Right now, we have to analyze on a case-by-case basis the consequences of a malfunction in a Machine Learning system and protect the parties through the use of contractual provisions.
The specific risks you should be aware of with AI can be classified as follows:
It is a forbidden practice and it’s strongly regulated in legal systems, like USA’s and EU’s. However, AI development has created a new collusion type, the algorithmic collusion.
The technology revolution hinders the prosecution of cartel behavior because regulation isn’t suitable enough. We can say competitors' agreement are illegal but we can’t safely detect when it happens, and that might imply cases of inappropriate bans. At present, you should pay attention to suspicious conducts like competitors using the same algorithm or going to the same algorithm provider.
Discrimination is the most typical issue within Machine Learning, and it can happen unintentionally. To avoid this problem, sometimes a Machine Learning system requires manually code introduction, in order to reconcile some natural functioning with civil rights.
A biased database generates a biased system, and as consequence some groups and individuals can be discriminated against. This discrimination implies legal liabilities. And since a database can be naturally biased the manual introductions of behavioral patterns is a priority.
Data protection is highly regulated in the EU and US, and in other countries, therefore with regard to privacy, is all about regulatory compliance. Still, access rights can be dissolved due to the Machine Learning system complexity.
Supervision is the key to manage Machine Learning’ privacy risks, as well as security and access protocols.
Also, it must be considered the prohibitions set out in Article 22 of the EU GDPR, according to which automatic decisions based on personal data are generally forbidden.
The basic rule about product liability is that a manufacturer will be responsible for the damages and losses caused as result of a faulty product.
However, misuse is a reality within the Machine Learning system, thus the manufacturer or supplier are obliged to provide training.
The lack of specific legislation in addition to AI systems typical risks turn Machine Learning in an issue to address notably through contractual negotiations. Some of the typical terms to be included are unexpected and malfunctioning, liability, algorithm rights, and developer guarantees.
How Can We Help You?
One of our top priorities within the technology-based IA services is the role of developers, for whom at Gowper we offer a set of specialist counsel aimed at covering every single legal they might have, such as:
- Regulatory compliance.
- Legal study and report drafting about personal data susceptible of discrimination.
- Diseño de modelos de cumplimiento normativo en materia de datos personales.
- Negotiation and drafting of contracts for the creation and commercialization of Machine Learning systems.
- Defense and representation on disciplinaries proceedings in Antitrust matters.
Learn more about our offer of Individualized Solutions Plans (ISPs), in particular about our Orange Solutions, or about our wide range of sophisticated Services & Industries designed for helping business like yours thrive.