As part of our firms’ innovative character, three focus areas have been created, one of them being “Artificial Intelligence and Digitalisation”. Aside from growing our core practice, we strongly believe that our clients will have a growing need for competent legal advice in this area, in particular as an increasing number of companies are using Artificial Intelligence (AI), be it to classify and label documents, translate emails, make recruitment decisions, or recommend what to buy or watch next.
Aside from the question of how to responsibly implement such technology in businesses, we first need to understand what AI is. As part of our AI series and to help you tackle your next AI project, we have drafted a list of terms that may prove helpful to you.
Artificial intelligence or machine learning sounds like something slightly intimidating from the future, which is not (yet) part of our daily lives, but this is not true.
Does the sentence “other people have been buying…” or “you might also be interested in…” sound familiar? These are phrases we often encounter in online shops suggesting additional or other products. Such platforms are only able to show you relevant products by using an algorithm that selects such relevant products, often using your own and other customers’ data, including purchasing histories (big data and analytics). This is a form of AI which we encounter on a daily basis, without actively taking note of it. Other similar applications of AI can be found in smartphones, smart home devices or even cars.
To help you better understand AI including potential legal aspects relating to it, we have prepared the following list of terms from A to Z.
A. ARTIFICIAL INTELLIGENCE (AI)
Alan Turing defined AI as the “science and engineering of making intelligent machines, especially intelligent computer programs”.
AI is a set of algorithms that can be fed with both structured and unstructured data to complete a task without being explicitly programmed on how to do so. It can be considered as a subset of computer science that aims to build machines capable of doing human-like tasks such as making decisions in new situations, classifying previously unseen objects, recognising speech, and providing translations.
B. BIG DATA AND BLACK BOX
Big data is a large amount of structured and/or unstructured data that is too complex to be handled by standard data-processing software. The huge amount of data used in big data systems often leads to compliance questions, in particular concerning data protection laws.
For many AI algorithms, understanding how decisions were made is not easy. The unknown part of the decision-making is called a black box. This makes it difficult to apply AI in situations that require complete transparency, for example for highly-regulated processes.
A chatbot is a program that is used to run the messenger interface to recognise the request of a person and to correctly respond to it. To do so, a chatbot simulates a real conversation with the user through text or voice messages. State of the art chatbots use machine learning and AI to identify communication patterns. Through interactions with people, chatbots learn to imitate real conversations and to help them find answers, making them smarter and more personable after each conversation. However, providing the chatbots with incorrect or incomplete training data, or opening them up to public access can result in unexpected behaviour, so care is needed.
D. DATA MINING AND DEEP LEARNING
Data mining refers to the process of digging through large sets of data to identify previously unknown patterns and correlations to generate insights. This often requires a lot of high-quality data which need to be obtained in a legally compliant way. Special care needs to be taken when dealing with personal data, in particular health care data or other types of sensitive personal information.
Deep learning describes the neural networks and the algorithms that are used to generate patterns for the use of decision making. Neural networks are models that are inspired by biological neural networks constituting of neural layers. These types of systems usually “learn” to perform tasks by identifying abstract correlations and patterns in a large number of labelled examples, which is contrary to traditional programming where algorithms and rules need to be explicitly designed. Other types of deep learning systems are capable of unsupervised learning, which relates to finding patterns in unlabelled or unstructured data.
E. EMBODIED AI
Embodied AI puts AI into the context of robotics by combining AI software with sensory input. The aim is to improve the cognitive function of the AI, allowing it to better understand its situation and surroundings for more accurate data analysis and improved response processing.
F. FACIAL RECOGNITION
Facial recognition is a biometric software application capable of uniquely identifying or verifying a person by comparing and analysing patterns based on the person’s facial features. Facial recognition is mostly used for security purposes, thereby opening it up to potential liability if it has defects. This is an issue that needs to be addressed correctly in particular in software development contracts.
G. GENETIC ALGORITHM
A genetic algorithm is based on the principles of genetics to more efficiently and quickly find solutions to difficult problems, in particular related to optimization and search. In this, a wide variation of model parameters are generated and tested for their fitness, the best performing of which are more likely to be used for generating the next “generation” of model parameters. Successive generations of this procedure can efficiently identify good model parameters for performing a particular task.
H. HEURISTIC SEARCH TECHNIQUES
Heuristic search techniques aim to solve problems quickly by using approximations and shortcuts where classical methods and approaches cannot easily or directly provide a solution (e.g. in Chess – evaluating all possible moves and their far-reaching consequences quickly results in an intractable number of game positions). Heuristic search techniques are often used where a solution can be reached iteratively – the heuristic provides a mechanism for evaluating the options at each step and following only the most promising options.
The internet of things (IoT) is a system of interconnected computing devices, mechanical and digital machines, objects, animals or people that are provided with unique identifiers and the ability to transfer data over a network without requiring human-to-human or human-to-computer interaction. IoT is often used in the context of connecting devices to the internet which are not traditional computing devices, such as vehicles, appliances and sensors.
Jeopardy is an American television game where contestants are presented with general knowledge clues in the form of answers and must phrase their responses in the form of questions. IBM Watson (an AI application) played Jeopardy against human opponents on television, beating them. Playing games like Jeopardy can help AI systems to learn strategy, negotiation and the ability to predict what humans will do, which can also be applied to business problems.
K. KNOWLEDGE ENGINEERING
Knowledge engineering is a field of AI that tries to emulate the judgment and behaviour of a human expert in a given field. Knowledge engineering is the technology behind the creation of expert systems to assist with issues related to their programmed field of knowledge.
L. LIMITED MEMORY
Limited memory systems are systems with a short-term memory limited to a given timeframe. This is a function which is often used in the area of automation (e.g. self-driving cars, self-working robots or Apple’s Siri). For example, self-driving cars use limited memory systems by combining pre-programmed and observational knowledge. The program retains various data for a short period to automate a task like driving along a specific road.
Concerning limited memory systems, it is to note that once the data or a set of instructions are programmed into the system, they are not able to change the programmed behaviour anymore. From a legal point of view, the short data retention period makes limited memory systems an interesting proposition from a data protection point of view, while the fact that the rules of behaviour cannot be changed may result in a liability issue.
M. MACHINE LEARNING
Machine learning (ML) is a set of algorithms that can be fed only with structured training data to complete a task without being programmed how to do so. The algorithms in the set build a mathematical model from the training data to enable them to make predictions or decisions based on new, unseen data.
N. NATURAL LANGUAGE PROCESSING (NLP)
Natural language processing (NLP) is a field of study focused on the interactions between human language and computers. NLP helps machines “read” text, understand speech, and generate text/speech by creating models of human language and speech. NLP plays a large part in applications such as the Google Assistant.
O. OPTICAL CHARACTER RECOGNITION
Optical character recognition (OCR) is the conversion of images of typed, handwritten or printed text into machine-encoded text. This was one of the first areas in which neural networks showed their potential, by accurately classifying handwritten numbers. Today, OCR enables efficient and accurate digital processing of handwritten text.
P. PATTERN RECOGNITION
Pattern recognition is the ability to detect arrangements of characteristics or data that yield information about a given system or data set. This can be used to identify previously unknown clusters of people, for purposes as diverse as identifying marketing segments and identifying patients who may respond well to a particular drug.
Q. QUANTUM COMPUTERS
Quantum computers directly utilise the possibilities of quantum mechanics to efficiently solve a number of computationally difficult problems. Currently, quantum computers are in the proof of concept stage and scaling up the number of qubits (the quantum analogue to traditional binary bits) they can use. At the current pace of progress, it is only a matter of time before many traditional methods of digital cryptography are rendered insecure to quantum computers.
R. ROBOTIC PROCESS AUTOMATION (RPA)
Robotics is the branch of technology that deals with the design, construction, operation, and application of robots. Robots today are used to perform repetitive actions or jobs considered too dangerous for humans. In combination with the robot, robotic process automation (RPA) is often used. This is software with AI and ML capabilities that enables the robot to perform repetitive tasks that were once completed by humans. This means that AI equipped computers become part of the brain of a robot, leading to questions such as: who is liable for the robot – the AI developer or the engineer that created the robot, the company requesting the robot to be built or the company deploying the robot in a situation where it can cause damage?
S. SUPERVISED LEARNING
A supervised learning algorithm analyses training data and produces an inferred function which can then be used to map new examples. This is in contrast to the alternative method of unsupervised learning, which works by taking information that is neither classified nor labelled and allowing the algorithm to analyse that data (for example, identify clusters of data points or patterns within that data) without guidance.
T. TURING TEST
The Turing test is a test which was created in 1950 by the computer scientist Alan Turing to see if machines could exhibit intelligence equal to or indistinguishable from that of a human. It involves one person trying to identify, in a conversation between another person and a machine, which participant in the conversation is the machine. Many extensions and variations of the test have been proposed. In a twist of fate, these days it is more likely that a human has to prove that he or she is not a robot, e.g. when confronted with a CAPTCHA (Completely Automated Public Turing test to tell Computers and Humans Apart) than the reverse.
U. UNSTRUCTURED DATA
Unstructured data is information that either does not have a pre-defined data model or is not organised in a pre-defined manner. Structured data on the other hand consists of clearly defined data with easily searchable labels or patterns.
V. VIRTUAL ASSISTANTS
A virtual assistant (AI assistant or digital assistant) is an application program that understands natural language voice commands and completes tasks for you.
W. WEAK AI
Weak AI is another term for “artificial narrow intelligence” which is AI that is programmed to perform a single task. This is in contrast to Strong AI, which is AI which would have an intellectual capability equal to a human’s, which could rapidly learn to solve a number of unrelated tasks, and transfer concepts it has learned and identified from one problem context to another.
X. X.A.I. OR XAI (EXPLAINABLE AI)
XAI is AI that is programmed to describe its purpose, rationale and decision-making process in a way that can be understood by the average person. This form of AI is more likely to be considered compatible with consumer protection concerns regarding automated decision making.
You will increasingly encounter AI in your daily life and from a business perspective. To be diligent when implementing an AI solution in your business, we recommend that you consider compliance and liability questions. When drafting a contract relating to AI (e.g. development contract or a license contract) ensure that it is suitable for this form of technology and uses the correct language.
Z. ZHANG ZHAO
Zhang Zhao is a Chinese news anchor on which the first artificial intelligence news reader was modelled. The AI news anchor, which is deployed by the Chinese Xinhua News Agency, learns from live videos, reads texts and reports via social media and on the Xinhua website.