Mention the words Artificial Intelligence and Machine Learning today, especially in the context of systems, and you are sure to grab the attention of everyone in the room.
AI and ML have captured the public imagination in a manner unprecedented even during the dawn of the 21st Century when smart technology first came into its own. Smart systems, however, still had a more or less two-dimensional implication when we consider how things have exploded with today’s AI-integrated systems.
So, why are today’s digital nomads so fixated on AI? The most obvious answer is probably also the most accurate. AI is projected to boost human intelligence and help us maximise our potential. One of the primary ways it will do this, futurists concur, is through helping us enhance our collective intelligence. While it has always been a foregone conclusion that a group of individuals thinking collectively about a particular problem are likely to have a better rate of success in finding the optimal solution, AI has gone back to brass tacks to draw inspiration from the animal kingdom to investigate at length how this actually happens.
How does a colony of ants carry morsels of food that almost equal their body weight back home? Or how does a swarm of bees or locusts know exactly in which direction to fly – almost as if each individual member of the swarm has invisible transmitting antennae to communicate their intention? Well in a sense they do – ants secrete pheromones to direct their team members to resources, bees use vibrations to convey the message, fish set off tremors in the water around them, and birds detect motion spreading through the flock. Swarm AI, a recent advancement, refers to the same collective behaviour of decentralised and self-organised systems to maneuver quickly in a coordinated fashion. Clearly, Swarm AI foretells considerable potential when it comes to harnessing collective intelligence.
Collective intelligence, a central goal in introducing AI into modern systems, is the result of a group of individuals (or components) acting together to enhance efficiency, productivity, and resilience. There are four ways in which AI can enhance collective intelligence. Helping make better sense of data, finding better ways to coordinate decision making, helping us overcome our inherent biases and highlighting unusual solutions that are often overlooked.
AI has come into its own largely due to the demand to analyse and sort through the enormous volumes of data we find ourselves in possession of today. With the world we live in now flooded with smart devices of all shapes and dimensions, we are witnessing an actual shift from cloud AI to on-device AI. Data is usually transmitted from device to cloud, where it is used to train or establish models, and then transmitted back to be deployed on the device again. Such transfers of data certainly have cost and security implications that cannot be disputed.
Furthermore, the transmission of user data to cloud and back following storage in the cloud, is open to interference and capture. Stored data risks unauthorised access problems. Significantly, cloud-based AI and ML models have higher latencies, cost more to implement, lack autonomy, and, depending on the frequency of model updates, are often less personalised.
So, how does one deal with the challenge of huge volumes of data computation and spatial distribution of computing resources? One solution is Distributed Artificial Intelligence (DAI) – an approach to solving complex learning, planning, and decision making problems. This approach facilitates reasoning, planning, learning and perception to solve problems that require large datasets, through distributing the problem to autonomous processing nodes (agents). In doing so, it enables us to move the computing to where the data is – instead of the other way around and to avoid the previous challenges encountered.
To optimise this approach, DAI requires a well distributed system with robust and elastic computation on unreliable and failing resources that are loosely coupled, as well as coordinated action and communication among the nodes, and sub-samples of large datasets and online machine learning. Clearly, there are multiple reasons for wanting to distribute intelligence or rely on multi-agent systems.
Parallel problem solving is a core advantage of the DAI approach and focuses on how classic artificial intelligence concepts can be modified, so that multiprocessor systems and clusters of computers can be used to speed up calculation.
Distributed problem solving (DPS) is another plus – the concept of agent autonomous entities that can communicate with each other, was developed to serve as an abstraction for developing DPS systems.
In addition, the DAI approach helps leverage multi-agent based simulation (MABS), building the foundation for simulations that need to analyse not only phenomena at a macro level but equally, at a micro level too.
Through the DAI approach, we are able to efficiently analyse vast volumes of centralised data, if we succeed in designing an airtight AI system.
Predictably enough for such a dynamic field, as the complexity of AI systems continues to grow, we will need to go beyond the DAI approach and leverage a Federated Learning (also known as Collaborative Learning) approach. This is a machine learning technique that trains an algorithm across multiple decentralised edge devices or servers holding local data samples, without exchanging them. This approach is in sharp contrast to traditional centralised machine learning techniques where all the local datasets are uploaded to one server. This also varies from more classical decentralised approaches which often assume that local data samples are identically distributed.
New UAE visa rules set to have ‘tremendous’ impact on AI, emerging technologies
Federated Learning enables multiple actors to build a common, robust machine learning model without sharing data. As we highlighted earlier with collective intelligence, this approach allows researchers to address critical issues such as data privacy, data security, data access rights and access to heterogeneous data. While Federated Learning holds significant potential for applications across multiple industries including defence, telecommunications, IoT, and pharmaceuticals, speed and scale are still some hurdles we need to overcome in enabling such advanced solutions to gain critical mass in the short-term.
