Listen to the highlights of the podcast: How does AI reason and learn?
The basics
Although machine learning and machine reasoning are two powerful AI technologies, they have two different approaches that solve different kinds of problems. In machine reasoning, we talk about human-like common sense, where ideas and concepts are represented as symbols in a computer system. Then logic or rules are used to combine those symbols to reason your way to a result. Machine learning is different. It is best described as the use of advanced statistical techniques to find subtle patterns in very large collections of data.
The early approaches to AI were in the reasoning fields. Logical reasoning tackled problems related to different kinds of games, like playing chess, that is, problems that tend to have all the information and options to hand, and where you reason through to a specific result from some initial conditions. This approach was popular for the first three or four decades of the field. But people rapidly discovered that there were certain problems that were not amenable to this type of approach.
How to make sense of big data sets, analyzing the contents of an image or understanding human speech are capabilities that machine learning has been successful at. Most of what you hear about today, referred to as AI is some type or form of machine learning, which is where statistical patterns are found in very large collections of data so that we can classify new kinds of outputs.
Let’s take a relatable example of the two models in action. Imagine you’re in a self-driving car, or an autonomous vehicle. When you're traveling from point A to point B, the car must find or plan a route – that is typically done with reasoning. There's a simple set of preconditions; there's a set of logical connections. Decisions must be made as to where the car should turn and how best to get from your home to your city office, for example. This route planning is a clear example of machine reasoning. On the other hand, when your car uses its cameras to determine that it’s at a junction and should stop – that’s an example of machine learning. So, both these two technologies combine and collaborate to allow us to solve certain kinds of problems.
Machine learning
|
Machine reasoning
|
| Machine learning can process large volumes of data and capture the hidden patterns needed to effectively predict outcomes. It tackles a pre-determined problem, with clear inputs and expected outputs. |
Machine reasoning can be seen as an attempt to implement abstract thinking as a computational system and apply human-like common sense to analyze and translate vast knowledge and learned network data into clear explainable insights. It does that by providing more contextual knowledge, concepts, and rules by which systems can obey, and from which they can start creating a model of the world around them. |
ML and MR in the telecom industry
For some time now, machine learning models have been implemented in many different places. But over time, the complexity of the systems in which they’ve been implemented has increased. Such is the case in telecom networks, where operations are becoming more intelligent and more automated.
With machine learning being industrialized across telecom networks we start to see some shortcomings of this technology:
- Machine learning relies on large amounts of learned data to make recommendations, so another solution is needed when there is less historical data to draw on.
- Machine learning models do not offer a simple way to trace the reasons behind the recommendations.
- It can be difficult to consolidate and prioritize differing advice from separate machine learning agents.
This is where machine reasoning can complement machine learning.
Machine reasoning solves problems by applying human-like common sense to learned data. It builds on the possibilities brought by machine learning, by analyzing vast knowledge and data sources to offer clear, explainable insight into the increasingly complex world of network operations, and ultimately aim for intent-based networks. Machine reasoning is able capture business intent and break it down into achievable network goals and KPIs. It can then balance and autonomously prioritize these based on the defined business intent to make recommendations and decisions enabling further automation of network operations.
Along with tackling increased complexity, we’re also seeing increased end-user expectations on the network at a level not seen before. For example, gamers’ expectations have never been higher when it comes to latency and throughput. This is also a huge market, so expectations really need to be met. More people are looking to networks to provide instant, smooth, zero-lag connectivity – essentially, great user experiences. This is the reality that service providers should be preparing for. It's also where there can be significant benefits with machine learning and machine reasoning working together. Independently, machine learning is a great solution for solving a single problem but is less suited to tasks that require more careful, deliberate, and explicit thinking. Generalizing or addressing problems that are different from the original task is difficult for machine learning. Machine reasoning however, adds these much needed skills to machine learning, contributing more abstract thinking and giving machines the power to make new connections between facts, observations, and the various things they can already be trained to do with machine learning.
The challenges and the outlook for machine learning and machine reasoning
The immediate challenge with the expansion of AI in networks is that with the growth of AI, products, systems and applications must reflect human values. Considerations also need to include how the design – and the transparency, or lack of transparency – of these systems might be codifying values or promoting or obscuring certain human-based points of view. Machine reasoning is crucial here as a complement to machine learning, as it provides recommendations that are explainable. This allows humans to trace any decisions made back through the process, which increases the auditability and explainability of the system. However, to ensure the implementation of fully non-bias AI systems, laws and technical approaches must also be implemented industry wide to allow to identify when that’s happening and to eliminate it or outlaw it as we continue to move forward with this important technology.
The telecom industry is based on trust and it’s crucial to maintain this trust as we move into a post-Covid world. For example, the way we adapted to, and used technology to work from home during the pandemic was something we weren’t expecting to see for another three years, so that puts the bar high.
As new requirements and technologies come in to play, we’ll also see the expansion of technology platforms, of limitless connectivity and a digital-first society. There's a lot to come, which is why there’s a lot of focus on organizations like the European Commission and the guidelines for trustworthy AI. We need to make sure that we build secure and trustworthy AI into our processes and our management systems as we further implement AI technologies into the telecom industry.
We have an exciting future ahead of us. Machine learning and machine reasoning are a critical part of building the future of autonomous networks, trustworthy AI systems, and platforms, through which diverse customer expectations, use cases and connectivity will run. AI-driven solutions are an essential part in making this evolution happen.
Listen to the full podcast episode, How does AI reason and learn?
Read our previous blog post in this series:
Can we use AI to build our future society?
What is the relationship between AI and 5G?
Want to know more?
Read our previous blog post in this series, What is the relationship between AI and 5G?
Learn about AI bias and human rights: Why ethical AI matters.
Read more about explainable AI and how humans can trust AI.
Read more about 5G monetization via intent-based network operations
