AI & ML – what’s the difference?

3-minute masterclass
AI & ML what's the difference blog projective visual
AI & ML what's the difference blog projective visual

Our clients often ask us what the difference is between artificial intelligence (AI) and machine learning (ML). Lots of people seem to use these terms interchangeably but they aren’t quite the same.

Before diving into the details let’s clarify the difference. The English Oxford Living Dictionary refers to AI as “the theory and development of computer systems able to perform tasks normally requiring human intelligence, such as visual perception, speech recognition, decision-making, and translation between languages”. And ML is “the capacity of a computer to learn from experience, i.e. to modify its processing on the basis of newly acquired information”. So, we could say AI is an umbrella term and ML is one of its many subsets.

Rule-based system

Another subset of AI is rule-based systems. These are usually built from the combined information of human specialists in the issue space. The specialists indicate every one of the means taken to settle on a choice and how to deal with any exceptional cases. This information is then incorporated into the rule-based system. So, the principles of the rule-based system are statistically defined and human created.

Why use rule-based system?

One of the key advantages of a rule-based system is that the composition and execution standards are very simple. Once research has taken place into the area of interest, we can proceed to create rules to define the application logic. It’s also very easy to compose rules. For example, if you need to represent 50 potential results or responses for your AI, you would need to compose 50 rules to cover them. In the event that, you have composed 50 definitions and you find an uncommon case you hadn’t considered previously, you can create an extra rule to handle the desired logic.

Machine learning

Unlike rule-based methods, machine learning is probabilistic and uses models rather than static rules. The basic operation of a machine learning process is to gain insight from historical data and provide future predictability. A good example of use within the financial sector is when detecting fraudulent behaviour (unusual transactions based on historical data). Another example could be an application for a loan and then to detect the probability of a loan being repaid.

With machine learning the creation of rules is typically replaced by the creation of features. Features are inputs associated to trends in historic data. If our model was predicting the acceptance criteria of a loan, a feature might be the average spend or saving amount of the user, or whether or not the user receives a sufficient income. The process involves finding parameters which give accurate outputs most of the time. The taught model is generated from the data itself rather than external supplied information.

Let’s recapitulate

Rule based Artificial Intelligence:

  • Requires static rules to be defined, predictable outcomes
  • Statically trained, easy to add data and represent/modify rules
  • Designed for static logic AI, limited to the ruleset
  • Short-term accuracy, knowledge base can be updated and extended
  • Practical for simple solutions

Machine Learning Artificial Intelligence:

  • Requires constant or large data
  • Dynamically trained
  • Designed for growth AI models
  • Strong long-term accuracy, continuous adaption
  • Practical for difficult and scalable solutions

When to use what?

We’ve learned that ML works very well in the long run as it is open to continuous adaption and improvement through data preparation, algorithm selection and algorithm parameter tweaking. Its algorithms tend to be one step away from human involvement in favour of optimisation for computers. Rule-based systems can offer quick, tactical solutions and workarounds. The requirement for human specialists’ input can help wider business and make it easier to explain how decisions were decided. Most projects begin with a rules-based system to explore and understand the system and later incorporate machine learning algorithms.

Both concepts are already quite old as they were developed in the 1950s. But this doesn’t mean they’ve become obsolete. With the advances in processing power, machine learning algorithms have become more popular. They’re even defining a new standard of AI to solve complex problems.

At Projective we’ve already helped several companies transform and optimise their processes with the help of AI. If you’d like to know more about what we’ve done, we’re always up for a good chat! In the meantime, don’t hesitate to check out our other specialisations and solutions.