Artificial Intelligence (AI) has been in the public consciousness for decades. Probably not long after the phrase was coined by John McCarthy and others in 1955 it captured people’s imagination.
Since then, there has been a great deal written, filmed and discussed in relation to AI, both fact and fiction. Much of the fiction portrayed AI in a negative or dangerous way. Often describing a future where AI has become too smart and powerful, and taken over humanity.
Obviously this is fiction and not going to happen anytime soon. Regardless though, AI, like any area of research and innovation, should be closely monitored. Although recent advancements do not take us anywhere close to signalling the rise of our AI overlords there is always risk that mistakes could be made, and that people could be impacted.
But how likely is it that “rogue” AI could lead to a person being impacted in a negative way?
One area that is progressing rapidly in business where a person is usually affected, though not necessarily negatively, is the use of AI to inform decision-making. Whether this is eligibility for government services or credit worthiness for a loan, AI is increasingly being used. To understand the risk, we need to understand how it works and what decisions are being made.
Over the next two blog articles we will discuss how and where AI is being used in decision making, what this means to individuals and whether it is something we should be concerned about.
For decades, businesses have attempted to find different ways to improve operations, save money and enhance efficiency and productivity. Whether it is in terms of new systems, process redesign or outsourcing there have been many waves of transformation as organisations have sought to find a competitive advantage. Most recently AI appears to be part of the next wave of transformation that could offer a much sought after competitive advantage.
To achieve this, businesses are utilising AI that replicates some elements of human activity. This includes analysis of data to make judgements, predictions, and decisions.
The latter, decision-making, can be a challenge for many businesses as they are often heavily reliant on humans to complete the activity. As a result, it can be slow and inefficient leading to bottlenecks – impeding productivity. This can be because those involved are overloaded with work, slow to make decisions or, there might be too many people involved in the decision-making process.
With ever-increasing data storage and computing power, AI has the potential to augment human intelligence and enable smarter decision-making and even automate decision making without any human involvement.
Here are some examples of how AI technology are being used for decision-making:
Rules-based decision making involves understanding inputs involved in a process or decision and then selecting an outcome based on the combination of these inputs. A very simple example is when a person decides on what clothes to wear, they would reach a simple decision based on inputs such as the weather and what is happening that day. The person actively searches their memory to recall the matching rule to find the right answer. Software solutions can mimic this behaviour to process complex decisions at high speed.
Rules based decision making is generally considered at the more traditional end of AI. This is because the software does not automatically “learn” based on large sets of data. Consequently, in this data rich world, technology that does not utilise data is considered simple.
One key benefit of using this approach is that the basis for any decision is that it can be explained, understood and therefore justified. This can be essential when decisions are being made that can impact customers or individuals.
Machine Learning decisions
Machine learning decisions are when computer algorithms are created and used to make decisions or predictions based on data inputs.
The essential difference with machine learning based decisions is that they are based on what the software can learn from data rather than being programmed based on human knowledge or expertise. For complex tasks, it can be difficult for a human to create the software program to solve the problem. Therefore, it can be more effective to enable the software to create the algorithm itself, rather than having to develop every step manually.
In very simple terms this is done by providing the machine learning software with a data set that includes the outcome or decision which it then analyses to identify all the combinations that result in each outcome. This is the “training” of the machine learning model. When the software is provided with new data it uses the model to identify, or predict, the outcome. Predicting an outcome can be used to make a decision in a business process.
The drawback is when there is a need for the prediction to be explained or justified. This is not easy to do with a machine learnt model. Without transparency then there is a risk that the algorithm could be unintentionally used in the wrong way or even unethically.
Whether rules or machine learning based, AI is being utilised for decision making more and more in different sectors. Here are some examples:
(Ok, so Marketing is not really an industry sector. But, it is where most of us come into contact with AI the most so worth discussing briefly.)
AI is being used to understand customer behaviour and identify what services, products, and engagement they are likely to respond best to. By analysing previous data on customer engagement and outcomes machine learning algorithms can predict when specific outcomes will be achieved.
This is then used to inform decisions by marketing companies or to automatically push adverts that target customers and potential customers, often with freakish accuracy (if you’ve ever wondered whether your phone is listening to you because it suggests that you search for something that you’ve just been talking about, you know what we mean!).
Repetitive processes and decisions based on information available in large volumes, means the insurance sector has great potential to be transformed by AI. A key area of focus is the prediction of risk and using that to inform decision making.
AI is already being used to make decisions on insurance claims. Those leading the way are using AI to assess claim applications against historic claims and the customer information to determine whether it should be paid out directly or investigated further.
AI is also being used in the underwriting process or detecting fraud. By analysing historic data on insurance policies and claims algorithms are being utilised that predict the likelihood, or risk, of a claim being made. This risk is then applied to the underwriting of a policy and the pricing of premiums.
Interesting but not particularly a cause for concern, right? Well, in our next blog we will discuss some slightly more controversial examples and whether we really do have something to worry about…