How artificial intelligence can help your business? This is a question being asked by many business people. So, I asked Artificial Intelligence expert, Darren Bentham, to share his expertise with you.
Do You Use Artificial Intelligence?
I’ll begin this article with a question… do you use Artificial Intelligence? You might begin to think about the Terminator movies, C3P0 from Star Wars or (if you’re of a certain age) Metal Micky. The reality is that Artificial Intelligence (AI) has been around for a long time, and most, if not all of you, will be using AI on a daily basis without realising it.
You’ll be using predictive text on your phone, AI will be deciding which adverts to show you on the internet, labelling your friends in photos, powering the latest voice controlled Alexa devices, adding cat whiskers to your photos or helping keep you safe in your car when you drift out of lane.
What AI Is Useful For
Not only this, AI can spot potential fraud in loan applications, pick out early signs of breast cancer from scans, tell parking companies when you’ve over stayed your allocated time and recognise if the photo on your passport matches the person in front of the camera.
On the darker side of technology, Artificial Intelligence can track a person as you walk around a city and inform the authorities when you leave the house, at what time, and who you speak to along the way. Scary stuff, eh? All of these areas are science fact, not science fiction, and are here now.
The Future Of AI
In the future AI could replace your accountant, lawyer or doctor and offer a cheaper alternative to a trained human professional. It will probably be Artificial Intelligence software that drives you to the airport or takes you home after a night out.
AI is a useful technology that can automate or enhance human decision making. Most people use it on a daily basis and many people don’t even realise they’re using Artificial Intelligence software. It can provide many benefits to society, but it isn’t perfect and can have unexpected side effects that you wouldn’t expect a human to make.
Similar to other new technologies, not everything is positive and there are some pitfalls we should be aware of. The best way to describe the negative side of AI is by explaining how it works and illustrate past failures that have produced unsatisfactory results. After which I’ll move on to describe various ways in which you can make use of AI within your own organisation.
Firstly Artificial Intelligence systems aren’t programmed, as is usual with most of the software you will be used to, such as Microsoft Office. As with the education human children, Artificial Intelligence is ‘trained’ by being shown examples of what is to be learned and told when a correct response is produced; whether it’s given a right or wrong answer. With enough examples and strict supervision, the system will slowly improve until it can reach a satisfactory level of response for most of the time.
An example of an AI system – automatic recognition of dogs
A system was created to recognise dog breeds from supplied photos. During the training process the software was provided with multiple photos of each breed of dog, with an associated label to state what the photograph contained. Numerous photos of a Boxer were shown during training and the system was told ‘this is a boxer’. Likewise for Pugs, Alsatians, Sausage Dogs, Retriever’s etc. etc. The finished software was capable of recognising the breed from a photo it had never seen before with a very high level of accuracy.
Everything worked well until someone decided to show the AI a photo of a chicken drumstick and the program incorrectly decided it as seeing a photo of a Cockapoo dog. This behaviour may seem absurd and clearly a mistake, until you remember the software had only been trained with dog photos and the closest example of a dog it had be shown that was light brown in colour and fuzzy at the edges was a Cockapoo.
Likewise, when shown a photo of a blueberry muffin, the software decided it was seeing a Chihuahua – due to the shape of the blueberries forming a potential triangle pattern of two eyes and mouth.
Artificial Intelligence Learns By Example
As been shown an example of something it can learn to recognise the unique characteristics within the example and use this ‘knowledge’ to classify future examples encountered. However, if the software encounters something it’s never seen before it will try and locate it from among the examples it has been shown – which may be acceptable where a near match is required, but if accuracy is what you’re looking to achieve the solution may be unsuitable and problems can occur with disastrous consequences; as was seen in the case of the car accident involving a self driving Tesla car that mistook a shadow for an oncoming vehicle and drove into the path of another car resulting in a crash killing the driver.
Unlike humans, AI works best when it is able to learn from a high number of examples of what it is being trained to recognise. Compare this with a human, who will struggle to extract much meaning from 5 million examples of credit card applications or photos of dogs. A situation where they would quickly become overwhelmed by the sheer volume of information. By comparison, a computer program would use the data to enforce its learning and become more and more accurate as each new record is read.
This is in part because humans take a long time to absorb information and have a limited memory capacity for complex data sets. A limitation a machine can easily overcome, especially when made to run on a powerful computer to reduce the time needed.
Humans are better at data abstraction
It’s not all bad for the human and good for a computer, there are cases where a human can out perform a computer and provide better responses. This is specifically relevant where the data being used in the training process is incomplete or of a small sample size. A child shown a photo of a car will quickly learn to identify those features of a car that make it a car and not a van, truck or motorbike.
A computer by comparison will need to be shown lots and lots of examples of each, from all angles (by turning it at angles and saving as a new image) and of different sizes within an image (one that is far away and more than are closer). A human only needs to be shown one image before learning takes places and even a young child can tell you that a blue Ford Focus is the same as a red Ford Focus.
Uses of AI within your own business
Despite public perception, Artificial Intelligence isn’t limited for use within large business organisations, small enterprises can make use of this technology. If you have a store of data or images from which you would like to extract meaning and automatic decision making, the information can be turned into software that will automatically evaluate new data and tell if the data matches a scenario previously seen – such as the probability an application is going to end in a loan default, a production line failure on an electric motor about to happen, who a person in a photograph or video stream is, recognising a spoken word that can be used to lookup records in a database, learning normal patterns of behaviour from unstructured data and detect changes in a persons electricity usage, spam detection in emails… the list goes on and on.
In short, if there’s a large set of data that exists and in which there is a pattern to the data, the chances are it can be used to provide new insights and automatic decision making. This could be a list of previous customer applications, vibration and heat data from an electric motor or a series of labelled photos.
Who Do You blame?
Who do you blame when things go wrong and who has accountability?
When a human makes a mistake they would be able to explain why they made a prognosis, which a computer AI wouldn’t be able to do. This leads to a moral question – which would you trust the
most with a medical diagnosis, a human who is 92% accurate and could explain why a certain decision has been made, or a computer that’s historically been 96% accurate but is unaccountable should anything go wrong?
Further more, when mistakes do take place who is at fault, the user, the software developers or the person/company providing the data used to train the Artificial Intelligence system. All were involved in creating and using the software but no single one was in full control.
This lack of a single control point within Artificial Intelligence software has been causing interesting conversations in circles where AI is being used to produce original creative pieces of work; such as art. In this case who would own the copyright on the work produced, the creator of the software, the source of the sample images used to train the system, the owner of the software license or the actual computer program itself?!
These are all issues yet to be discussed and ironed out. Until then we’ll continue to see AI used more and more in our everyday lives and workplaces, often being overlooked and silent helping keep us healthy and more efficient.
Types of Artificial Intelligence
As with other types of software there are multiple types of Artificial Intelligence software around. Each works in a different way from one another and has different strengths (and weaknesses).
Artificial neural networks (ANN’s) are one of the main tools used in machine learning. As the “neural” part of their name suggests, they are brain-inspired systems that are intended to replicate the way that we humans learn. Neural networks consist of input and output layers, as well as (in most cases) one or more hidden layers that transform the input into something that the output layer can use.
ANN’s are used to identify patterns in large data sets that are too complex or numerous for a human to identify. For example, the correlation between a patients medical data and the chances of them developing specific forms of cancer. Click here to find out more.
A genetic algorithm is software for solving optimisation problems that is based on natural selection; the process that drives biological evolution. The genetic algorithm repeatedly modifies a population of individual solutions.
At each step, the genetic algorithm selects individual solutions at random from the current population to be parents and uses them to produce the children for the next generation. Over successive generations, the population “evolves” toward an optimal solution, and the system slowly improves.
Chatbots are programs that have been designed to simulate how a human would hold a conversation with a human. Replicating what an expert would usually ask and then offering further questions depending on what response has been received.
Intelligence can be incorporated into the conversation using evidence-based responses, where the next question is dependent on the series of previous answers provided.
Unlike Humans, chatbots can operate 24 hours a day, 365 days a week and without needing to be paid.
About Darren Bentham
Darren Bentham is a software developer with over 25 years experience and an academic background in Artificial Intelligence at both undergraduate and postgraduate levels. His business, Internet IT, offers custom software development and AI services to businesses, helping them use big data to produce effective decision making systems to further improve their processes.
Darren is the Managing Director of full service digital marketing agency, Interact Digital.
Now you know how AI can impact your business, get in touch with Darren directly to find out more.