Artificial Intelligence will impact many aspects of our lives. There are lots of possibilities (and limitations) of this exciting technologyThe title is meant to be flippant being based on a comment my wife made when I proudly showed her some visual recognition demo I’d done for a client and she then tried it with a picture of our grandson, and it told her there was an 82% confidence level that we had a ‘good widget’.
The problem was, of course, that the model used had only been trained to identify a particular object (damaged or not damaged) with any degree of success. It was, again, of course easy to try it with any image and still get a score from the system.
So, I’ll try to set out in this short blog some of the exciting things AI can do when set in context and some of the implications it may have for the future in work and society in general. A warning! Many have written whole books on this subject, this blog is no more than a teaser to build some awareness of the subject and hopefully gain the interest of the reader to delve further.
AI and its relationship to Machine Learning and Deep Learning can be summarised by the diagram below:
Image credit: Graph reproduced with permission from author, Steve Lockwood
Techniques such as classification, clustering, regression, supervised and unsupervised learning, alongside the latest deep learning approaches, constitute aspects of what some refer to as AI. The primary goal of these approaches is to be given input and output and let a ‘black box’ work out how to build a function to approximate the prediction of input to output.
Big Data has been essential for the development of AI, creating large data sets to train for AI to train against. More data is one approach to enabling better trained models (other techniques can be applied such as dimensionality reduction). We also need to combat bias in models created. This article from the MIT Technology Review clearly describes these problems. This is important as biased models have the potential to cause havoc in all aspects of our lives. Protecting against biased models in the data used to generate the models and in the models themselves will be a key goal as we move forward in this field.
Industry needs tools to help it understand where bias may occur and be able to show how results were obtained and fix them. Remember the black box mentioned earlier? Our solutions also need to become more transparent to those who have permission to scrutinise them. This is a fledging part of building dependable AI systems and several companies are building toolkits to help with these challenges. This excellent article on Medium explains the issues and the state of things currently: ethics, bias and transparency in AI.
We will all be impacted by these innovations in the years to come, whether as an employer, employee or citizen. It’s worth getting to grips with the ramifications of these technology breakthroughs.
In short, we will all be impacted by these innovations in the years to come, whether as an employer, employee or citizen. It’s worth getting to grips with the ramifications of these technology breakthroughs.