By Duncan Everett (MD Optimal Monitoring)
Artificial Intelligence (AI) is currently perceived in many ways. Some believe the Hollywood view of the self-aware, fully thinking autonomous entity, that will take over the world. Others that is no more than a complex, rules-based search engine that clever programmers have disguised as your new friend. They’re both wrong.
Busting the Plug & Play Myth
Let’s think for a moment about us or more precisely you. Firstly, think back to when you first started school. At that point did you have all the knowledge to do your current job? No. Now what about when you left school and got your first job, did you have all the knowledge to do your current job? Still no. Throughout our life, we learn the ‘rules’ which we then apply to do our job, some technical, some social; that’s learning. Throughout our life we fine tune these rules by promoting those that work or deliver the outcome we want and demoting those that don’t. Using this process, we hopefully become an expert at something.
AI is a process of reproducing this learning process in a machine. To do that the machine needs to go through the same process as we did, learning the logic by trial and error or being programmed with it (like you were by your parents and teachers) so that, just like you, it will ultimately become an expert in something.
How AI ‘learns’
Machine Learning currently has three primary methods of achieving this learning;
This is where a machine creates a complex set of rules by being told existing outcomes by a human. To do this first it needs someone to look at lots of data and tell it how you classify the meaning e.g. heating on a hot day is bad , cooling on a hot day is good, heating and cooling at the same time is always bad! The machine then builds a model which can apply this learning to new data.
Here you give the machine lots of data and ask it to group it into what it believes are similar conditions. A human then decides which of these conditions are of use and which to ignore. e.g. where the pattern shows nights when things were running when everyone had gone home and where the pattern looked like the cleaner was in (because the power went off three hours after normal business hours when nobody was expected to be in the building so who switched it off?)
In this case you can start with either a supervised or unsupervised technique as a basis. Feedback is then used to help the machine learn to perform a task better. Feedback can be a measure of performance against a goal or human feedback (good, bad). The machine will try to maximize rewards it receives for its actions (e.g., maximizes points it receives for increasing returns of an investment portfolio). When it makes a good investment it ‘banks’ the knowledge that it used to make that decision, when it makes a bad investment it banks that knowledge to not use it again.
Ultimately the machine is ‘learning’ a complex set of logical rules of the chosen task from its teacher, just as we did in school and throughout our careers. The advantage of the machines that it can apply this learning 24/7/365 very quickly at a low cost. As a result, tasks which were previously not cost effective to perform or where the result was delivered too late to be useful can now be realised.
Historically AI has been used to implement relatively simple, narrow specialisations. That is changing quickly as the technology is deployed across ever wider environments and as computing power increases.
So is AI a self-aware, fully thinking autonomous entity, that will take over the world? Well no. But it can be used to perform complex analysis and problem solving giving the impression of understanding and reasoning.
What could the future look like
Looking at my field ‘Energy Management’, AI can be used to interpret utility consumption data and make a reliable assessment of how-to best deal with unexpected or unsupported consumption. The commercial benefit is obvious; the system sends out instructions on what to do having analysed the data instead of just sending the data to a human to perform the same analysis, inevitably much more slowly.
By continuously applying machine leaning processes and continuously gathering user feedback, such systems learn about the specific nuances of a location or type of location, so becoming better and better at their jobs (exactly like you have done at school, university and, in your career).
It will change the way we work. No more will we spend time and effort collecting and processing data. The machine will do that. Our role becomes to take informed action and oversee the machines decisions, not unlike a pilot does when the aircraft is on autopilot.
So, humans become the higher level management of situations and outcomes whilst the machine does the grunt work. Everyone wins.
Artificial intelligence: A definition
AI is typically defined as the ability of a machine to perform cognitive functions we associate with human minds, such as perceiving, reasoning, learning and problem solving. Examples of technologies that enable AI to solve business problems are robotics and autonomous vehicles, computer vision, language, virtual agents and machine learning.
Machine learning: A definition
Most recent advances in AI have been achieved by applying machine learning to very large data sets. Machine-learning algorithms detect patterns and learn how to make predictions and recommendations by processing data and experiences, rather than by receiving explicit programming instruction. The algorithms also adapt in response to new data and experiences to improve efficacy over time.
More useful reading
McKinsey An Executives Guide to AI
If you would like to know more you can contact Duncan on 01494 435137, email Duncan@optimalmonitoring.com or contact Optimal Monitoring on 020 7439 9259.