{"id":1628,"date":"2023-03-21T08:43:53","date_gmt":"2023-03-21T08:43:53","guid":{"rendered":"https:\/\/quantiv.com\/?p=1628"},"modified":"2023-03-22T13:51:16","modified_gmt":"2023-03-22T13:51:16","slug":"the-secret-behind-artificial-intelligence-data-in-context","status":"publish","type":"post","link":"https:\/\/quantiv.com\/the-secret-behind-artificial-intelligence-data-in-context\/","title":{"rendered":"The secret behind artificial intelligence \u2013 data in context"},"content":{"rendered":"

<\/div>
<\/div><\/div>

\"ChrisChris Conway
\nChief Architect, Quantiv<\/p>\n<\/div>

It seems everyone is talking about artificial intelligence, with many of its enthusiasts lauding it as the potential answer to just about everything. AI is now being used in everything from web search to photography\u2026 and even in toothbrushes.<\/p>\n

But while AI may be ubiquitous, its definition is much more elusive.<\/p>\n

In part, this is because the term AI is often used to describe several ways in which processes<\/a> can be improved, to support improved service levels or cost savings.<\/p>\n

Automation is the first of these techniques. Here, the actions to be performed are predefined in code and so can happen automatically when specific criteria are met. A user still clearly sets the rules about which values of the criteria are associated with a particular action, and the machine acts in accordance with those rules.<\/p>\n

Examples of this automation could be having a phone change to silent mode in certain locations, or for shares to be sold at a defined price.<\/p>\n

[As an aside, I\u2019m writing this on a train and my phone has gone into driving mode, presumably detecting motion. This shows the definition of a situation can be more subtle than it might first appear.]<\/p>\n

You might think of this as just \u2018coding 1.0\u2019. In other words, it\u2019s what IT<\/a> has been doing for years (my phone and trading systems both work this way already).<\/p>\n

From automation to machine learning<\/h3>\n

Machine learning (ML) \u2013 another technique often labelled as AI \u2013 takes automation a step further. The actions still need to be coded, but ML adds coding of the decision-making process, too. The criteria associated with the actions are predefined by the ML coder, but the \u2018machine\u2019 (the computer) now learns the appropriate action based on the values of the criteria from previous user decisions.<\/p>\n

This could mean having an option for a phone to do whatever is normally done in a situation. So, if the user usually puts the phone to silent, that would be done. But if the user usually sends a text message, that would be done instead. And for the shares, this could mean when specific market conditions occur, the usual financial transaction could take place. So, as well as selling, perhaps buying could also occur.<\/p>\n

In effect, this is \u2018coding 2.0\u2019, and as the examples show, is possible today, although not that common yet (my phone isn\u2019t so intelligent, but financial trading systems can be).<\/p>\n

True artificial intelligence could be seen as something that needs coding of the learning process itself. So, many\/all possible actions would be tried unilaterally, instead of waiting for a user to select an action.<\/p>\n

AI \u2013 learning in action<\/h3>\n

With artificial intelligence, the initial conditions are logged, and a decision-evaluation process is used to assess the outcome of the action. The results are used to inform future action selection \u2013 all independent of any user decisions. Some outcomes could be annoying or inappropriate but that\u2019s learning in action. An example for my phone could be \u2018set my out-of-office notification\u2019.<\/p>\n

This could be coding 3.0 (although I\u2019d probably only go for 2.5). It\u2019s simply about learning to decide. And for all the talk about AI, this isn\u2019t common yet. That\u2019s partly because decision processes are actually well known, so machines are catching up. But it\u2019s also because trying random actions can be dangerous.<\/p>\n

But no matter how vague the definition of AI is, what is clear is that for any form of intelligence to work well, good information<\/a> is essential.<\/p>\n

Data in context<\/h3>\n

But good information doesn\u2019t just mean lots of data. Instead, the data<\/a> must be in context, even if that context is automatically derived.<\/p>\n

To support this, human intervention is still needed to establish the methods for:<\/p>\n