Advertisers, their agencies and the whole marketing ecosystem are beginning to understand what the consumer wants rather than where they have been, thanks to AI – writes Joe Nguyen of the IAB Singapore.
A few years ago, ‘big data’ was the in thing. Marketers, agencies and publishers, as well as every other company, were told that this could change their businesses.
There was a plethora of information available that they could use to better compete, better market and even better strategise if they could leverage big data.
But what many companies realise now was that they just opened a can of worms. Yes, there was tonnes of data out there – too much in fact – but a lot of it was confusing and hard to harness correctly.
Companies were trying see how big data could help change the way they do business, open new opportunities or compete on analytics. The reality was that this took more time and resources than many thought it would.
Then in came the data management platforms – or DMPs – particularly for the advertising industry. These DMPs (not to be confused with data marketplaces) would enable advertisers, agencies and publishers to connect up and integrate various data sources that a company has access to or own.
What these platforms have actually done is to enable the connection between data sources and increase the efficiency or usability of datasets. Ultimately, with any data that was ingested, there was still a need of a ‘key’ – a way to link the data sets together.
This is where artificial intelligence comes in. So just what is meant by artificial intelligence in this context? Well, artificial means human-made. Intelligence is when the term AI gets murky. The Merriam-Webster Dictionary has a number of definitions, and the first two are:
(1) The ability to learn or understand or to deal with new or trying situations :reason; also :the skilled use of reason.
(2) The ability to apply knowledge to manipulate one’s environment or to think abstractly as measured by objective criteria (such as tests).
Are computers able – at this point in time and with current technology – to learn, understand or deal with situations? Can they reason? Can they really think abstractly?
Perhaps those questions are too cerebral and not practical enough. But marketing terms like ‘AI solutions’ or ‘AI-powered engines’ want us to reference HAL 9000 from 2001: A Space Odyssey, Mr. Data from Star Trek: The Next Generation, and R2D2 and C3PO from Star Wars.
This just seems like hype, because the big difference between those and anything that exists today is sentience and self-awareness. We are supposed to think that there is an intelligence within these platforms that can access and process infinite amounts of information to help us market better.
This is just not the case. And this does not mean that the definition of AI is sentience or that the Turing Test is not valid.
Current platforms are using algorithms, machine-learning and deep learning to teach computers how to process data. These are all very exciting developments and will immensely help us. These innovations may eventually lead to the AI that we all think of, but in business, we need a reality check.
Let’s consider the deterministic versus the probabilistic. We need to understand the fundamental difference between what we have been working with and what machine learning is. Up until recently, most who use data and databases were working on a deterministic premise.
That means, in order to connect datasets, there had to be ‘hooks’ or data fields to link one data set to another – like a cookie match, device ID, IP addresses or browser ID. Using this data required all these connections be made directly.
With machine learning and deep learning, the platform makes connections between different data sets based on patterns and correlations. This means that connections are modelled on statistics and probability.
Of course, when a cookie is targeted for example, there is still a certain amount of probability involved, but most people do not factor that in or it is taken for granted. Now, with the introduction of the General Data Protection Regulation rules in the European Union, deterministic data matching can become very challenging.
There are many platforms that use machine learning and deep learning today. Google, Bing and Baidu have been doing this for years to enable better search results. Amazon’s recommendation engine that shows shoppers items ‘for you’ is based on algorithms built according to what your behaviour on the platform.
Netflix does the same with its own algorithms. Beyond just making recommendations, this is really personalisation and improving the customer experience for a brand. In marketing advertisers, their agencies and the whole ecosystem are beginning to understand what the consumer wants rather than where they have been.
Hence, they are able to provide more relevant advertising. Brands want to reach the right consumers, at the right time, with the right message. And AI-enabled programmatic platforms are making this a reality.
But are we optimising towards the end of the world? With machine learning, humans are teaching computers what patterns to investigate and it is a continual learning process. The platforms need a lot of data in order to learn, and they learn by trial and error.
Programmers, data scientists and fellow geeks must continually monitor and tweak the algorithms so that the accuracy gets better and better with time. So be patient. Machines can process more information much faster than humans can, but without proper instruction they can easily go down the wrong path.
The wrong path in the 1983 movie War Games saw the computer – based on parameters given to it – decide that the best way to win a nuclear war was by starting one and getting most of the opponent’s nuclear arsenal first. It did not care that millions of people would die or think about the radioactive fallout because it was not programmed to do so.
So when working with machine-learning and deep-learning computers, the responsibility is with the humans that programme them and provide the parameters within which to operate and optimise.
So in business, when we use computers for machine-learning and deep learning, we need to be thinking in terms of probability and likeliness rather than just ones and zeros. We need to take responsibility and invest in resources to guide these computers in the right direction so that the probability for success increases as the system gets better and better.
Everything that is done now contributes towards the evolution of AI. Whether sentience will be a reality or not is almost impossible to guess; that ‘spark of life’ and self-awareness is a notion on which humankind has been pondering for a long time now.