So far, Machine Learning has given us self-driving cars, practical speech recognition, effective web search, and many other implementations. Giorgio Cam uses image recognition to label what it sees and turn objects into song, Forkable can predict what you want for lunch before you’ve even decided, and just last week, Google released Autodraw which allows people with limited drawing skills to better express their ideas.
Machine Learning’s potential to reveal hidden information in data is amazing and it’s already a part of our daily lives. However, as MediaMonks’ Technical Director Michiel Brinkers reflects in this post, its possibilities are still largely unexplored.
He gives his view on what Machine Learning can mean for creative digital production going forward and the challenges that need to be overcome to get there.
They’re coming for us, or are they?
For those unfamiliar with Machine Learning (ML), it’s a type of algorithm which learns from patterns in data and then, based on what it learned, can recognise and predict similar patterns in new data. It’s an application of Artificial Intelligence, Read more on this here.
The technology is evolving at a phenomenal pace and every few weeks a new API, paper or prototype is released, continually raising the bar. Currently, the bulk of the research in ML is being developed by the big tech companies such as Google, IBM, Amazon and Microsoft. All have been researching ML for years and recently Facebook has also been making enormous strides.
A lot of work is also being done by universities, independent research groups and individual developers. And through the collaboration of Partnership on AI, as well as open source projects, tech companies are working alongside all varieties of engineers and creative technologists to advance the field.
So, what’s stopping us?
The first challenge we face in is that models, which contain all the information needed to classify data, are difficult to create. It takes thousands of classified input data points to create a model, and unless a client already has the right data available, it’s hard to come by. The good news however is that creating a completely new model isn’t always necessary. Using an existing system, such as the Google Vision API, may in fact be preferable from a time, cost, and features perspective. So, smart choices in the creative approach can usually cover up any shortcomings.
The second challenge is fixing bugs. ML can be a bit of a black box, with even scientists often only able to speculate at its inner workings. This means that fixing bugs is a major challenge and vastly unlike regular programming. It’s not a matter of simply changing a few lines of code, but requires engaging in a whole new cycle of trial, error and discovery to develop the right model.
The last challenge is that we need to learn how to design with ML in mind. While the success rates for output can be impressive, it’s not perfect. For example, human lipreading has a tested success rate of around 20%, while ML has a 50% success rate. So, if we want to build a campaign around this, we need to come up with a solution for the remaining 50%.
With machine learning we need to take the user on a journey and help them appreciate what’s going on. It shouldn’t be a binary experience if it’s to be impactful, and there needs to be room for error built into the concept. So, to solve this we have to be creative with the tools we have, and come up with smart fallback solutions which don’t break user engagement.
For marketers, ML can be integrated with a multitude of touchpoints, including Social, DOOH screens, platforms and campaign sites, and can play a powerful role that influences the consumer decision-making journey.
Some possibilities include using object recognition so a brand can recognise types of products from competitors and show the user an equivalent product in its range. Or, using voice recognition to develop conversational UI or voice controlled websites.
For digital agencies opportunities lie in creatively implementing available applied Machine Learning solutions, such as the Google Vision API, TensorFlow, IBM Watson, Microsoft Cognitive Services, and API.ai,. We’re seeing tremendous potential for clients and users, and are particularly excited about:
- Object recognition: Google Vision, Cloud Video Intelligence, segmentation of images, helping the blind see photos.
- Voice recognition: home automation systems, websites and mobile apps.
- Language processing: Natural Language Understanding and emotion detection in text.
- Gesture recognition.
- Manipulating Neural Networks: Deep Dream, Google Arts & Culture Experiments.
Through trial, error and deployment, here at MediaMonks we’re working with lead researchers to explore ML’s potential and how we can build solutions at scale. We’re quickly moving from a position of “we think this is possible” to “we know this is possible”.
Can we build serendipitous experiences through the use of predictive algorithms? Can we recognize fashion trends based on Instagram and what people are wearing in the street? Can we create awareness by detecting cyberbullying? Can we recognize endangered species being sold at a market?
Client briefs that were once out of reach are suddenly becoming a reality and we are excited. Machine Learning will enable us – and others – to craft the most captivating digital work seen in years. Just watch this space…