41 ROUND TABLE Machine Learning
BEN DAVENPORT: Even with machines doing the heavy lifting, someone – and someone with domain expertise – needs to train the machine. This is certainly the case with machine learning in media, so while skill sets might need to change, humans will still be needed. Additionally, video consumption is still on the rise and a lot of where we see AI being implemented is about gaining efficiency, meeting demand and protecting interests – through compliance, for example – and not just about cutting costs. TONY JONES: Machine learning’s real benefit is determining difficult-to-discern patterns among complex data sets. This can free up expert resources to focus on more valuable activities, increasing productivity. This allows organisations to deliver more value and high-quality content, helping to make the businesses more successful in terms of increasing subscribers/viewers and brand reputation.
Machine learning has the ability to unlock serious strategic and monetisation potential. As with most technological advancements, this means human operators can be reutilised in other ways in an organisation, and further efficiencies achieved when delivering high-value content and services to consumers. ARASH PENDARI: Definitely, but the question is not that simple. It will also create jobs. It will push us to educate ourselves better and adapt to a new future. Getting a coder ’s licence to work as a software engineer behind the scenes of self-driving cars could become as attainable as it is today to get a driver ’s
FEED: THERE IS A LOT OF TALK ABOUT HOWMACHINE LEARNING CAN REPLACE WORK DONE BY HUMANS. IS MACHINE LEARNING GOING TO COST JOBS?
licence to drive a taxi. So it will cost jobs, but I believe it can push humanity to a more educated future.
FEED: MACHINE LEARNING HAS THE POTENTIAL TO CREATE CHANGE AT A LARGE SCALE. HOW CAN SUCH A POWERFUL TOOL BE SAFEGUARDED AGAINST ABUSE OR HUMAN ERROR?
BEN DAVENPORT: As with any new technology, the capabilities of machine learning have the potential to be misunderstood by those implementing it, and also exploited. The important thing is for any organisation to understand what the technology can do for them, and others, and have a strategy – which tools to implement, why and how. We see a lot about the ‘cool’ things machine-learning applications can do, like creating football highlights, but there are a number of applications around protecting interests. One interesting example is detecting deep fakes – as machine learning improves, these are becoming harder for humans to detect, but machine learning can be employed to detect them. TONY JONES: Machine learning is an architectural technique that’s nothing without being trained. With suitable training, however, it can be useful for solving particular tasks that correspond to that training. The important part is to ensure appropriate training methodologies, and that the data sets used are appropriate and unbiased. One of the real dangers is that the training data sets, if they contain data
affected by human decisions, could themselves be biased. Any training that uses such a data set will inevitably reflect that same bias and the resulting ML processing will replicate those biases when processing new data. It is therefore important that the data used for training is carefully screened, to ensure it doesn’t cause a perpetuation of existing societal biases. In the case of image processing for HD-to-UHD conversion for entertainment purposes, these types of issues don’t exist. But in other environments – for example, in law and order environments – there could easily be such biases, so great care is needed to ensure the data set is appropriate and doesn’t lead to existing biases being perpetuated. ARASH PENDARI: I like Elon Musk’s initiative with Microsoft OpenAI. I would say it’s going to be very hard to control every aspect of it, so regulations are definitely needed. But just as with the tough privacy questions around the internet, the regulations around machine learning will need to adapt and change form. No matter what regulations we set today, we will need to change them as the technology evolves.
feedzinesocial feedmagazine.tv
Powered by FlippingBook