AI has been overhyped, but offers good opportunities

With more and more data being used, consumed and transferred in both the telco and media space, it is important that providers find more efficient ways to process it all.

Data is seen as a key way of monetising the increased demand for content. Every million TV households will have generated 0.5TB of data per month in the set-top box era according to figures from Viaccess-Orca, which provides OTT and TV platform solutions. Add in augmented, virtual and mixed reality and this skyrockets to a possible 500TB of data per million households, per month.

Processing that much data is no small feat, and given that much of it will pass through IP, the relationship between broadcasters and telecoms companies will play a key role.

• Read more: How AI is reinventing Visual Effects

For Orange – the former state-owned French telco – AI and machine learning are already key areas of focus.

“I’ve been learning AI for many years myself and it isn’t a brand new technology. In the past, it has been overpromised,” explains Patrice Slupowski, SVP of digital innovation at the telco.

Slupowski, who is a speaker at this year’s IBC conference, spoke to IBC365 ahead of the event. He explains that because we are using “more and more data” networks need to become more efficient.

“Our network management is using more and more AI. We’re consuming data but need a good experience.” Patrice Slupowski, Orange

“The first thing is we’re using more and more data,” he adds. “Everywhere we have data and AI internalised data. All of the things around networks being more efficient in terms of network management and seeing a good service everywhere requires AI.”

Orange is trialling, testing, and developing a number of solutions around AI aimed at making sure the end user’s experience is better if, say, they are consuming high definition video.

“Our network management is using more and more AI. We’re consuming data but need a good experience. We’re also trying to use AI to enhance the experience of the customers.”

Djingo Unchained
Another use case he points to is digital assistants. The French telco built a digital assistant, Djingo, in partnership with Deutsche Telekom. It provides access to Orange services such as making hands-free calls at home, controlling Orange TV as well as other connected home services.

Deutsche Telekom also has a separate solution, called Magenta, based on the same platform. Both are smart speakers – think Amazon Echo for reference.

• Read more: Winning audience engagement with artificial intelligence

“What we’ve been doing inside some chatbot interfaces through our service Djingo, which is a digital assistant available on several devices and interfaces,” adds Slupowski. “It is a way of driving customer interaction by being available to our customers 24 hours a day 7 days a week. It gives us the ability to use to a human interface whenever it is needed. With sport, most of our customers can raise frequent questions through the chatbot and get answers as needed from a human instead.”

Slupowski acknowledges one of the key challenges around AI is that in the past, the technology has been overpromised, partially down to confusion as to what it can actually do or what it actually is. Machine learning, AI and automation are often confused, yet all serve different functions.

“What we’re seeing is there are a number of types of problems that can be handled by AI but other problems cannot,” he says. Things that cannot easily be analysed by the human brain, “such as finding the correlation of a needle in a haystack” is one example of AI functionality, he adds.

He names three areas where AI can be efficiently applied:

• natural language processing
  -> assistants, chatbots
  
  
• computer vision 
  -> radiology, self-driving cars, identity check
  
  
• data pattern and correlation identification 
  -> smart data, fast data, trading
  
  
• games with well-known rules 
  -> chess, go

“We are focussed on the first three domains and have seen interesting results. It doesn’t mean that it is easy of efficient every time. We’re in a domain now where expectations can be met by reality. But the problem is there are high expectations around AI.”

A lot of the solutions that are being talked about, he explains, such as AI that can create a painting or judge who you should hire, create “high expectations and that is challenging”.

He concludes: “A lot of AI systems have been around for over 30 years and they have been disappointing to the end users. We have a strong idea of what is workable and what is not. We are looking at AI and while people act like there is magic behind it, it is more about having the right data and the right techniques, to find tangible results which are always being tuned to bring it up to an acceptable level.”