Autor: Redaktion

Artificial Intelligence is For Real (2 / 3)

No, AI isn´t just a hype.

The Approach 

For the purposes of this DPP AT HOME session, the content supply chain was divided into a number of key stages, which were then grouped into three phases: 

The Content Supply Chain

It may seem odd that commissioning was placed in the final stage, but commissioning decisions are informed by data derived from the Phase 3 stages. In the increasingly iterative, and decreasingly linear, world of content production, commissioning can be seen as the key stage that closes the production circle and initiates the next piece of work. 
When considering each of the stages above, our group of experts first captured the potential for AI to impact that stage. The list of actual and potential functions they identified will not be definitive of course, but they will provide a good indication of the kinds of ways in which AI is being, or could be, deployed. They also attached a benefits score around this potential. The benefits were categorised as: 
saving cost and/or time 
the potential for creative and/or business benefits by delivering new or better services or output 
innovation or improvement in consumer experience to an extent noticeable to the consumer 
The scores ranged from 1 (non-existent or negligible) to 5 (game changing).  
Once our experts had considered the potential for AI to impact the content supply chain, they then documented the actual AI-led activity that is occurring today in each of the stages. Activity was defined as a deployed implementation – even if still in beta. Research and development activity and closed pilots were taken into account when considering potential; but not classified as actual use cases today.  
Once again, our experts will not have created an exhaustive record of current use cases, but what they have discussed can be taken as indicative of current levels of activity. 
These present day implementations were then assigned a maturity score of 1 to 5, on the following basis. 
1: No evidence of any capability 
Evidence of: 
2: Some capability, but not yet used fully 
3: Marginal benefits  
4: Implementations replacing human activity  
5: Implementations that are transformational 
As will be evident, this maturity scoring was weighted towards being able to differentiate levels of maturity in emerging applications. The leap between 4 and 5 – from the replacement of some human activity to the application of the technology in ways that are truly transformational – is considerably greater than the jump between any of the other steps. 
As can also be seen, there is no assumption in the scoring that any implementations are universal or widespread, since if they were, the conversation about AI would be at a very different stage.  
Finally, the experts considered the overall challenges to achieving greater maturity of AI across the content supply chain and formed some concluding remarks. 

The Discussion 

Our group of experts were set the task of creating an audit of both current and potential applications of AI, across the whole content supply chain, and for all content types.  
This was a considerable task to achieve within a single session. The results, which are summarised in the pages that follow, cannot claim, therefore, to be definitive. Some readers may be able to identify some current or future applications that have not been included here.  
Nonetheless, this rapid-fire task achieved its principal purpose: 
  • it identified key distinctions in the relevance of AI for different content types 
  • it identified where and how AI will be most effectively applied at different stages in the content supply chain  
  • it created a benefits map which in turn is likely to become an implementation map for AI, since application follows need  
  • it created a picture of the gap between the ‘as is’ and the ‘could be’ for AI in media In short, the discussion created perhaps the clearest business development route map for AI yet presented in the media sector. 

Phase 1: Create 


The starting point in the creation of content, once it has been commissioned by a client, takes place within a production company or agency. Often such companies are small and don’t have access to sophisticated technology of their own. Increasingly however they are able to buy web-based services. A number of such services are now emerging, and AI is beginning to be applied as a service in acquisition and post production. But the planning process itself still tends to be very manual. This means that most of the planning knowledge remains in the heads of expert individuals rather than in machines – in the case of long form production at least.  
In fast turnaround production areas – notably news and sport – the planning process happens in near real time, and in a far more technical environment.  
So in any discussion about planning it becomes important to separate live or near-live production from other forms of output such as scripted content and documentaries. 


The production process is complex. It typically involves the deployment of a wide range of different resources and people, sometimes over long periods of time. Although every production has its idiosyncrasies, most of the processes are nonetheless predictable and repeated from production to production. If data relating to the utilisation of people and resources was to be captured, and modelled against different production types, it would generate highly valuable intelligence that would improve budgeting, cost control, reporting and future planning. It would assist greater profitability. 
Resource scheduling intelligence can be separated from wider business intelligence (BI) that can be gained from the performance of particular content, and the opportunities for the generation of further content. BI for content that isn’t fast turnaround – such as scripted and documentary – often sits with the commissioning entity, unless the producer is creating web native content. From the point of view of the production entity, therefore, the potential of such BI is limited. 
BI for fast turnaround content is very different however. A news operation, for example, could constantly reassess how a story is playing out; for which consumers; and on which platforms. Decisions can be made about areas of focus, new content generation, use of platforms, amplification, and so on, based upon the real time analysis of data coming back into the newsroom. 


This use of BI in fast turnaround production environments already exists, to a more or less sophisticated degree. In the case of online news providers such intelligence can be very sophisticated indeed, including A/B testing of particular story edits. However our experts struggled to think of AI-led business intelligence generated in longer term production environments. And although there are now a number of increasingly sophisticated and well-integrated resource management tools, none of our experts was aware of one that is AI-led. 

Planning SCORES 
[size=2]Potential [/size][size=2]AI Type [/size]
[size=2]Business Intelligence: real time [/size][size=2]B2 [/size]
[size=2]Business Intelligence: long term     [/size][size=2]B2 [/size]
[size=2]Resource scheduling [/size][size=2]B2 [/size]

[size=2]Benefits [/size]
[size=2]Efficiency [/size][size=2]3 [/size]
[size=2]Opportunity      [/size][size=2]3 [/size]
[size=2]Experience [/size][size=2]3 [/size]
[size=2]Total [/size][size=2]9 [/size]
[color=white][size=2] [/size][/color]
[size=2]Actual [/size][size=2]AI Type [/size]
[size=2]Business Intelligence: real time     
in news and sport
[size=2]B1 [/size]
[size=2]MATURITY SCORE [/size][size=2]2 [/size]

Content Acquisition /Ingest and Logging 

In considering the content acquisition and ingest logging processes, our experts could not draw a meaningful separation between the two stages. Ingest and logging of fast turn around content is integral to the content creation process: to log a piece of content in sport, for example, is also to prepare it for highlight generation.  
So for the purposes of this discussion, both processes were considered together. 


Content acquisition in fast turn around environments is so fundamentally different from longer term content making that they seem almost like different industries.  
Longer term content making is fragmented, complex and occurs across multiple locations. Much of the act of gathering the raw material is oddly physical. It is difficult to apply technology to it (apart from capture technology) until the moment that raw material is made available to be shaped.  
Fast turn around content, on the other hand, is shaped in near industrial, highly technical, and far more predictable environments. In many ways it is well suited to the application of AI. 


There is a common theme to the opportunities around content acquisition: they tend to be focused on fast and efficient ways to select from large amounts of content, and to deliver that content to consumers. 
Sport represents the greatest opportunity of all. The action takes place in standardised, predictable environments, and is given further predictability by the rules of the sport. This can enable automatic generation of clips, highlights and graphics. The application of AI and commoditised technology is also enabling the high quality coverage of nonpremium sports – notably college sport in the US. 
The use of AI in consumer technology is already leading to the auto generation by photo and social media platforms of video packages with music and graphics, and this capability is likely to come to professional content also. 
The creative opportunities to be derived from AI are significant, since automation is enabling the creation of content that otherwise wouldn’t exist. But the real benefit is in efficiency. Historically, fast turnaround, multi-camera content has required an enormous number of people. Robotic cameras, automated logging and automated clip selection offer the potential to create the same, or more, content with far fewer people. 


Many of the functions described by our experts as having potential, already exist, at least in an embryonic form. In some cases they are commonly available but the algorithms are not yet well trained enough to deliver the accuracy required for high end use. But the business benefit is clear, so there is good reason to expect there will be wider adoption in the next few years. 

Ingest and Logging SCORES 
[size=2]Potential [/size][size=2]AI Type [/size]
[size=2]Face detection  [/size][size=2]B2 [/size]
[size=2]Shot classification      [/size][size=2]B2 [/size]
[size=2]Speech to text [/size][size=2]B2 [/size]

[size=2]Benefits [/size]
[size=2]Efficiency [/size][size=2]4 [/size]
[size=2]Opportunity      [/size][size=2]2 [/size]
[size=2]Experience [/size][size=2]2 [/size]
[size=2]Total [/size][size=2]8 [/size]
[color=white][size=2] [/size][/color]
[size=2]Actual [/size][size=2]AI Type [/size]
[size=2]Highlights clipped from live and sent to social  [/size][size=2]B1 [/size]
[size=2]Sports commentary auto transcribed to subtitles      [/size][size=2]B2 [/size]
[size=2]Auto clip generation for sport [/size][size=2]B1 [/size]
[size=2]MATURITY SCORE [/size][size=2]3 [/size]

Post Production 


Much of what applies to content acquisition, ingest and logging applies also to post production.  
Fast turnaround environments, where production and post production are difficult to separate, are most susceptible to the application of AI. A case can be made on paper for how AI could drive the whole process of scripted production, from idea to auto assembly of the rushes, but the reality is far more difficult to implement. AI is more likely to impact individual processes such as rushes selection, image enhancement and visual effects than to transform the whole story making process. 


Opportunities were identified around premium production. One example was the selection of best takes for ultra high definition scripted content. Such files are so big that there is simply not enough time to transfer all of a day’s rushes to the cloud. AI can help select the most important takes, and prioritise them for transfer. 
There is also opportunity around the application of AI for the incorporation of product placement into shot footage. The same kind of technology can be used to make other changes to the shot image.  
But it was striking that some of the greatest opportunity was considered to be around the original generation of content, such as the generation of ‘B roll’ from existing archive, or the analysis of existing social media content to identify storylines. There is also opportunity to make existing content more usable by generating or repairing frames. 


Once again, many of the opportunities already exist – but are not widely applied. The most developed applications are probably occurring in specialist fields such as archive enhancement, or virtual product placement. Perhaps more significantly, major edit platforms are now able to integrate with AI tools. So even if AI-led activity is not widespread, the platforms are being developed to enable it to increase. 

Post Production SCORES 
[size=2]Potential [/size][size=2]AI Type [/size]
[size=2]Content enhancement  [/size][size=2]B2 [/size]
[size=2]Auto graphics templates  [/size][size=2]A1 [/size]
[size=2]Digital effects  [/size][size=2]B2 [/size]
[size=2]Best shot identification  [/size][size=2]B2 [/size]
[size=2]Auto assembly  [/size][size=2]B2 [/size]
[size=2]Auto audio editing  [/size][size=2]B2 [/size]
[size=2]Archive recommendation  [/size][size=2]B2 [/size]
[size=2]Intelligent product placement and replacement       [/size][size=2]B2 [/size]
[size=2]Intelligent content reframing [/size][size=2]B2 [/size]
[color=white][size=2] [/size][/color]
[size=2]Benefits [/size]
[size=2]Efficiency [/size][size=2]4 [/size]
[size=2]Opportunity      [/size][size=2]3 [/size]
[size=2]Experience [/size][size=2]4 [/size]
[size=2]Total [/size][size=2]11 [/size]
[color=white][size=2] [/size][/color]
[size=2]Actual [/size][size=2]AI Type [/size]
[size=2]Auto editing of bumpers for different endpoints        [/size][size=2]B1 [/size]
[size=2]Phonetic search  [/size][size=2]B1 [/size]
[size=2]Edit platform AI integration [/size][size=2]B1 [/size]
[size=2]Enhancement of poor quality frames  [/size][size=2]B1 [/size]
[size=2]Generation of missing frames in archive  [/size][size=2]B1 [/size]
[size=2]Intelligent product placement and replacement  [/size][size=2]B1 [/size]
[size=2]Intelligent content reframing [/size][size=2]B1 [/size]
[size=2]MATURITY SCORE [/size][size=2]3 [/size]

Related Articles: