Your browser is not supported. Please update it.

12 July 2018

How can media markets survive the inevitable AI bubble burst?

The global AI market will reach just $39 billion by the end of 2023, giving rise to a bubble which is destined to burst – dragging with it millions of dollars of venture capital funding into the abyss. Most AI riches will become conglomerate exclusives, while only a handful of niche companies will thrive in specialist sectors.

This contrarian view towards otherwise over-hyped AI markets comes from the AI: Show me the Money report published this week by the research arm of our Internet of Things sister publication Riot – diving into which sectors will make a killing in AI with a market forecast from 2017 to 2023.

AI in media markets will be one such niche, where many of the companies regularly featured in Faultline Online Reporter have jumped into in recent years – but where few will be successful. Other AI reports gloss over the difficulties of making money from AI, convinced that churning out enough buzzwords will be a recipe for success in itself. The research conducted by Riot may not be music to investors’ ears, but it has approached AI cautiously and in doing so has identified the size of key sector opportunities – research we think is intrinsically more valuable to a business in the long run than sugarcoated Gartner magic quadrants.

Of the $39 billion total market size by 2023, machine vision and natural language processing will be the two largest horizontals, reaching $14.1 billion and $15 billion respectively. However, the significance of these two technologies means crossovers into other vertical markets is inevitable. We have therefore picked out four specific sectors from the 68-page report of relevance to our readers – news and media, software engineering and debugging, hardware and chip design, and marketing and advertising.

The main media applications of AI, according to the report, are in identifying trends and personalizing services. Broadcasters and newspapers in particular are showing increased interest in fishing out news which has broken beyond their immediate reach, for example on social media sites. AI is also being applied to similar effect in spotting and quickly removing fake news.

A couple of companies seeing success here have been highlighted in the report, the Silicon Valley start-up Banjo, funded by Japan’s SoftBank among others, and China’s ByteDance which has raised $3.1 billion so far.

AI for advertising is an interesting area, where the current focus is on placing ads where they resonate most with a viewer – effectively replacing the job of media agencies. The report cites Volkswagen as a prime example. After the German car maker found that AI-based predictive analytics was capable of generating more car sales for a given amount of money, it swiftly recruited an AI-based system supplied by Dutch media agency Blackwood Seven.

AI-based tools are even becoming part of the user interface, changing the code itself in a way that was previously done by Quality Assurance (QA) staff. Machine learning is being used to understand input data and adjust it in response to test output, generating variations in tests much faster than humans while also turning them to the task and ensuring that as many potential exceptions or failures as possible are covered.

The report therefore warns that a reduction of QA and testing staff on the back of the emergence of AI tools looks likely – but it will not eliminate them and may actually increase the skills required to make work more interesting. This is an intriguing takeaway; we had to read that last point over a few times before registering what is not a typical standpoint on the AI versus employment debate.

Hardware isn’t often an industry painted in flattering light, yet the report highlights how the role of AI in hardware design is more fundamental than in other vertical sectors. This can be all be traced back to the discovery that GPUs for gaming, such as those developed by Nvidia, were well suited as parallel devices for machine learning algorithms that involve arithmetical operations on arrays of numbers.

This is because GPUs perform much better than general purpose CPUs on a large number of problems in the AI domain across all sectors. Nvidia, the GPU market frontrunner, has not shown any interest in developing a chip dedicated to one specific AI algorithm as this would have too small a market, while a few start-ups are taking a more radical approach.

Silicon start-ups such as SambaNova Systems and Mythic have even talked of established players being ripe for disruption, arguing that general purpose GPUs can only go so far. What does appear certain, concludes the report, is that as algorithms under the heading of AI and machine learning become incorporated in mainstream computation, more and more chips will evolve to execute them efficiently – both general purpose CPUs and GPUs.

Worryingly, we stumbled upon a report this week which found companies claiming to have AI technologies were guilty of employing people overseas to pretend to be AI – for mind-numbing exercises such as chatbots. Quite simply, human labor is cheaper and much easier than developing AI. The Wall Street Journal uncovered how Google allowed hundreds of third-party app developers to access Gmail accounts and instead of using algorithms to identify and learn from the data, it transpired that certain companies, namely Edison Software, were manually sifting through private emails – in a practice dubbed “pseudo-AI”.

For more information contact:

Chloe Spring (Marketing Manager): [email protected].

Download a copy of the executive summary

Purchase a 5-user license

Purchase a corporate license