How do we balance the opportunities and threats presented by AI?

Analyst firm Omdia recently hosted an afternoon of presentations and discussion of the current state of play with artificial intelligence.

While AI as a concept has been around for almost 70 years, the latest developments around generative AI in particular have revved up the hype cycle once more. Almost uniquely with this technology, any hype is matched by an equal amount of panic about its potential negative consequences. While some of this may be Hollywood-induced hysteria at this point, there are plenty of legitimate immediate concerns.

The event started with a look at how AI is affecting the TV and video industry. Omdia Analyst Eden Zoller highlighted a few contemporary use-cases, which include automated metadata tagging and interactive video such as attempted by Netflix through its Black Mirror episode – Bandersnatch. Zoller warned, however, of the risk of bias in generative AI, given that it’s output is a product of its training and source material.

There followed a conversation between Omdia Analyst Maria Rua Aguete and Jonathan Broughton, who is head of strategy for the European Broadcasting Union which produces, among other things, the Eurovision Song Contest. Broughton revealed that AI already has a profound and growing influence over what they do, especially in the area of transcriptions and translations. The EBU even has a tool for translating voice in real time and, as an illustraton of where media AI technology is headed, he flagged up an AI-generated radio station platform called RadioGPT.

In the subsequent discussion, the matter of AI bias was once again raised, specifically with respect to demographic representation. While iit’s reasonable to monitor the outputs of generative AI models and seek to fine-tune them for accuracy, the prospect of manipulating outcomes for ideological reasons risks creating more problems than it solves. It was agreed that rigorous regulation is called for but AI seems like a classic example of an industry in which regulation will always significantly lag innovation.

Next we heard from Omdia Analyst James Crawshaw, who looked at the history and mechanics of generative AI. He revealed that the main data set used by ChatGPT is Common Crawl and pointed to a couple of sources for anyone wanting a deep dive into how generative AI works: an academic paper called Attention Is All You Need and a public educator called Andrej Karpathy, who recommends that, due to its limitations, AI only be used in low-risk environments. Crawshaw concluded by stressing that AI is just a tool, not a replacement for people, and that fully open models, such as Meta’s LLaMa, may eventually out-compete proprietary ones.

The theme of keeping AI in context was picked up by the final speaker, Omdia Analyst Camille Mendler. “AI is destined for banal invisibility,” she said, meaning most applications of AI will take place in the background, simply solving first-world automation problems and generally making life a bit easier. While there is widespread concern about human jobs being made redundant by AI, Mendler insisted that the hope is it will enable us to refocus on higher matters.

One application of AI that will require significant assistance from the telecoms industry is augmented reality which, to become truly useful, will require ubiquitous, robust and plentiful connectivity, as well as affordable devices. On the whole, among industry segments, telecoms seems to be roughly in the middle when it comes to uptake of AI.

Overall this event was a thought-provoking look at the current state of play with AI, offering a balanced summary of the strengths, weaknesses, opportunities, and threats it presents. Much public discussion of AI has become polarised between unconditional cheerleaders and apocalyptic doom-mongers but, as ever, nuance is called for. It’s clear that AI has the potential to solve many contemporary problems but there are real dangers associated with its unconstrained proliferation. How to mitigate those dangers is a major challenge of our times.

UPDATE – 10:00 13/7/23: We mistakenly identified the AI-generated radio station mentioned by Jonathan Broughton as RadioGPT when he was instead referring to Couleur 3. Here’s a French press release about it: https://www.rts.ch/entreprise/espace-pro/communiques-de-presse/13976724-intelligence-artificielle-sur-rtscouleur-3-rendeznous-nos-humains.html 

Get the latest news straight to your inbox. Register for the Telecoms.com newsletter here.