Artificial Intelligence is increasingly relevant to broadcast and streaming production, but it helps to select the most appropriate solution to solve a production problem. Here we explore the taxonomy of the different kinds of AI to discover their inner workings and practical applications.

Being circumspect in how you use AI can be misinterpreted as having a negative attitude towards it, but being cautious is fundamentally important because using AI is not entirely risk free. With every advanced tool comes a mandate to use it responsibly.

In a simplistic way, AI embodies any kind of pseudo intelligent behavior. The word pseudo is highlighted because this is not real intelligence in the human sense – it is still a rule based deterministic algorithm. Until AI evolves to modify its own code, that algorithm is still designed by human programmers.

AI systems cannot create new content but they can appear to create new and unique text and data by aggregating the stimuli that they were trained on. This is not the same kind of creativity that humans are capable of. There is nothing fundamentally original happening although unrelated ideas may be combined in novel ways to appear to be creative.

A Timeline

In common with most overnight successes, Artificial Intelligence has been around for some time without being noticed until recently. Here is a snapshot of some important milestones:

New Terminology To Assimilate

AI developers use a glossary of arcane terms to describe what they do as a form of condensed language. It sounds complex but as usual, there is nothing to fear:

Applying AI Tools To Production

Notwithstanding perceived negative aspects of AI, there is a great deal of useful functionality to deploy within a broadcast infrastructure. Separate the marketing hype from the real capabilities (and limitations). Then develop a working understanding of how AI systems work.

Unless you are in the business of building them, it is not necessary to go very deeply down the rabbit hole and a quick peek under the hood equips us with useful knowledge to facilitate an effective deployment. Understanding the limitations of the technology shows where your infrastructure needs additional support and solutions to augment the AI facilities.

Used correctly and appropriately, AI tools can be of great benefit and provide significant time-saving leverage. The most obvious area that immediately benefits broadcasters is gathering of metadata to describe content in a media storage system; AI empowers search engines to use new and interesting techniques for discovering content.

AI is useful for generating ideas and starting points but if you don’t fully understand what it is doing then betting your reputation on what it thinks you want to do is a high risk. Here are some important criteria to consider:

• Know the provenance of the training data used for the AI model.
• Fully understand the output the AI system generates.
• Make sure the AI system is not creating content that exposes you to a copyright infringement liability.
• Constrain the LLM to be an SLM that has full knowledge of a small topic extent.
• Be aware that using public systems to analyze proprietary data can leak confidential and security sensitive information into the wild for anyone to obtain.
• Be mindful that the global landscape of AI and copyright law is still evolving, both in terms of the ownership of AI-generated output at one end, and the use of copyrighted materials to train AI models at the other. Expect more development of policies that attempt to balance AI content with the protection of intellectual property rights. Broadcasters and other content providers should keep abreast of changes in international and regional copyright law to avoid any legal and licensing penalties in the future.

Taxonomy Of AI Systems

Here is an example taxonomy that identifies different paradigms within the AI world based on their practical use and application. These are arranged somewhat into a sequence where they evolve in complexity and autonomy. This starts with the least sophisticated decision making capabilities up to experimental and somewhat hypothetical thinking machines.

There are other abstract ways to characterize AI but for practical applications this is the most appropriate:

Ecological & Financial Impact

At some point it will be important to consider the ecological impact of using AI systems. There are unintended consequences of deploying AI on a large scale and they show no signs of easing soon. In fact, the situation is getting worse.

A single AI prompt dissipates much more energy than a search engine query ever did. There is a significant proportion of the world’s energy generating capacity and water resources now being consumed by data centers.

The financial impact is reflected in the much higher prices being charged for memory chips and hard drives. Graphics cards have been in short supply for years. That may ease as Tensor Processing Units and Nvidia AI chips become more widely available but it is likely to be some time before memory and disk storage prices return to normal.

Relevant Standards - Where To Find Out More

As yet, there are very few standards to describe AI systems. This is bound to change as we move from a competitive approach to something more collaborative where AI data needs to be distributed and shared between systems developed by different companies.

These are important initiatives and standards:

Conclusions

Big tech companies are developing the AI technology and competing to be first to develop highly capable artificial general intelligence. Clearly, they won’t stop there and will evolve onwards to super intelligent machines: the genie is out of the bottle and up to mischief.

It is altogether exciting and frightening in equal measure but somewhere in this complex mix are the tools that broadcasters can use creatively to make better media content.

These Appendix articles contain additional information you may find useful: