Image Credit: Panumas Nikhomkhai

Over the past two weeks, technology companies have taken a massive hit in valuation, with the market performance of said companies missing expectations. Despite initial interest in the capabilities of generative AI across multiple verticals and business sectors, advocates and adopters have gone on to form a lackluster impression of its actual impact. As AV seeks to take advantage of AI's potential (some advocates even going so far as to call it "inevitable"), the shortcomings of AI may bring with it a reality check.

As a caveat, AV has a number of use cases and implementations that take full advantage of how AI and machine learning operate. Machine learning works by constantly taking in data, forming a method of operation based on the patterns learned. This is greatly beneficial for use applications where the variables are limited, like optimizing a conference room's input and output; the only thing that changes is the people in the room. Similarly, machine learning also benefits everyday technology, such as Active Noise-Cancelling, or ANC. While ANC technology ingests many variables in the form of different sounds, the output is limited because it's only trying to counter these sounds. In these limited applications, AI and machine learning can and do provide a significant benefit.

How Generative AI (Sometimes) Works

The issue lies with generative AI, how it's been explained, and the reality behind what generative AI is actually capable of. For many, generative AI has been sold as an ever-learning, ever-refining machine that is steadily becoming smart enough to act in the stead of real human input. However, generative AI has a significant amount of issues, from both from a business and technical perspective. Generative AI is trained on data models, known as Large Language Models (or LLMs). Because the data that goes into these models is often unverified, the information shared can face accuracy concerns, which makes it more prone to give a wrong answer as more complex prompts and use cases are attempted. Additionally, when attempting to emulate human behavior, these tools can act erratically, or are easily manipulated by user inputs.

Increases in AI's usage also has the effect of exacerbating climate issues, as data centers require huge amounts of power and water to operate

AI's Business Hurdles

The chief reason behind the massive tech shortfall is the lack of spending returns on AI. Enterprise data processing has been significantly increasing in price, with LLMs, such as ChatGPT, passing these costs down to the businesses that utilize them. As products or services come to rely on these models, this also means that these cost increases can become traps of overhead expenditure. The cost increase produces a cascade effect that threatens the longevity of a software-as-a-service product; if a company has to end support because of prohibitive maintenance expenses, that ultimately damages trust and reputation with that company.

Finally, generative AI's legal considerations also pose a massive question mark. Businesses still have yet to grapple with the implications of AI's capabilities and pitfalls, leading some businesses (including AV companies) to request the ability to disable AI features while they determine what they do and do not control. More broadly, the training data for LLMs is often sourced from the internet at large, with no consideration for copyright law or legal consent for usage. In addition, if data security is not managed properly, LLMs can draw on sensitive data, presenting major confidentiality risks for companies.

Generative AI is an excellent example of not putting the cart before the horse, in terms of capability. While many marvel at the seemingly effortless nature of how generative AI produces a response, it's important to assess the actual impact when testing implementations, and whether or not that impact is worth the looming price over its head. As mentioned previously, machine learning very much has a place in AV and still makes a strong case for the ways in which it currently benefits AV, but with proven results to justify its use.