In technology, buzzwords come and go. It’s easy to become cynical, especially because in recent years the public has been sold many big new concepts, all of which have promised to change how we work and play.
Yet, while we continue to wait for concepts like the Metaverse and Web3 to deliver, generative AI has swept in with a whole new wave of creative and technological innovation.
But let’s be clear: this is a milestone that goes well beyond the hype. Large language models (LLMs) are likely to revolutionize conversational AI in much the same way as neural networks transformed the field of machine learning forever—and the buzz around this technology isn’t just from those building it.
On launching ChatGPT, OpenAI went directly to consumers. As a consequence, we quickly saw the kind of public enthusiasm last experienced at the advent of the app store. Suddenly, corporate executives were learning about “the next big thing in AI” from non-experts, and ever since, businesses have been grappling with what this potent new technology could mean for them.
Having worked at the cutting edge of technology and innovation for the last two decades, I believe this is undoubtedly a period of disruption for many companies (including mine)—but the potential for these new AI capabilities is huge.
Just as with the app store, we will see startups emerging with until-now-unseen business models, fueled by generative AI. And although we don’t yet know what the new class of Whats Apps and Instagrams will look like, I can offer my thoughts to business leaders looking to understand what it means for their own operations and questioning how they can ride the wave.
1. CONVERSATIONAL AI EVERYWHERE
For many years, the creation of AI-driven conversational experiences has been technically challenging and resource intensive. Now LLMs and generative AI are creating new possibilities, amounting to a revolution in the space.
We can expect conversational AI to become much more prevalent—particularly as a core business-to-consumer channel.
2. VOICE AS THE NEXT BIG INTERFACE
If people will be holding more involved conversations with LLMs, it stands to reason that we will see a much bigger emphasis on voice interfaces.
Not only is voice technology more accurate than ever, but it also has the advantage that it’s much quicker (and more natural) to speak than to type. At the same time, it is much faster to read than to listen, so we can expect to see more dynamic, voice-led multimodal systems to come to the fore.
3. A NEW OWNED CHANNEL FOR BUSINESSES
That all said, public concerns about privacy and Big Tech dominance in recent years have made it less likely that all queries will be mediated through one single domestic voice assistant. Similarly, businesses want to own customer data and the consumer experience of their brand, so will be looking to field their own conversational AI channels.
Just as we saw with websites and then apps, we can expect each business to create its own branded space where they provide conversational experiences to customers without any third-party intervention.
Although this exciting new proposition will allow brands to cultivate a more intimate relationship with their customers, there are also a number of “watch outs” to observe when experimenting with LLMs.
NO REAL-TIME INFORMATION
Though generative AI can give extremely sophisticated answers to nuanced questions—and with the right skills onboard, these can be tailored to a particular business’s products and services—on their own, LLMs fall short when it comes to anything in real time.
That means consumers cannot obtain the information they may require about the availability of a product, or current parking or traffic conditions near business premises, or any other significant detail that updates regularly.
In order to make it work in a live, real-time environment, generative AI needs to be engineered to function alongside other relevant domains. This can be done but means thinking beyond a mere API integration.
AI HALLUCINATIONS
If you try to ask generative AI real-time questions, you may get misleading—even harmful—responses that are deceptively convincing. These are known as “AI hallucinations” and are an acknowledged form of misinformation.
Usually, it’s a case of the AI system reaching for any semi-relevant answer in the absence of access to live data. For example, if you ask: “What flights are running between New York and San Francisco?” you may receive a list of airlines rather than specific flight times.
Mostly these hallucinations are innocuous, if unhelpful, but they can propagate the spread of inaccurate information that could represent a risk to any business looking to deploy generative AI.
BUILDING TRUST
Businesses will need to think carefully about their conversational AI channels before making them live. Channels need to be robust, voice-enabled, and able to deal with utility queries about real-time data as well as having generative capabilities. Otherwise, they could put the trust of loyal consumers at risk.
That’s not to put a negative spin on this exciting time. What is clear is that we’re now at a critical junction whereby this technology is good enough to give customers what they’ve wanted for years—a natural, fluent conversation across topics and domains without the fatal, “I’m sorry, I didn’t get that” that frustrates the whole experience.
It’s now time for businesses to think seriously about how they harness generative AI.
Moving too slowly could be a mistake, as could jumping in without the right guardrails or AI understanding. At SoundHound, we’re engaging with our customers on how generative AI can be deployed in a way that works safely and seamlessly. I’d advise other businesses to work alongside AI experts as a way to navigate and execute in this new and evolving world of nascent AI technology—and to remember that conversational AI can’t be done as a side project.
Keyvan Mohajer is the Co-Founder & CEO of SoundHound.