It’s unlikely that we’re going to come across any technology more consequential and necessary than AI in our lifetimes. The presence of artificial intelligence has already altered the human experience and the way technology can reshape our lives, and its trajectory of impact is simply getting wider.
With that in mind, AI innovators and leaders have spent the past quarter of a century aggregating data and advancing the models to realize the software that powers generative AI. AI represents the height of software: An amorphous tool that may reproduce tools to resolve problems across abstraction layers. Firms constructing compute empires or those acquiring LLMs to bolster their software offering are actually common sights.
So, where can we go from here?
Even with limitless compute, the gathering of deductions using all existing data will asymptotically approach the prevailing body of human knowledge. Just as humans have to experiment with the external world, the following frontier in AI lies in having the technology interact meaningfully with the physical realm to generate novel data and push the boundaries of data.
Interaction through experimentation
Exploring AI’s potential requires transcending its usage on personal computers or smartphones. Yes, these tools are prone to remain the best access points for AI technology, however it does put a limit on what the technology can achieve.
Although the execution left much to be desired, the Ray-Ban Smart Sunglasses powered by Meta’s AI system demonstrated a proof of concept in wearables infused with AI technology. These examples of hardware-first integrations are critical to constructing the familiarity and value of AI outside of a tool setting because they illustrate make these grand technological advancements seamless.
Not every experiment with AI in the true world goes to be successful, that’s precisely why they’re experiments. Nonetheless, demonstrating the potential of hardware-first AI applications broadens the spectrum of how this technology may be each useful and applicable outside of the “personal assistant” box it’s put in now.
Ultimately, firms showcasing make AI practical and bonafide can be those to generate experimental data points that you simply simply cannot get from web applications. After all, all of this requires compute and infrastructure to properly function, which necessitates a greater influx of investment in constructing out AI’s physical infrastructure.
But are AI firms ready and willing to try this?
The hardware and software dialogue
It’s easy to say that computationally intense AI applications in physical products will turn out to be the norm eventually, but making it a reality demands far more rigor. There’s only a lot resources and can available to go down the road less traveled.
What we’re seeing today is a type of short-term AI overexuberance, mirroring the standard market response to disruptive technologies poised to create recent industries. So, it’s clear why there could also be hesitancy from firms constructing AI software or dabbling in it to embark on costly and computationally intense hardware outings.
But anyone with a wider outlook can see why this is perhaps a myopic approach to innovation.
Unsurprisingly, there are plenty of comparisons made between the AI boom and the early web’s dot-com bubble, where projects focused on short-term goals did die off once it burst. But when we were to collectively write off the web due to dot-com bubble’s aftermath as an alternative of refocusing on the long-term ideas which have survived gone it, we can be nowhere near the technological landscape we’re in today. Great ideas outlast any trend.
Moreover, compute is the linchpin for any AI innovation to maintain progressing. And as any AI developer will let you know—compute is price its weight in gold. Nonetheless, that also puts a limit on what number of projects can feasibly afford to explore real-world AI applications when model development alone already eats up resources. But no company can maintain market dominance on software alone—regardless of how impressive their LLM is.
It’s comfortable for AI firms to steer with software and wait patiently for a hardware provider to swoop in and acquire or license its technology. Not only is that this severely limiting, it leaves many incredible projects on the mercy of outsiders who may never come knocking.
AI is a multi-generational technology that may only turn out to be more customized and designed for people as time progresses. Nonetheless, it’s as much as projects to reap the benefits of a mostly-even playing field software-wise to take real strides into the physical realm. Without daring experimentation, and even failure, there can be no path forward for AI technology to comprehend its full potential in improving the human experience.