It’s doubtful that anyone working on early iterations of the world wide web could have predicted that more than half the world’s population would eventually use internet-enabled mobile devices to share selfies on social media. Though this unforeseen use case was catalysed by recognisably groundbreaking technology, the conjoined rise of social media and wireless internet highlights how technological change is generally non-deterministic. In this respect, AI is a case in point.
As of the 2020s, most, if not all, AI use case predictions have been proven wrong. We have yet to see the advent of either the narrowly functional “expert systems” imagined by early AI pioneers or the all-knowing and ultimately malicious AI devices predicted in films like “2001: A Space Odyssey.” Instead, as AI capabilities evolve beyond just novelty functions and the technology “gets real,” AI is taking a different, but no less transformative, developmental path. Counterintuitively, as today’s developmental trends continue, AI will simultaneously proliferate, transform our world, and fade from popular consciousness.
In the Next Decade, On-Device AI Will Become an Invisible Enabler
The future development of real-world AI is likely to mirror present-day developments in digital technology. Services such as AWS have already seamlessly integrated technologies like machine learning and natural language processing to help businesses forecast demand, launch targeted promotions, and analyse documents rapidly. As a result, consumers now take for granted the advanced personalisation offered by their favourite brands, social media platforms, and streaming services.
When physical devices are able to achieve a similar level of seamless integration to what is currently possible in the digital space, the result will be profound economic change. Ran independently within devices themselves, edge AI will allow consumers and businesses to leverage advanced deep learning capabilities regardless of limitations like network connectivity or bandwidth restrictions. Even in sectors traditionally resistant to automation like agriculture, fleets of intelligent IoT devices such as drones and smart sensors will unlock new efficiencies as devices recognise crop patterns and optimise cultivation themselves.
Embedded within the devices people use on a daily basis, ubiquitous AI will also redefine the functional relationship between devices and humans. For businesses and consumers alike, intelligent devices will bring the same level of personalisation currently offered in the digital world into real-life situations. No sector will be immune from the transformative effect of this development. Even in educational environments, AI-enriched learning devices will customise learning scenarios for children and assist teachers in grading.
Ubiquitous Real World AI Is a Prescient Challenge
For forward-looking organisations focused on meeting consumer and business expectations around smart device capability, enabling seamless AI-powered capabilities in devices is likely to be an immediate concern. Unfortunately, as any organisation trying to deploy large numbers of devices with the kind of AI capabilities consumers expect will have already realised, doing so means overcoming significant technological challenges.
On one level, running advanced AI algorithms onboard devices themselves means coming up against a hard limit in terms of processing ability and power supply. Because edge devices like smartphones, cars, and drones are limited by finite physical hardware constraints, running high spec AI-ready computing is often incompatible with practicality. Ultimately, these limitations mean that powering advanced AI pits essential specifications like battery life, processing power, and design at odds both with each other and a product’s business case.
Another unavoidable facet of real-world operation is that critical sensors, such as cameras and microphones, inevitably get damaged or obscured during day-to-day use. As a result, models trained on perfect information, such as pure audio, are likely to dramatically underperform in situations when the microphone gets muffled or a critical instruction is said in a noisy room.
Aside from hardware constraints, fundamental limitations within current AI machine learning models and algorithms can further complicate real-world deployment. While a well-trained deep learning model may seem perfectly fit for purpose in a controlled environment, even the most advanced models are surprisingly fragile and lack what we would call “common sense.” As a result, the infinitive variability of the real world can create unforeseen negative outcomes — situations that can easily become catastrophic when combined with human overconfidence.
Mitigating these and other inherent limitations within deep learning algorithms means constantly monitoring and updating models. Critically, however, this solution requires the kind of secure high bandwidth network connectivity that is often absent outside the laboratory.
Final Thoughts
Unlocking ubiquitous AI undoubtedly poses significant challenges. Fortunately, by networking AI devices and enabling parallel learning to occur, many of the barriers to large-scale ubiquitous AI are surmountable.
As a result, developments currently in progress have the potential to transform AI into an essential utility far more rapidly than currently envisioned. With advances in device intelligence made ubiquitous and highly capable, real-world AI will change our world just as profoundly as wireless internet did during the previous decade. In doing so, AI use cases in 2031 will likely look as strange to us as TikTok would to a 1990’s computer scientist.
