The analogy of artificial intelligence (AI) is seasonal; it’s known as the AI summer when it’s at its height and AI winter when it’s at its trough. The past decade was probably the hottest summer that AI had ever experienced as we witnessed a boom in related startups; relentless research and testing of various domains, as well as active discussions on how the technology should be implemented. Nevertheless, some experts feel there are recent signs of slowing down, as we gradually realized certain capabilities of AI may have been overestimated and it’s time for a new round of re-evaluation. Since the support for AI has yet to cool off, it will be more appropriate to call this period of time an “AI autumn”.
A mismatch between expectations and reality
The Lighthill Report might have ignited some controversies when it was published in 1973 but an obvious mismatch between expectation and reality was perhaps the main reason leading to the first AI winter. Things have obviously improved over the years, especially with the availability of big data. AI can now outperform human being at strategy games; diagnose or predict diseases with accuracies mimicking a real professional, generate hyper-realistic images and videos, reply conversationally and so on. However, every time when we expect AI to perform more, the technology will hit on its own challenges.
A self-driving car can be easily fooled by tainted road signs, machine learning systems may suffer from “catastrophic forgetting”, and AI fails to understand causation; this means they have to be laboriously taught for each and every new task rather than learning them through logic and common sense. Unless AI masters causality, some scientists believe artificial general intelligence (AGI) will remain in the same dream that we have conjured a decade ago. Others, such as Edmond Banayan, Chairman, Los Angeles Venture Association (LAVA); Healthcare Founder and Chief Executive Officer of startup ventures such as Chronaly Inc., has a different thought.
As Banayan told AIMed in a recent interview, “AI is an enabling technology where a single innovation can unlock solutions to hundreds, if not thousands, of problems. Such breakthroughs don’t happen often or regularly. They can take time and can depend on many other chips falling in place”. Banayan suggested one should look at the “timeliness” of a solution or how timely one’s innovation is relative to the market, industry, and environment. AI is not the only technology that needs perfection. “For AI breakthroughs to occur at a higher batting average, greater focus should be placed on this formula,” he adds.
Insufficient communications between developers and users
As mentioned, as startups compete for funding and attention, they may have peripherally prevented the AI industry from entering a possible winter. At the same time, these new companies may have unintentionally introduced confounds which prevented the industry from moving forward. In medicine and healthcare, insufficient communications between developers and users makes it harder to establish trust. Some medical professionals thought AI startups do not fully understand the nature of their work, thus, the proposed solutions may not fully address the challenges they face.
With that, Edmond Banayan has a different thought. He believes competition is good for users. “When an industry lacks a competitive spirit, it’s reasonable to conclude that innovation suffers. To the extent these startups are ‘staging a war’ among themselves without crossing the lines of integrity and ethics, I’m a proponent of it,” he says. In terms of trust between developers and users, Banayan thought it’s due to entrepreneurs and innovators not performing adequate user research and analyses to discover relevant user needs and insight.
“We call it ‘consumer insight’ in the marketing world. Innovators need to conduct sufficient user research (i.e., surveys, heuristic evaluation, unobstructed observations, focus groups) to derive user insight – before product development and launch. Most importantly, the products or solutions we develop needs to be user-centric,” Banayan explains.
Part of an ongoing cycle
Ultimately, Edmond Banayan feels AI does not have to be flawless to come up with a compelling solution that solves a particular problem or need. As he wrote in a recent blog article, “AI adoption is on the rise, going from 48% in 2018 to 72% in 2019. In healthcare alone, 68% of the executives reported their businesses are currently utilizing AI, an increase from 46% in 2018”.
Banayan continues to share in that “AI will remain relatively hot in the next few years. By 2023, it’s expected more than $98 billion will be spent on AI and AI systems, which is more than two-and-a-half times the #37+ billion that will be spend in 2019. And when you consider AI and its integration for healthcare, AI in healthcare soared in 2019 with investors pouring $4 billion-plus into the sector across 367 deals – which is a significant increase from 2018 where $2 billion-plus was poured into the sector across 264 deals”.
As such, even if the AI hype is wearing off, it may be part of an ongoing cycle. As explained by the Gartner Hype Cycle, once technology hits the “trough of disillusionment”, it will mount onto a “slope of enlightenment” before arriving at a plateau eventually. Hopefully by then, AI will not be what we come to know as today.