The Future of AI

Seeing the forest for the trees, and the forests beyond The future of AI A report by the Deloitte AI Institute The question of whether a computer can think is no more interesting than the question of whether a submarine can swim. —Edsger W. Dijkstra, computer science pioneer 2 Contents 01 Everything and 03 The machines 06 Guardrails nothing at all (or, a technological 25 4 history of the future) 8 07 Conclusion 02 Something old, 29 somehow new 04 Us and the machines 6 (or, an anthropological history of the future) 14 05 Us as the machines (or, a biological history of the future) 20 3 Everything and nothing at all 01 As futurists, my team and I secretly And so, when tech headlines begin to Like any gold rush, there’s hope beneath the spend the lion’s share of our time increasingly read as breathless brochures hype. To be sure, there is plenty of actual 02 studying the past. I like to say that for Artificial Intelligence, we grey-hairs can’t gold afoot insofar as we’re seeing a genuine, we’re closet historians. Specifically, help but be reminded of a certain upstart evidence-based phase shift from AI as “cherry- 03 we research the history of various technology category some 20 years ago called on-top” curiosity to “key ingredient” at leading technologies and how they’ve impacted, “the world wide web.” organizations. 61 percent of respondents to or failed to impact, the way the world a recent Deloitte Insights report say AI will 04 works and lives. When you spend 25 The similarities are striking. Just as startups and substantially transform their industry in the next incumbents alike once appended “dot.com” to 3–5 years. Furthermore, adoption is significant years up to your eyeballs in all-things- 05 newfangled, you can’t help but start to their names to supercharge their marketplace on a per-organization basis2, with 53 percent of recognize patterns. Scripts. Déjà vu. perception (and IPO values) in the early 2000’s, those polled spending more than $20 million so too are today’s players cloaking themselves during the past year on AI tech and talent.3 06 in the “everything to everyone” halo afforded by AI. Statista reported over 2,000 AI-focused AI’s increasing centrality to business processes, 07 companies in the US alone as recently as 2018,1 and even strategy, is no longer up for debate. a number that’s further exploded over the last The “Age of With”—human work augmented and two years. enhanced with AI—is upon us. As with any exponentially accelerating emerging AI’s increasing centrality to business processes, technology, the abundance of news has, however, and even strategy, is no longer up for debate. given way to an even greater abundance of noise. 4 Everything and nothing at all 01 One time-honored means of getting past On the flipside, zoom too far out, and we’re left hyperbole is to double-click into the crunchy with Artificial Intelligence and Machine Learning 02 research itself. Zoom in too far, though, and (AI/ML) as a blanket platitude. Like “synergy,” 4 we’re blinded by a blizzard of buzzwords too “innovation,” and “leverage,” the fast and loose 03 numerous, and too arcane, to reasonably keep inclusion of “AI” into a business conversation track of. Papers with Code, an open source AI is at once obligatory, and an eye-roller. When research community founded by the Facebook every story is an AI story, none of them are. And 04 AI research team, itself hosts 37,000 published worse, some of them actually aren’t. 40 percent research papers and 3,000 datasets. A scrappy of European AI startups, upon closer inspection, 05 leader might reasonably visit such a community don’t even use AI technologies or techniques.5 to derive a coherent first-person understanding of, say, “Weakly-Supervised Action Localization The key, it would seem, is to peer beneath AI-as- 06 by Generative Attention Modeling.” Woe, though, platitude and above the a la carte jargon. To shoot to that leader if she were to try to put together a the curl towards the “Goldilocks Zone,” where 07 boardroom-relevant “So What? And Now What?” clusters of coherent, high-growth, impactful6 sub- executive summary at that altitude. movements are both easier, and more useful, to understand. In this way, we can not only better sense what’s happening, but make sense of where these movements are likely headed. When every story is an AI story, none of them are. 5 Something old, somehow new 01 AI is not new. For starters, there’s the AI Effect. Namely, the idea that as machines become more capable, 02 Founded as an academic discipline in the tasks considered to require “intelligence” are 1955, it’s practically as old as the first often removed from the definition of AI. Take 03 digital computer. As with most emerging chess, for example. A fascinating AI problem if technologies, a gradual (though anything there ever was one, at least until IBM’s Deep Blue but smooth)7 convergence of cost defeated grandmaster Garry Kasparov in 1997, 04 reductions, performance improvements, after which it was reframed as not really the realm of AI. As ever more powerful AIs racked up ever and network effects has only recently 05 conspired to make AI a boardroom- more impressive championships in Poker, Go, and relevant agenda item. What, specifically, the TV game show Jeopardy!, the expectations has changed, and why are we talking only continued to rise. 06 about so much about it right now? As such, AI has become a catch-all term for 07 “whatever computers can’t do yet.” Hence AI’s ability to continually grab our attention, despite it having been around for nearly 75 years. Less than 10 years ago, a natural language voice-assistant on our phone would be considered absolute magic. Today, the average American home has more such smart devices than people.8 6 Something old, somehow new 01 Behind this curious expectations-management results. It’s only with the recent explosion proposition increases markedly. Specifically, whiplash: Our tendency to prize intelligence as of cloud-native production datasets that organizations have begun to realize that, properly 02 a uniquely human virtue. AI’s maturation has, in organizations can efficiently build and deploy real- embedded in business processes, AIs can reduce 9 part, been a battle between our pride in creating time, point-of-need models that perform quickly their fully-loaded cost per decision. 03 something exceptional and our pride in ourselves enough to fit into existing business processes. being exceptional. Our profound need as humans Or in the plainest possible English: AIs can, of late, to believe that we’re capable of creating amazing We can liken things to the difference between save us time and money. 04 things, just so long as they’re not as amazing as C-3PO and Chewbacca in the cockpit of the us. This tension fuels the ever-moving goalpost as Millennium Falcon. Legacy analytics systems, So, as we prepare to project likely AI futures, 05 to what really constitutes interesting AI. like 3PO, could point out the odds (i.e., insights) let’s recap: AI itself is not new, but there is, and tell us verbosely, after the fact, what we and will always be, a new milestone afoot in AI. AI has always felt, and will always feel, “new.” might consider doing differently. Today’s AIs, Each of these milestones seems improbable 06 like Chewy, are fast enough to grab ahold of the upon approach, and in retrospect, “no big deal” That said, the contemporary burst of enterprise controls and get us out of trouble. Co-pilots as upon achievement. AI’s increasing centrality 07 attention is most attributable to the recent point- opposed to critics. to corporate strategy is a result of its recently in-time convergence of cloud-based architectures achieved agency—Its ability to be cost-effectively and open-source AI toolkits. Bespoke machine This relatively recent development, agency, embedded into business processes. learning models have been built for decades on sits at the center of AI’s newfound business offline copies (dumps, snapshots, or slices to cite utility. As ML systems graduate from “out-of- a few terms of art) with interesting and insightful band” analyzers to “in-band” actors, their value 7 The machines (or, a technological history of the future) 01 Charles Babbage and Ada Lovelace It would be roughly 100 years until the first digital designed the first general-purpose computers would actualize the concept in the 02 computer, their Analytical Engine, in the 1940s. Enormous, expensive, and requiring PhDs 1830s. Frustrated by manual errors in and punched cards to operate, early computers 03 their friend’s astronomical tables, they like ENIAC were hailed by the press as “electric hoped to bring the tireless precision of brains,” and by scientists as Turing Complete: That the steam-powered Industrial Revolution is, capable of solving any classical math or logic 04 to arithmetic. Economies worldwide problem they were told to solve. were being transformed by mechanical 05 muscles (e.g., locomotives, looms, But as it would turn out, a huge set of real-world steam-shovels, etc.) and it stood to decision-making isn’t readily framed as a tidy reason that mechanized math was itself math problem. A painter, of course, follows their 06 an attainable goal. Alas, the engineering muse, not their math, in deciding which colors tools and techniques of the day (and it and lines to use next.

Load more