AI is too busy serving consumers to bother taking over the world

By Kevin Fogarty  |  January 3, 2018

Business dithers while consumers lead the way – as they did with social networks, smart phones, the cloud, the web, the internet, LANs, the PC…


2017 was a big year for artificial intelligence. It wasn’t THE year – the one in which au courant versions of AI replaced hunt-and-peck, brute-force computing as the preferred way to write software to solve problems. During 2017 we were still 45 years from a time when AI would do almost anything better than humans, even according to a pro-AI poll from a pro-AI site.


Of course, it also wasn’t the year AI took over the job of destroying human civilization, either, despite warnings from expert sources including Stephen Hawking, Elon Musk and the Terminator movie franchise. (2017 was also not the year AI saved human civilization – Hawking also suggested – or that artificial intelligence merged with our own to make humanity far greater than would ever be possible if it continues to rely on meat-based computing platforms, as Ray Kurzweil and other utopians have suggested.)


2017 was the year businesses realized there is no such thing as AI,” and that they’d have to figure out whether a specific function could be delivered best using machine learning, natural language processing, deep learning, neural networks or a dozen other ways to get software to emulate some aspects of human memory or judgement.


Despite the increasing reliance of business on sophisticated analytics for everything from strategy to marketing to design – and widespread acceptance of AI methods as the best way to improve the scale and ability of those analyses – 59% of companies were still researching their options and only 6%  had deployed cognitive software in production roles during 2017, according to a Gartner summary from October.  AI will add $29 Trillion of business value by 2021, the report said, but only 4% of companies said in 2017 that they plan to roll out AI software during 2018.


A June, 2017 McKinsey Global Institute survey report of a survey of more than 3,000 companies showed AI still mostly at the fringes. Only 20% of respondents said they used AI in core business functions. When McKinsey examined the 160 use cases it found only 12% used AI for genuinely commercial purposes.


Analysis paralysis


Part of the reason for the logjam in planning-and-analysis could be the complexity of cognitive computing methods, which require developers with much greater skill in linear algebra, statistics and predictive modeling than are typical in brute-force programming.


Part of the reason could also be a fuzzy understanding on the part of non-developers about how to use a new set of capabilities and how much a business had to change itself to get that advantage.


 “The value you receive from AI is proportional to how much you rethink your business,” read one slightly off-putting Gartner report that was typical of advice during late 2016 and early 2017. “You have to fundamentally rethink how we are going to work.”


Gartner now recommends “narrow AI” approaches that demonstrate value using huge improvements in the performance of a single application or feature. The leading example is a consumer application, not business analysis. In late 2016 Google converted the primary code of Google Translate – a 10-year-old app that was often adequate as a dictionary with automated search – using Google Brain and neural machine translation. Almost overnight its translations became good enough to impress a human-computer-interaction professor at the University of Tokyo, who posted personal tests of English-Japanese translations using not technical documents, but samples of literature  in which phrasing and texture were as important as vocabulary.


2017 was not the year businesses took advantage of that potential. It was the year they realized it existed and that they should figure out how to get for themselves the kind of benefit consumer-facing applications from Google and Facebook and other cloud-service providers were getting from the efficiency, novelty and customer-pleasing impact of cognitive computing.


“Take My Money!”: Consumers to the rescue


2017 was the year, in fact, that the real champion of AI turned out to be not wonks obsessed with analytics, but ordinary consumers, whose interest will make consumer-oriented AI a $2.7 billion business by the end of 2017, according to a September prediction by market researcher Tractica.


It was consumers, for example,  who thought it was so cool that Google Translate might be able to translate 40 languages on the fly that they raised a huge buzz over the impending shipment of the $159 Google Pixel Bud headset.


Consumers talked to devices, too. During 2017, 60.5 million consumers in the U.S. used a voice-based assistant, usually on a smartphone, at least once per month and 35.6 million used a voice-activated device at least as often, according to market-research firm eMarketer.


Usage and product purchases through devices such as Amazon’s Echo – which holds a 70% share of the smart-speaker market -- have increased to the point that  researchers at RBC Capital estimated Amazon could generate $10 billion in revenue via voice-driven shopping, services and extra functions provided by third parties using Amazon Web Services as a back end.


That helps explain why 31% of healthcare executives told PwC that virtual personal assistants would have an impact on their business and only 29% said the same of automated data analysts.


It also explains the growing competitive battle among developers of virtual assistants and the pucks, smart speakers and smartphones in which they live.


Instant installed base?


Smart speakers are surprisingly popular, but if you want to put AI in the hands of most consumers, a phone is your best option. Hence the September announcement that Apple would build a “neural engine” into the A11 processor it developed to drive the iPhone X. The neural engine, which is designed to improve facial recognition and Siri’s voice recognition and expand iPhone augmented reality functions, is an integrated circuit designed to accelerate inference – the client function of most machine-learning applications. Inference engines use models built using massive, datacenter-based neural networks to build data models that allow relatively low-powered devices to deliver surprisingly powerful AI functions.


Apple’s version will come out during 2018; so will me-too versions from Qualcomm, Huawei and others in a smartphone market far too competitive to allow players to ignore Apple’s AI play.


By 2022, AI/ML inference engines will be built into 1.7 billion client devices, including smartphones and Internet of Things things, according to Linley Gwennap, president of the Linley Group, and editor-in-chief of Microprocessor report.


Expecting business to step up


Consumers made up 26% of the market for AI software, hardware and services during 2016, according to Tractica, but that number will drop to 8% by 2025 as businesses accelerate their adoption and use of AI. The 266 use cases Tractica identified in 29 industries may show they’re already doing so.


Despite “AI washing” by companies claiming cognitive computing capabilities for software that hadn’t gotten them yet, almost every business app will have some AI component by 2020, according to Gartner.


Researchers disagree on growth rates. IDC predicts software, services and hardware revenue for AI and cognitive computing will grow to a total of $57.6 billion by 2021.


Market researchers Tractica predicted in May that AI revenue would total of $59.8 billion by 2025. After research that identified 266 different current use cases of AI in 29 markets, it revised its estimate – predicting annual revenue would hit  $89.8 billion by 2025.


Ubiquitous and…invisible?


That still doesn’t mean AI will take over the world in 2018.


It probably won’t “take over” at all, in fact. It will just seep up into the world in one application after another – in finance and medicine and logistics and consumer services – until we stop wondering which applications are “AI” and simply judge applications on how well they do something and how innovative the thing is that they do.


Consumer-oriented companies like Google and Facebook are already plastering the world with AI; there will be a lot more of that in 2018. By 2020 we’ll be using it so heavily we’ll consider cognition just “decent software” and may not see much “AI” about it at all.