Predicts 2018: Artificial Intelligence - Squarespace
←
→
Page content transcription
If your browser does not render page correctly, please read the page content below
This research note is restricted to the personal use of catarina.freitas.martins@novabase.pt.
Predicts 2018: Artificial Intelligence
Published: 13 November 2017 ID: G00343423
Analyst(s): Whit Andrews, Moutusi Sau, Chirag Dekate, Anthony Mullen, Kenneth F. Brant, Magnus Revang,
Daryl C. Plummer
AI technologies, especially deep learning, are poised to diffuse rapidly
through cloud services, APIs and the Internet of Things, driven by growing
consumer use of virtual assistants in smartphones and smart homes. CIOs
should start now to lay their organization's AI foundation.
Key Findings
■ Expectations will soar for AI-enabled assistance as it becomes pervasive in consumer services
and customer/citizen-facing applications through the use of virtual support assistants.
■ AI and machine-learning strategy development/investment is already in the top five CIO
priorities.
■ Organizations face growing threats across an array of digital channels, such as AI-generated
malicious "dis-information" that is intended to damage their brand value.
■ AIs will fuel a broad reaction in terms of growing concerns over liability, privacy violations, "fake
news" and pervasive digital distrust.
Recommendations
CIOs responsible for AI initiatives:
■ Start planning, developing and deploying intelligent virtual support capabilities in business
process areas that customers and citizens increasingly expect to be mediated through AI-based
assistants.
■ Use mature tools and preintegrated AI stacks to minimize development from scratch, with a
focus on customizing their applicability to the organization's specific use cases.
■ Protect the trust in your brand by adopting a digital ethics strategy.
■ Extend the ability of AI initiatives to create valuable content in parallel with using AI to verify the
authenticity of that content in corporate and government contexts.
This research note is restricted to the personal use of catarina.freitas.martins@novabase.pt.This research note is restricted to the personal use of catarina.freitas.martins@novabase.pt.
Table of Contents
Strategic Planning Assumptions............................................................................................................. 2
Analysis.................................................................................................................................................. 2
What You Need to Know.................................................................................................................. 3
Strategic Planning Assumptions....................................................................................................... 6
Replay Prediction......................................................................................................................18
A Look Back...................................................................................................................................18
Gartner Recommended Reading.......................................................................................................... 19
List of Figures
Figure 1. AI Capabilities Will Diffuse Quickly in 2018............................................................................... 3
Figure 2. Staffing Skills Is the No. 1 Challenge in Adopting AI..................................................................5
Figure 3. Deployment of AI Initiatives, 2018.......................................................................................... 10
Strategic Planning Assumptions
By 2020, 20% of citizens in developed nations will use AI assistants to help them with an array of
everyday, operational tasks.
By 2022, 40% of customer-facing employees and government workers will consult daily an AI virtual
support agent for decision or process support.
By 2020, 85% of CIOs will be piloting AI programs through a combination of buy, build and
outsource efforts.
By 2022, enterprise AI projects with built-in transparency will be 100% more likely to get funding
from CIOs.
Through 2020, AI-driven creation of "counterfeit reality," or fake, content will outpace AI's ability to
detect it, fomenting digital distrust.
By 2022, most people in mature economies will consume more false information than true
information.
Analysis
Our 2018 predictions for artificial intelligence (AI) are cued to the way it is diffusing (see Figure 1).
First, recent AI technology breakthroughs coupled with APIs and cloud architectures make still-
nascent AI capabilities and services more widely available (see "Smart Machines See Major
Page 2 of 21 Gartner, Inc. | G00343423
This research note is restricted to the personal use of catarina.freitas.martins@novabase.pt.This research note is restricted to the personal use of catarina.freitas.martins@novabase.pt.
Breakthroughs After Decades of Failure"). Second, the success of vendors in applying AI
technologies and conversational interfaces in smartphone and smart home virtual assistants creates
higher end-user expectations for these capabilities.
Figure 1. AI Capabilities Will Diffuse Quickly in 2018
Source: Gartner (November 2017)
What You Need to Know
Gartner believes that 2018 will mark the beginning of a democratization of AI, extending its impacts
across a much broader swath of companies and governments than previously. This expectation
offers CIOs new opportunities to experiment with AI and start to lay a foundation for successfully
piloting and exploiting it.
Several factors are driving this diffusion of AI capabilities:
■ Adding AI to applications and major platforms, such as cloud office suites. This means that
organizations and most employees will encounter AI outside specialized initiatives or
investments and without specialized skills (see "Maximize the Effectiveness of Office 365 and G
Suite With Everyday AI").
Gartner, Inc. | G00343423 Page 3 of 21
This research note is restricted to the personal use of catarina.freitas.martins@novabase.pt.This research note is restricted to the personal use of catarina.freitas.martins@novabase.pt.
■ The continued shift of AI toward cloud accessibility, as APIs are easier to consume and
experiment with than on-premises server farms. Conversations with Gartner clients about AI that
also mention "cloud" are growing faster than AI inquiries that don't mention it.
■ AI consumerization through the growing use and effectiveness of deep neural network (DNN)-
based virtual assistants, such as Alexa and Siri. This is boosting end-user awareness of and
expectations for intelligent conversational interfaces to products and services.
Gartner previously characterized AI as "a technology approach to enable machines to do what we
formerly thought only humans could do." Examples of what is popularly thought of as AI change
with technology's progress (see "Hype Cycle for Artificial Intelligence, 2017"). Deep learning
(machine learning via DNNs) qualifies as AI today. So do some forms of natural-language
processing (NLP). But many valuable and useful technologies that were once known as AI are now
taken for granted — including text-to-speech, which dates to the early 1980s.
In current popular cases, AI refers to systems that change behaviors without being explicitly
programmed, based on data collected, usage analysis and other observations.
Throughout 2018, AI performance will continue to improve. Segments such as speech recognition
have leaped in performance, thanks to deep-learning work by major cloud providers during the last
few years. Similar advances will occur both in specific vertical domains and in broader horizontal
areas, as spreading AI use creates incentives to capture better data with more detail.
CIOs face two key challenges in exploring and adopting AI: the availability of skilled and
experienced staff, and the lack of IT and business understanding of AI's potential (see Figure 2).
Many organizations are just starting to develop an AI strategy. The increased availability of AI
capabilities embedded in applications will alleviate the staffing issue (but exacerbate the hype). The
challenge of creating an AI strategic development plan parallels the staffing challenge, as having AI-
savvy workers and executives benefits organizations actively working to set strategy. Actions to
grapple with both constrains will reveal the organizations that are striving to improve their
understanding of what AI is best-suited for and how to employ it.
Page 4 of 21 Gartner, Inc. | G00343423
This research note is restricted to the personal use of catarina.freitas.martins@novabase.pt.This research note is restricted to the personal use of catarina.freitas.martins@novabase.pt.
Figure 2. Staffing Skills Is the No. 1 Challenge in Adopting AI
Source: Gartner (November 2017)
Skills are the greatest challenge in AI deployment, according to a recent Gartner Research Circle
survey. Management skills need to evolve — it's only recently that more managers have come to
understand and rely upon advanced statistical techniques that extract "signals from noise" to
improve decision making. This is the starting point for managing AI-based systems and services.
Technical skills — especially for deep learning — remain limited and are still evolving. We still do not
understand how to reliably configure a DNN to deliver useful results, and the long turnaround time
on DNN training makes for a long evaluation cycle. Universities are now producing many graduates
with valuable deep-learning skills, but few of these graduates have the intuition that delivers great
foundations for a successful DNN model.
Although not part of the above survey, another of the most difficult issues that forward-looking CIOs
will confront is the problem of trust in relation to AI-based automation. As in previous eras of
automation, such as industrial and office automation, AI will continue to demonstrate what could be
called "The Sorcerer's Apprentice temptation" — a situation where a person summons help from
allies who ultimately cannot be controlled (a lesson derived from the 18th-century poem of the same
1
name by Johann Wolfgang von Goethe). This will inevitably result in very public failures that will
Gartner, Inc. | G00343423 Page 5 of 21
This research note is restricted to the personal use of catarina.freitas.martins@novabase.pt.This research note is restricted to the personal use of catarina.freitas.martins@novabase.pt.
puncture AI's inflated hype, but also lead to new legal and regulatory regimes addressing privacy
and liability.
The documentation of AI action points and algorithms will become a significant obstacle for the
most strategic implementations of AI. It will be necessary to expose, in some fashion, the workings
of AI in status reports and explanations, to satisfy the very human concern that AI advice and
recommendations are founded on sound and ultimately understandable analysis (see Note 1). Such
concerns will be particularly keen in regulated industries.
Strategic Planning Assumptions
Strategic Planning Assumption: By 2020, 20% of citizens in developed nations will use AI
assistants to help them with an array of everyday, operational tasks.
Analysis by: Whit Andrews
Key Findings:
In developed nations, the individual's interactions with AI-based services are already becoming
more complex and more "operational" — users can now do more with virtual personal assistants
(VPAs) than simply ask a question and receive an answer. Speech error rates from major speech
recognition vendors now hover near 5%, which is close to the rate that one would accept from
human transcribers. This greater accuracy makes consumer-oriented speech services such as
Apple's Siri, Amazon's Alexa and Google Assistant more effective for millions of smartphone users.
Greater effectiveness sets higher expectations for what AI assistants can and should be able to do.
Gartner has already predicted, for example, that by 2018, more than 2 billion people will use
regularly conversational AI to interact with VPAs, virtual customer assistants (VCAs), virtual executive
assistants (VEAs), chatbots and other AI-enabled services on smartphones and connected devices.
Currently, most of these interactions are informational (query-response). But increasingly, these
interactions are functional — they result in actions. Consumers are becoming more familiar with the
process of using these services for simple tasks such as setting alarms or reminders. They will shift
easily into using these AI conversations for more-complex tasks, such as timing future tasks or
interacting in other ways.
Market Implications:
Today, Amazon offers developers the Alexa Skills Kit, enabling them to create "skills" or tasks that
Alexa can use in response to voice queries or instructions. For example, both Domino's Pizza and
Pizza Hut have created a pizza "skill" for Alexa. Users create an account at the retailer's website,
activate the associated skill and tell Alexa to order the pizza via that website. How this model will
evolve is uncertain — Amazon could create its own Alexa pizza-ordering skill and monetize the
transactions through its own payments system.
The key challenge for retailers, service providers and many others will be how the AI-powered
assistants of Apple, Amazon and so on "learn about" their product and services. Increasingly, the
criteria for surfacing services and products through these third-party-hosted conversations will
Page 6 of 21 Gartner, Inc. | G00343423
This research note is restricted to the personal use of catarina.freitas.martins@novabase.pt.This research note is restricted to the personal use of catarina.freitas.martins@novabase.pt.
hinge on how "smart" the assistant is and what relationship the vendor has established with the
assistant's creator.
Search engine optimization (SEO) concepts will be upgraded to ensure that a brand's knowledge
base and expertise can be demonstrated to third-party marketplaces — in order to be solicited.
Advertising and promotional agreements with such AI assistant services will become lucrative for
vendors that have previously not pursued such opportunities.
Large companies will be able to respond to AI assistants in two ways:
1. Expose fulfilment and knowledge services via APIs to the AI assistant (Alexa, Siri, etc.). No
conversation directly with user — services only.
2. Build their own virtual agent that can converse with the AI assistant about its relevance. A
possible use case is Amazon's Alexa creating conversations with a subset of Amazon sellers —
through their respective virtual agents — to establish which vendors can be presented on the
shortlist of options to an Amazon customer inquiry. (These relationships could use a transaction
or a placement model, possibly with some paid advertising, forming the future ecosystem
business model.)
Recommendations:
■ Begin planning to "market" products and services in personal and commercial environments
where AI-based assistants play an ever-more important role.
■ Adopt, adapt and extend SEO techniques to encompass interactions with AI assistants.
■ Experiment with APIs to VPAs or other AI-based services as these become available, and
expose information, products and services through AI assistants.
■ Anticipate the emergence of clearinghouse-style services that allow smaller organizations, both
commercial and governmental, to establish visibility with VPA services.
■ Explore couponing tactics to leverage the aggregation made possible through services that will
expose coupons to customers at new product research crossroads.
Related Research:
"Architecture of Conversational Platforms"
"When Will AI Virtual Support Agents Replace Your IT Service Desk?"
"Innovation Insight for Conversational Commerce"
"Market Insight: How to Collaborate and Compete in the Emerging VPA, VCA, VEA and Chatbot
Ecosystems"
Strategic Planning Assumption: By 2022, 40% of customer-facing employees and citizen-facing
government workers will consult daily an AI virtual support agent for decision or process support.
Gartner, Inc. | G00343423 Page 7 of 21
This research note is restricted to the personal use of catarina.freitas.martins@novabase.pt.This research note is restricted to the personal use of catarina.freitas.martins@novabase.pt.
Analysis by: Moutusi Sau
Key Findings:
The increased adoption of AI virtual assistants will drive greater usage of conversational platforms
as the starting point for communications with decision- or process-support agents. AI capabilities
will power virtual support agents at two levels:
1. As a resource enabling human support agents to respond faster and more effectively to
customer/citizen inquiries or actions
2. As a first-line conversational interface that responds to basic inquiries
The opportunity to reduce friction in relationships and improve service transcends business.
Citizens who have experience with the AI-powered improvements in consumer virtual assistants
such as Apple Siri or Amazon Alexa have growing expectations for similarly responsive intelligence
from government services.
Many CIOs have not yet realized the full potential of AI-based virtual support agents, including
chatbots. NLP coupled with machine learning enables virtual agents to understand what words
mean in different combinations and to ask questions to uncover intent and create context. Based on
this understanding, the virtual agent will be able to take or propose intelligent actions in response to
a customer or citizen query, either on its own or in tandem with a human support agent. A virtual
agent will be able to do this faster and more successfully compared to an agent — human or virtual
— lacking access to an intelligent body of research.
Near-Term Flag: AI virtual support agents will become commonly acquired and employed as part of
collaboration and content suites sold in organizations.
Market Implications:
Every enterprise in the future will use AI-empowered virtual assistants as part of their customer-
facing engagement. But the more significant evolution will be extending the use of AI into the back-
end decision and process frameworks. These capabilities will be increasingly used by customer-
facing and constituent-facing workers.
Gartner has predicted that by 2020, 25% of customer service and support operations will integrate
VCA technology across engagement channels, up from less than 2% in 2015. We are already seeing
very good progress with developments in virtual support agents (VSAs). VSAs typically are virtual
assistants that provide support to the IT service management process alongside the IT service
desk. They provide information to common questions and also have chatbotlike features.
The key challenges at this stage are market immaturity and product limitations. Most of the market
development has been around perfecting the simple chatbot transaction. As a result, there are as
yet few proven use cases for virtual agents that support front-office agents.
Page 8 of 21 Gartner, Inc. | G00343423
This research note is restricted to the personal use of catarina.freitas.martins@novabase.pt.This research note is restricted to the personal use of catarina.freitas.martins@novabase.pt.
Recommendations:
■ Analyze the business process areas where VSAs could augment the front-office customer
agent's performance, especially areas where there are more data-driven activities.
■ Have a strategic plan about implementation and usage before investing in any particular vendor
or product.
Related Research:
"When Will AI Virtual Support Agents Replace Your IT Service Desk?"
"Impacts of Artificial Intelligence and Machine Learning on Human Capital Management"
"Smart Machines: Consulting and System Integration Services Market Forecast and Opportunities"
"The Business Case for Buying Application Management Services With Intelligent Automation"
"The Impact of Intelligent Automation on Managed Workplace Services"
Strategic Planning Assumption: By 2020, 85% of CIOs will be piloting AI programs through a
combination of buy, build and outsource efforts.
Analysis by: Chirag Dekate
Key Findings:
Current AI trends mean that most organizations will not have to launch their own AI research
program from scratch. Instead, CIOs will be able to shift from their current phase of knowledge
gathering and strategy development to piloting and even implementing AI initiatives across business
units for specific use cases.
The vast majority of enterprises today are in the earliest stages of AI initiatives — but they are
moving forward quickly. Data from Gartner's most recent CIO survey reveals that about 4% of CIOs
have AI deployed now, while another 21% are piloting it or have it in short-term planning. Another
25% have AI initiatives in medium- or long-term planning (see Figure 3).
Gartner, Inc. | G00343423 Page 9 of 21
This research note is restricted to the personal use of catarina.freitas.martins@novabase.pt.This research note is restricted to the personal use of catarina.freitas.martins@novabase.pt.
Figure 3. Deployment of AI Initiatives, 2018
Source: Gartner (November 2017)
CIOs face a range of critical challenges, including isolated data silos, poor or uncertain data quality,
lack of digitalization and minimal AI skill sets. On top of this, DNNs open up new possibilities for AI.
Along with business and other IT leaders, CIOs struggle in understanding the capabilities of the new
slate of AI techniques and in selecting the right use cases where they can productively apply AI.
Web-scale companies, cloud-native organizations and cloud service providers are among the select
minority that are deploying statistical machine learning and DNNs in production today. But the field
is rapidly expanding in all business areas. The skills gap over the next three years will shrink
dramatically as more universities include and expand AI courses in their curricula and in widely
available retraining courses. Many early DNN adopters are aggressively prototyping and openly
sharing their techniques, providing enterprises with templates for replicating AI success stories.
Over the next three years, more software companies and cloud providers will integrate DNN
capabilities into their products, further reducing the complexity and barriers associated with AI
projects.
These developments enable most CIOs to begin experimenting with AI technologies. Over the next
two years, they will be scrambling to make sense of machine learning and other AI technologies, to
figure out their roles in digital business and to launch the internal pilots that will test that knowledge
and insight. At the same time, they will have to sift through competing vendor claims and promises
to identify and assess the genuineness of AI capabilities.
Page 10 of 21 Gartner, Inc. | G00343423
This research note is restricted to the personal use of catarina.freitas.martins@novabase.pt.This research note is restricted to the personal use of catarina.freitas.martins@novabase.pt.
Market Implications:
The technology "mystique" of AI can easily obscure the primary requirement for CIOs to identify the
use cases where AI technologies can have the greatest effect. Actively monitor AI success stories
and possible industry usage patterns for use cases to replicate in your own organization. Use cases
can be weighed realistically against current and emerging AI capabilities. To bootstrap AI initiatives,
pragmatically blend capabilities from outsourcing, building from scratch or buying off the shelf.
Expertise in machine learning and DNNs will be a scarce resource, and relatively few organizations
will have the money or incentive to devise DNNs from scratch. The limited availability of AI skills and
experience will mean a bidding war for talent. Most undergraduate and graduate STEM programs
feature machine learning and statistics courses, and these will expand. CIOs will have to balance
hiring from this talent pool with using external or internal training programs to develop and improve
internal skills.
An array of vendors will embed AI expertise, tools and capabilities into products and services. The
most promising of these vendors will be those that proactively integrate (and simplify access to)
infrastructure solutions, thus minimizing end-user adoption pain. CIOs should ensure that AI/DNN
initiatives prioritize identifying the tools and techniques that will enable a fast launch. In other cases,
outfit existing middleware products with machine-learning add-ons, or buy or subscribe to
externally sourced machine- or deep-learning middleware.
Gartner expects cloud service providers and others to introduce robust machine-learning
environments that are exposed via cognitive APIs. Examples of these are IBM Watson products and
services, Microsoft Azure Machine Learning Studio and Amazon Machine Learning. Amazon has a
three-layer AI "stack": frameworks and infrastructure with tools such as Apache MXNet and
TensorFlow, API-driven services that let developers add intelligence to applications, and machine-
learning platforms aimed at data scientists. Cloud service providers are gearing up to offer
comprehensive IaaS, SaaS and PaaS in scalable public cloud environments.
Recommendations:
■ Select proven approaches to implementing narrow AI projects that can deliver meaningful
results by objectively analyzing broader AI adoption patterns.
■ Nurture agile development practices with cross-organizational teams, enabling rapid
prototyping and validation of engineered products.
■ Use mature tools and preintegrated AI stacks to minimize development from scratch, with a
focus on customizing their applicability to your organization's specific use cases.
■ Architect effective outsourcing strategies for software development and production deployment
— especially when in-house skill sets are lacking or oversubscribed.
Related Research:
"Three Elements of a Scalable Enterprise Machine Learning Infrastructure Strategy"
Gartner, Inc. | G00343423 Page 11 of 21
This research note is restricted to the personal use of catarina.freitas.martins@novabase.pt.This research note is restricted to the personal use of catarina.freitas.martins@novabase.pt.
"Doing Machine Learning Without Hiring Data Scientists"
"How to Start a Machine-Learning Initiative With Less Anxiety"
"Applying Artificial Intelligence to Drive Business Transformation: A Gartner Trend Insight Report"
"Machine Learning: FAQ From Clients"
"Preparing and Architecting for Machine Learning"
Strategic Planning Assumption: By 2022, enterprise AI projects with built-in transparency will be
100% more likely to get funding from CIOs.
Analysis by: Chirag Dekate, Anthony Mullen
Key Findings:
CIOs are evaluating AI initiatives for mission-critical application scenarios. The potential benefits
from AI-driven process and capability optimization are immense. But so are the dangers associated
with creating systems that appear to deliver biased results, which could have devastating and highly
public impacts on AI outcomes. Consequently, decision makers need to ensure accountability and
transparency of the underlying methodologies, which is especially challenging in the case of DNNs.
For enterprise use cases, CIOs must be able to validate and defend AI system outcomes, so
transparency of methods is a crucial criteria. CIOs seeking to deliver impactful AI initiatives in
production will adopt capabilities that are transparent.
But transparency in AI varies. Statistical machine-learning techniques such as clustering are by
design transparent — the process by which algorithms derive their results can be explained with
relative clarity. By contrast, advanced AI techniques that improve their accuracy over time or as they
execute, such as deep learning, are opaque. The outcomes of these methods over time can be
highly accurate, but the "mechanics" of the constituent neural networks that generate the results
are not predictable. The output of a DNN can always be explained — but only after the fact.
Furthermore, perfectly reasonable inputs may still result in apparently biased results because of
social norms. Despite this, their accuracy ensures that DNNs will be deployed in various application
domains, making the demand for transparency and accountability more urgent and acute.
For many enterprise use cases, from financial services to autonomous vehicles, implementing DNN
techniques will be challenging. In financial services, regulations require that financial service
analytics be clearly defined and explainable. In these settings, while DNN could deliver extremely
accurate results, its intermediate stages of how data was transformed to derive outcomes tend to
be opaque. Consequently, for these and other related use cases involving litigation, regulatory
compliance, oversight and operational control, using more-transparent methods will be required.
AI ecosystems should provide tools to verify and validate the data sources and model outcomes.
The more transparent the machine-learning modeling environment is, the more effectively
organizations can analyze the efficacy of the underlying machine-learning models. Common tools in
this area are Spark ML and other similar frameworks. Additionally, researchers are actively working
to improve transparency of DNN methodologies.
Page 12 of 21 Gartner, Inc. | G00343423
This research note is restricted to the personal use of catarina.freitas.martins@novabase.pt.This research note is restricted to the personal use of catarina.freitas.martins@novabase.pt.
Over time, other protections will emerge where complete transparency will be difficult or impossible.
For example, some form of insurance may become an option in cases where the regulatory and
legal lines are less bright or less fixed.
Near-Term Flag: "Hatches," where organizations can view documentation of how decisions are
reached in DNNs and where visualization of results is presented for explicatory reasons, will become
familiar aspects of AI.
Market Implications:
The presence or absence of transparency features to help end users validate and verify AI models
will be a critical distinction among vendors and a key selection criteria for CIOs. Where absolute
transparency is not possible, leading vendors will be those that provide augmented tools to help
end users review intermediate data and improve debugging capabilities.
Along with new data initiatives for AI, CIOs also will need to institute new AI model auditing to
ensure veracity of the data, the AI models and the model results. Compared to traditional IT
environments, AI environments will require higher-frequency auditing of input streams and validation
of output signals. Vendors can simplify some of these challenges by integrating auditing capabilities
within their software offerings.
Proliferation of AI tools and methodologies will make advanced analytics more accessible for end
users. However, decision makers in charge of integrating AI in enterprise business cases will seek
those that deliver advanced AI while maximizing transparency of the models. Regulatory authorities
in charge of devising sustainable AI policies should enshrine transparency as a foundational
element.
Recommendations:
■ Select algorithms, tools and integration frameworks that maximize transparency and
accountability of AI-augmented environments.
■ Audit input data streams and validate model chains periodically to ensure the compliance of
algorithmic transparency and accountability, and proactively create policies to detect deviation
from this norm.
■ Use transparency as a key criteria in selecting AI algorithms for mission-critical enterprise
applications.
Related Research:
"Artificial Intelligence Primer for 2017"
"Questions to Ask Vendors That Say They Have 'Artificial Intelligence'"
"Hype Hurts: Steering Clear of Dangerous AI Myths"
Gartner, Inc. | G00343423 Page 13 of 21
This research note is restricted to the personal use of catarina.freitas.martins@novabase.pt.This research note is restricted to the personal use of catarina.freitas.martins@novabase.pt.
"What CIOs Should Ask When Someone Pitches a Project That Uses AI"
"Chief Data Officers Desk Reference for Artificial Intelligence"
Strategic Planning Assumption: Through 2020, AI-driven creation of "counterfeit reality," or fake,
content will outpace AI's ability to detect it, fomenting digital distrust.
Analysis by: Daryl Plummer, David McCoy
Key Findings:
■ The creation of content in multiple forms (e.g., entertainment, corporate, political, documentary,
news, etc.) that is not real is beginning to accelerate beyond human ability or will to flag.
■ AI systems have the ability to categorize the content of images faster and more accurately than
humans can, but AI is not applied to the problem often enough.
■ Neural network generation of "fake" video news is the next near-term frontier.
■ The creation of counterfeit images to make actors appear younger in movies (even from death)
occurs in over 90% of the biggest movie releases each year.
Near-Term Flags:
■ In 2018, a counterfeit video used in a satirical context will begin a public debate once accepted
as real by one or both sides of the political spectrum.
■ By 2018, there will be a 10-fold increase in commercial projects to detect fake news.
Market Implications:
"Counterfeit reality" is presented in digitally created images, video, documents or sounds that are
convincingly realistic representations of things that never occurred or never existed exactly as
represented.
The historical tendency of business and government entities has long been to accept data as valid
points of information that can further common understanding and acceptance of beliefs and
practices. However, in the past 30 years, the ability to create and to disseminate content that has
been subtly or overtly altered has increased by orders of magnitude. Massive numbers of people
having access to the internet with few controls on content distribution meant that this was an
inevitable outcome. Now comes the next wave of that distribution: machine-generated content.
As neural networks and AI have progressed, they have reached a point where they can successfully
categorize image content at least as well as humans can — but they can do it significantly faster.
Alongside this, the use of computer-generated imagery has progressed with the help of the
entertainment industry (movies, games, etc.) to a point where the convincing digital portrayal of
humans is now commonplace. Further, the concept of fake news (i.e., intentionally designed to
misrepresent the truth) as a political football has focused many on the proliferation of propaganda
and downright corrosive content through social media. These sites have now embarked on efforts
Page 14 of 21 Gartner, Inc. | G00343423
This research note is restricted to the personal use of catarina.freitas.martins@novabase.pt.This research note is restricted to the personal use of catarina.freitas.martins@novabase.pt.
to detect and minimize the spread of fake content. Much of this is destined to fail, although their
efforts will not be in vain.
The detection of counterfeit reality will best be accomplished by AI that is able to identify and track
markers in counterfeit content faster than human reviewers can. The process of detecting
counterfeit content will be applied to fake news as often as it is to video and audio content that can
present a threat to corporate brands and reputation. Unfortunately, the use of AI to detect
counterfeit reality lags behind the use of AI to create it.
The creation of counterfeit reality has accelerated using AI techniques in recent years, with multiple
universities and research organizations creating compelling yet completely inauthentic video
2
representations of famous people saying things that they never said. Computer mapping of facial
expressions as well as body language using neural networks have been paired with mimicking
3
technology to map one person's facial motions onto another.
For these reasons, among others, counterfeit reality creation and detection are increasingly
becoming a perfect fit for AI-based methods.
Market effects that continue to grow are:
■ Corporate branding and reputation are at stake due to counterfeit delivery of messages or
contract-related terms and conditions.
■ The political impact of counterfeit video and images will escalate as actors move beyond faked
textual content to content that misrepresents anything, from individual actions to speeches
before important audiences.
■ The legal profession will need new laws passed to combat the evolution of counterfeit video,
audio and documentation being presented in a legal context.
■ The entertainment industry, from movie production to video games, will continue to lead the
way in counterfeit reality creations.
■ Social media giants will lead in combating/detecting counterfeit content for the purposes of
accuracy, authenticity and even rapid removal of inappropriate video content.
■ Counterfeit reality detection as a business will enable new companies to market their expertise
in order to protect corporate, political, individual or legal entities, much as identity theft
protection has done.
Recommendations:
■ Do not underestimate the number of counterfeit reality sources that exist in the world today.
■ Tier your content decisions to place the highest level of "anti-counterfeiting" scrutiny on the
most critical content you produce and rely upon, whether internally or externally created.
■ Extend the ability of AI to create valuable content in parallel to using it to verify content
authenticity in corporate and government contexts.
Gartner, Inc. | G00343423 Page 15 of 21
This research note is restricted to the personal use of catarina.freitas.martins@novabase.pt.This research note is restricted to the personal use of catarina.freitas.martins@novabase.pt.
■ Use machine-generation platforms, such as the narrative science Quill platform, to leverage AI-
generated content.
■ Track the progress of social giants and universities in detecting and generating convincing
counterfeit reality.
Related Research:
"Market Insight: Creative AI — Assisted and Generative Content Creation"
"Four Ways to Get More Value From Data Visualization"
Strategic Planning Assumption: By 2022, most people in mature economies will consume more
false information than true information.
Analysis by: Magnus Revang, Whit Andrews
Key Findings:
■ Confirmation bias — a well-known human tendency — leads all people to seek out, select and
value information that parallels what they believe and expect to be proven true.
■ AI can detect false information. It can also generate it. AI applications will compete such that
quality of detection and generation improve with time.
■ Creating false information will always cost less in currency and effort than the cost of detecting
it. False information will consequently outpace true information where there is economic or
political reason to purvey it.
Near-Term Flags:
■ Before 2020, false information will fuel a major financial fraud, made possible through high-
quality falsehoods moving the financial markets worldwide.
■ Through 2020, no large internet company will fully succeed in its attempts to mitigate this
problem.
■ By 2020, a major country will pass regulations or laws seeking to curb the spread of false
information.
Market Implications:
A major theme for 2017 in politics and the media worldwide has been the creation of "fake news."
The terms has been used to discredit true information as well as properly used to describe wholly
false information, with many stops along a spectrum in between. Gartner introduced the term
counterfeit reality in 2004 to describe the much broader phenomenon of content created digitally for
good as well as for ill (e.g., in entertainment, the use of digital actors in movies, photographs, text
and audio content).
Page 16 of 21 Gartner, Inc. | G00343423
This research note is restricted to the personal use of catarina.freitas.martins@novabase.pt.This research note is restricted to the personal use of catarina.freitas.martins@novabase.pt.
While fake news is currently in the public consciousness, it is important to realize that the extent of
digitally created content that is not factual or authentic representation of information goes well
beyond news.
Without taking any political position on what currently constitutes true or false information, we can
say that society faces a profound challenge. Different media sources are reporting vastly different
facts and perspectives on the same event. False information as a "first perception" — especially
when collectively endorsed — can persist in subcultures long after it has been roundly debunked.
Trying to correct the misinformation can make the misunderstanding worse. "Corrections that
merely encourage people to consider the opposite of initial information often inadvertently
strengthen the misinformation. Therefore, offering a well-argued, detailed debunking message
4
appears to be necessary to reduce misinformation persistence."
A future where people can elect to live in a global subreality — regardless of that reality's alignment
with fact — and have support for that belief in the information they consume and media they are
exposed to is a very new phenomenon. In the digital world, anybody can look credible. "Spoof"
websites with real impacts for those who fail to get the "joke" date back 20 years. Fake news
follows in the footsteps of "The Protocols of the Elders of Zion," a thoroughly untrue document
published to appear true; the document has been and still is used as a foundation for anti-Semitism.
5
For enterprises, this acceleration of content in a social-media-dominated discourse — of which
much will be false — presents a real problem. Enterprises need to monitor closely what is being
said about their brands directly and in what context, so as to not be associated with content that is
detrimental to their brand value. They must also avoid establishing partnerships (or eschewing
them) based on intentionally spread misinformation. And they must defend their brands and those
of their partners against conscious attacks. Disinformation does not deserve the protection of "just
business; nothing personal."
Ultimately, brands that have diligently worked to cultivate unique and recognizable behaviors and
values will generally prevail, being much more resistant to any effort of undermining. Enterprises
that choose not to participate in short-term manipulation of truth and persist on preserving their
trustworthiness will win in the long term. The greatest risk to enterprises will lie in areas where
technological loyalty is based more in emotion than in fact — for example, the so-called "religious
wars" around competing mature platforms and ecosystems.
Enterprises that need to reach people must learn to tailor their messages in multiple ways to
penetrate different alternative reality bubbles.
Recommendations:
■ Protect the trust in your brand by adopting a digital ethics strategy.
■ Monitor information diligently to catch false information about your brand, and draft and update
strategies for swift reaction.
Gartner, Inc. | G00343423 Page 17 of 21
This research note is restricted to the personal use of catarina.freitas.martins@novabase.pt.This research note is restricted to the personal use of catarina.freitas.martins@novabase.pt.
■ Respond without recapitulating the falsehoods, as that reinforces them. Instead, present the
truthful narrative to replace the lie in recipients' brains.
■ Do not fund any activity that rewards the creation of false information.
■ Invest in the training of your employees to ensure that they are inclined and skilled to discern
the truth — it will be a competitive advantage for decision making.
Related Research:
"Use a Digital Trust Index to Maximize Digital Business Performance"
Replay Prediction
The replay prediction is a prediction from a previously published report that is so significant that it is
being republished here.
A Look Back
In response to your requests, we are taking a look back at some key predictions from previous years.
We have intentionally selected predictions from opposite ends of the scale — one where we were
wholly or largely on target, as well as one we missed.
On Target: 2015 Prediction — By 2018, one in three people with a smartphone will be using virtual
personal assistants (VPAs).
Based on the 2015 Gartner Mobile Apps Survey, 39% of respondents in the U.S. and the U.K. used
VPAs on their smartphone in the three months leading up to the survey (cited in "Predicts 2017:
Personal Devices").
Gartner's survey found that 45% of consumers in the U.S. and 32% in the U.K. had used a VPA app
on their smartphone. Across all respondents surveyed, VPAs were used by 31%. Those who were
using one seemed to do so regularly, with one in three using the app daily and another 38% at least
once a week. It was the sixth-most-used mobile app behind social media, messaging, gaming,
music and video, although it was used mainly for simple tasks: weather, location and calendar
checks.
As providers add new features, including integration with business services and support for
additional languages, Gartner expects usage to continue showing strong growth.
Missed: 2014 Prediction — By 2018, at least one smart-machine maker will have settled a liability
suit because its product made a negligent or criminal decision.
The number of scandalous AI-based projects continues to rise, but none has yet resulted in a
publicly disclosed liability settlement in a court case. The scandals include Facebook having anti-
Semitic ad categories; another is the algorithm that detects sexual orientation based on visual
recognition.
Page 18 of 21 Gartner, Inc. | G00343423
This research note is restricted to the personal use of catarina.freitas.martins@novabase.pt.This research note is restricted to the personal use of catarina.freitas.martins@novabase.pt.
We can see that it's only a matter of time, however, based on the lawsuits starting to emerge around
autonomous driving, which is impossible without machine learning. The Miami Herald reported on
6
details of a class action suit filed by a Florida driver against electric automaker Tesla. Tesla has
denied the allegations.
Acronym Key and Glossary Terms
DNN deep neural network
STEM science, technology, engineering and math
VPA virtual personal assistant
Gartner Recommended Reading
Some documents may not be available as part of your current Gartner subscription.
"A Framework for Applying AI in the Enterprise"
"Hype Hurts: Steering Clear of Dangerous AI Myths"
"Questions to Ask Vendors That Say They Have 'Artificial Intelligence'"
"How Enterprise Software Providers Should (and Should Not) Exploit the AI Disruption"
"Seek Diversity of People, Data and Algorithms to Keep AI Honest"
"Survey Analysis: Enterprises Dipping Toes Into AI but Are Hindered by Skills Gap"
Evidence
1 "The Sorcerer's Apprentice," Wikipedia.
2 K. Hao, "Researchers Have Figured Out How to Fake News Video With AI," Quartz, 19 July 2017.
3M. Niessner, "Face2Face: Real-Time Face Capture and Reenactment of RGB Videos," Stanford
University.
4M.S. Chan, C.R. Jones, K. H. Jamieson, D. Albarracín. "Debunking: A Meta-Analysis of the
Psychological Efficacy of Messages Countering Misinformation," Psychological Science, Sage
Journals, 12 September 2017.
5 "The Protocols of the Elders of Zion," Wikipedia.
6 "Tesla Autopilot Called 'Dangerously Defective' in Lawsuit by Driver," Miami Herald, 28 April 2017.
Gartner, Inc. | G00343423 Page 19 of 21
This research note is restricted to the personal use of catarina.freitas.martins@novabase.pt.This research note is restricted to the personal use of catarina.freitas.martins@novabase.pt.
Note 1 Making Computers Explain Themselves
Researchers are addressing how to train neural networks not only to make predictions and
classifications, but also to reveal the rationale for their decisions.
See L. Hardesty, "Making Computers Explain Themselves," MIT News, 27 October 2016.
Page 20 of 21 Gartner, Inc. | G00343423
This research note is restricted to the personal use of catarina.freitas.martins@novabase.pt.This research note is restricted to the personal use of catarina.freitas.martins@novabase.pt.
GARTNER HEADQUARTERS
Corporate Headquarters
56 Top Gallant Road
Stamford, CT 06902-7700
USA
+1 203 964 0096
Regional Headquarters
AUSTRALIA
BRAZIL
JAPAN
UNITED KINGDOM
For a complete list of worldwide locations,
visit http://www.gartner.com/technology/about.jsp
© 2017 Gartner, Inc. and/or its affiliates. All rights reserved. Gartner is a registered trademark of Gartner, Inc. or its affiliates. This
publication may not be reproduced or distributed in any form without Gartner’s prior written permission. If you are authorized to access
this publication, your use of it is subject to the Gartner Usage Policy posted on gartner.com. The information contained in this publication
has been obtained from sources believed to be reliable. Gartner disclaims all warranties as to the accuracy, completeness or adequacy of
such information and shall have no liability for errors, omissions or inadequacies in such information. This publication consists of the
opinions of Gartner’s research organization and should not be construed as statements of fact. The opinions expressed herein are subject
to change without notice. Although Gartner research may include a discussion of related legal issues, Gartner does not provide legal
advice or services and its research should not be construed or used as such. Gartner is a public company, and its shareholders may
include firms and funds that have financial interests in entities covered in Gartner research. Gartner’s Board of Directors may include
senior managers of these firms or funds. Gartner research is produced independently by its research organization without input or
influence from these firms, funds or their managers. For further information on the independence and integrity of Gartner research, see
“Guiding Principles on Independence and Objectivity.”
Gartner, Inc. | G00343423 Page 21 of 21
This research note is restricted to the personal use of catarina.freitas.martins@novabase.pt.You can also read
