INDUSTRY ADOPTION (y-axis): Signals include momentum of startups in the space, media attention, customer adoption (partnerships, customer, licensing deals).
MARKET STRENGTH (x-axis): Signals include market sizing forecasts, quality and number of investors and capital, investments in R&D, earnings transcript commentary, competitive intensity, incumbent deal making (M&A, strategic investments).
Trends seeing adoption but where there is uncertainty about market opportunity.
As Transitory trends become more broadly understood, they may reveal additional opportunities and markets.
- Cyber threat hunting
- Conversational AI
- Drug discovery
Trends which are seeing widespread industry and customer implementation / adoption and where market and applications are understood.
For these trends, incumbents should have a clear, articulated strategy and initiatives.
- Open-source frameworks
- Edge AI
- Facial recognition
- Medical imaging & diagnostics
- Predictive maintenance
- E-commerce search
Conceptual or early-stage trends with few functional products and which have not seen widespread adoption.
Experimental trends are already spurring early media interest and proof-of-concepts.
- Capsule Networks
- Next-gen prosthetics
- Clinical trial enrollment
- Generative Adversarial Networks (GANs)
- Federated learning
- Advanced healthcare biometrics
- Auto claims processing
- Checkout-free retail
- Back office automation
- Language translation
- Synthetic training data
Large addressable market forecasts and notable investment activity.
The trend has been embraced by early adopters and may be on the precipice of gaining widespread industry or customer adoption.
- Reinforcement learning
- Network optimization
- Autonomous vehicles
- Crop monitoring
Download the full CB Insights Report Artificial Intelligence Trends in 2019
Although science fiction may depict AI robots as the bad guys, some tech giants now employ them for security. Companies like Microsoft and Uber use Knightscope K5 robots to patrol parking lots and large outdoor areas to predict and prevent crime. The robots can read license plates, report suspicious activity and collect data to report to their owners.
These AI-driven robots are just one example of “autonomous things,” one of the Gartner Top 10 strategic technologies for 2019 with the potential to drive significant disruption and deliver opportunity over the next five years.
“The future will be characterized by smart devices delivering increasingly insightful digital services everywhere,” said David Cearley, Gartner Distinguished Vice President Analyst, at Gartner 2018 Symposium/ITxpo in Orlando, Florida. “We call this the intelligent digital mesh.”
- Intelligent: How AI is in virtually every existing technology, and creating entirely new categories.
- Digital: Blending the digital and physical worlds to create an immersive world.
- Mesh: Exploiting connections between expanding sets of people, businesses, devices, content and services.
“Trends under each of these three themes are a key ingredient in driving a continuous innovation process as part of the continuous next strategy,” Cearley said.
The Gartner Top 10 Strategic Technology trends highlight changing or not yet widely recognized trends that will impact and transform industries through 2023.
Read the full article on Gartner.com
Written by Microsoft Research team on November 2018.
1. Articulate the purpose of your bot and take special care if your bot will support consequential use cases.
The purpose of your bot is central to ethical design, and ethical design is particularly important when it is anticipated that a consequential use will be served by the bot you are developing. Consequential use cases include access to services such as healthcare, education, employment, financing or other services that, if denied, would have meaningful and significant impact on an individual’s daily life.
• Before beginning design work, carefully specify how your bot will benefit the user or the entity deploying the bot. If the bot is likely to affect the well-being of the user, such as providing access to a consequential service like healthcare, attention to these guidelines will be especially important. Be sure to pause to research, learn and deliberate on the impact of the bot on people’s lives. When in doubt, seek guidance.
• Assess whether the bot’s intended purpose can be performed responsibly. Some purposes may inherently require human judgment, empathy and expertise or a very high degree of reliability and accuracy, e.g., healthcare diagnosis or financial planning. Be sure to consider the nature and type of errors in the performance of the bot and their cost to users. Consider if you have access to relevant expertise in the domain in which your bot would operate.
• Develop metrics to assess user satisfaction. Metrics for your bot should cover not only whether the user feels that the bot served its intended purpose, but also the user’s sense of well-being and comfort while using the bot
2. Be transparent about the fact that you use bots as part of your product or service.
Users are more likely to trust a company that is transparent and forthcoming about its use of bot technology, and a bot is more likely to be trusted if users understand that the bot is working to serve their needs and is clear about its limitations.
• It should be apparent to the user that they are not having an interaction with another person. Since designers might endow their bots with “personality” and natural language capabilities, it is important to convey to users that they are not interacting with another person and some aspects of their interaction are being performed by a bot. There are variety of design choices that can be made to accomplish this that do not degrade the user experience.
• Establish how the bot can help and the limitations associated with its use. Users are more likely to find a bot to be trustworthy if the bot sets reasonable expectations for what it can do and what it does not do well. Users should be able to easily find information about the limitations of the bot, including the possibility of errors and the consequences that can flow from such errors. For users who wish to “learn more,” you should offer a more detailed explanation of the purpose and operation of the bot.
3. Ensure a seamless hand-off to a human where the human-bot exchange leads to interactions that exceed the bot’s competence.
If your bot will engage people in interactions that may require human judgment, provide a means or ready access to a human moderator.
• Respect individual engagement preferences, particularly if your bot deals in consequential matters. Bots designed for use in consequential matters should incorporate the ability to transfer an engagement to a human moderator as soon as the user asks, otherwise indicates, or if the bot recognizes (e.g., through sentiment analysis) that the user is dissatisfied. If users feel trapped or alienated by a bot, they will quickly lose trust in the technology and in the company that has deployed it.
4. Design your bot so that it respects relevant cultural norms and guards against misuse.
Since bots may have human-like personas, it is especially important that they interact respectfully and safely with users and have built-in safeguards and protocols to handle misuse and abuse.
• Limit the surface area for norms violations where possible. Every bot should be designed to follow a specific set of values and cultural norms. To reduce the possibility of conflicting with those values and cultural norms, limit the surface area for norms violations. For example, if your bot is designed to take pizza orders, limit it to that purpose only, so that it does not engage on topics such as race, gender, religion, politics and the like.
• Where appropriate, point to a relevant “code of conduct” for users. Consider whether your bot should be subject to a user code of conduct (from your organization or the entity deploying the bot) that, for example, includes prohibitions on hate speech, bullying and threatening others, and provides appropriate notice to the user of any code of conduct.
• Apply machine learning techniques and keyword filtering mechanisms to enable your bot to detect and — critically — respond appropriately to sensitive or offensive input from users. Deploy a two-way filtering mechanism with a customizable threshold of tolerance for what your bot takes in from users, as well as what your bot says in response. In most cases, we 3 recommend the bots simply steer clear of controversial subjects (especially hate speech). Open domain conversations are considered high-risk because they require significant investment in both content operations and social media monitoring capabilities and must be maintained 24/7 with bugfix service level agreements. You should leverage products that include offensive text classifiers, such as the Microsoft Bot Framework, , to protect your bot from abuse if it engages in open domain conversations. Sensitive categories include adult content, extremism, drugs, alcohol and tobacco, profanity, vulgarity, harassment, bullying, violence and gore, and hate speech (relating, for example, to ethnicity or race, gender identity or sexuality, religion, or people with disabilities). Public-facing bot APIs should also be reviewed to assess whether they could be used by people outside your organization to create a bot that would engage in hate speech or otherwise reflect poorly on your organization.
5. Ensure your bot is reliable.
Ensure that your bot is sufficiently reliable for the function it aims to perform, and always take into account that since AI systems are probabilistic they will not always provide the correct answer.
• Establish reliability metrics and review them periodically. Consider what questions your bot needs to answer and rigorously test its performance and ongoing effectiveness. Because the performance of AI-based bot systems may vary from development to implementation, and over time as the bot is rolled out to new users and in new contexts, it is important to continually monitor reliability. Reliability signals can be developed to help drive decisions about when to pass the baton to a human, or when a bot should announce that it cannot perform the requested function reliably. If an AI-based bot system can determine that it has made a mistake, that fact should be communicated to the user.
• Be transparent about bot reliability. Particularly for bots operating in sensitive domains, you should make available information concerning the reliability of the bot, such as summaries of general statistical performance, performance under particular circumstances, or in the context of specific examples.
• Build traceability capabilities into your bot. When something goes wrong with your bot during a high-value interaction, it is important to have traceability (monitoring and auditing)
• Provide a feedback mechanism. Users will feel more comfortable with bots if they can provide feedback on their operation (and feedback is essential in any event, as with all product development work). Bots should actively ask for feedback. Set expectations as to whether the user will get any response to feedback provided.
• For sensitive uses, obtain domain expertise. If you are building a bot to deliver services in areas such as health, employment, finance or law enforcement, ensure that you obtain and take account of input from experts in these areas as you design and deploy your bot.
6. Ensure your bot treats people fairly.
The possibility that AI-based systems will perpetuate existing societal biases, or introduce new biases, is one of the top concerns identified by the AI community relating to the rapid deployment of AI. Development teams must be committed to ensuring that their bots treat all people fairly.
• Systematically assess the data used for training your bot. Systematically assess the data used for training your bot to ensure that it has appropriate representativeness and quality, and take steps to understand the lineage and relevant attributes of the training data. As bias detection tools become more broadly available, adopt them as an additional means to ensure the fairness of your bot and make such tools available for customer use and adoption.
• Strive for diversity amongst your development team. Employing a diverse team focused on the design, development and testing of bot technology will help ensure that your bot operates fairly.
7. Ensure your bot respects user privacy.
Privacy considerations are especially important for bots. While the Microsoft Bot Framework does not store session state, you may be designing and deploying authenticated bots in personal settings (like hospitals) where bots will learn a great deal about users. People may also share more information about themselves than they would if they thought they were interacting with a person. And, of course, bots can remember everything. All of this (plus legal requirements) makes it especially important that you design bots from the ground up with a view toward respecting user privacy. This includes giving users sufficient transparency into bots’ data collection and use, including how the bot functions, and what types of controls the bot offers users over their personal data.
• Inform users up front about the data that is collected and how it is used and obtain their consent beforehand. Provide easy access to a valid privacy statement and applicable service agreement and include a “profile page” for users to obtain information about the bot with links to relevant privacy and legal information. Making this information available and easily accessible in the “first run” experience is highly recommended.
• Collect no more personal data than you need, limit access to it and store it for no longer than needed. Collect only the personal data that is essential for your bot to operate effectively. If your bot will share data (such as with another bot), be sure only to share the minimum amount of user data necessary in order to compete the requested function on behalf of the user. If you enable access by other agents to your bot’s user data, do so only for the minimum time necessary in order to compete the requested function. Always give users the opportunity to choose which agents your bot will share data with and what data is suitable for sharing. Consider whether you can purge stored user data from time to time while still enabling your bot to learn. Shorter retention periods minimize security risks for users and will help to position your bot as privacy-friendly.
• Provide privacy-protecting user controls. For bots that store personal information, such as authenticated bots, consider providing an easy-to-find “Show me all you know about me” button, and similar buttons to “Forget my last interaction,” “Delete all you know about me,” and so forth. In some cases, such buttons may be legally required.
• Obtain legal and privacy review. The privacy aspects of bot design are subject to important and increasingly stringent legal requirements. Be sure to obtain both a legal and a privacy review of your bot’s privacy practices through the appropriate channels in your organization.
8. Ensure your bot handles data securely.
Users have every right to expect that their data will be handled securely. Follow security best practices that are appropriate for the type of data your bot will be handling.
• Establish secure development and secure operations foundations. Traditional secure software foundations are critical. As with any AI system, your bot should ensure proper authentication, separation of duty, input validation and mitigations for denial-of-service attacks.
• Your bot should be resilient. Design your bot to identify abnormal behaviors and prevent manipulation. Pinpoint “users” (who could in fact be malicious bots) who deviate from norms established by large clusters of other users — such as users who seem to respond too fast, don’t sleep, or trigger parts of your bot code paths that other users do not.
• Ensure the integrity of your training data. All AI systems must be able to distinguish between maliciously introduced data (which must be purged) and data that is merely rare, yet valid and potentially important. This is particularly critical for bots which employ automatic or supervised learning techniques.
• Obtain security review. If available, work with the appropriate security team within your organization to conduct a security review on your bot and supporting services. Given the close relationship of security and privacy in this space, a joint security/privacy review is recommended to ensure the best depth and breadth of coverage.
9. Ensure your bot is accessible.
Bots can benefit everyone, but only if they are designed to be inclusive and accessible to people of all abilities. Microsoft’s mission to empower every person to achieve more includes ensuring that new technology interfaces can be used by people with disabilities, including users of assistive technology.
• If you are developing a bot, consider how your bot complies with commonly used accessibility standards, such as WCAG 2.0 AA, and U.S. Section 508 and EN 301 549 standards. Customers with disabilities should be able to use your bot as effectively as those without disabilities. Complying with the international web accessibility standard WCAG 2.0 AA (codified as ISO 40500:2012) and U.S. and European procurement standards will help enable users who rely on screen readers, navigate UI using only keyboard, are hard of hearing, require color contrast or cannot distinguish between colors, to use your bot. Many of these requirements carry dependencies on the conversational canvas.
• Have people with disabilities test your bots. In addition to complying with accessibility standards, getting feedback from users with disabilities on your bot prior to launch will help determine whether the bot can be used as intended by the broadest possible audience.
10. Accept responsibility.
We are a long way away from bots that can truly act autonomously, if that day will ever come. Humans are accountable for the operation of bots.
• Developers are accountable for the bots they deploy. If you are developing a bot that your organization will deploy, you should recognize that you are fully responsible for its operation and how it affects people. If you are designing a bot to be deployed by a third party, come to a shared understanding with them of who is ultimately responsible for the bot and document that understanding.
The intangible nature of AI technology can make it difficult for executives to build a business case for investing in new projects. Two examples show how well-thought out, high-impact projects can help with this.
The difficulty of making a business case with AI
The excitement around artificial intelligence has reached a tipping point, with its presence across sectors making it the primary battleground for technology vendors. But, despite the desperation on the part of companies, accelerators, and VCs to find a foothold, this vibrant market remains more conceptual than one of tangible substance. This makes an investment in AI technology a step into the unknown. A business case for such projects is not easy, needing reliance on intuition rather than ROI figures. Taking risks by investing in one or two high-impact scenarios can be very rewarding. The article covers two instances of how major organizations are doing this.
Sizing up the opportunity
AI requires using large amounts of data smartly, which the global law firm Linklaters is doing by turning its 175-year-old knowledge base into a competitive advantage. AI can create more sophisticated approaches for searching through this knowledge base, helping lawyers with quick information regarding legal precedents and previous projects. Linklaters’s CIO expects the ability to digitize and search contracts for key legal themes to become commonplace very quickly. The firm has already created an AI working group to analyze services in the marketplace and to work out how these technologies might impact the business.
But this change in how lawyers work also involves a cultural challenge. Senior partners will have to start trusting computers to do the same kind of work in seconds they have traditionally relied on associates to get done after spending hours with legal documents. Among other things, it’s the reputation of the lawyer and the firm on the line.
Using data to save lives
Moorfields Eye Hospital NHS Foundation Trust is involved with DeepMind Research, a project that involves the Trust sharing a set of one million anonymized eye scans. The hope is that these historic scans will improve future care, and lead to discoveries that make early detection and reduction in preventable eye disease possible. Challenges related to data security and confidentiality make it difficult to use non-anonymised data, which is actually more useful if demographic information is to be used to inform patient care. But stakeholders trust that similar projects will eventually lead to significant change in terms of how humans look at AI.
The future with AI
These experiments show that the potential of this self-learning technology is exceptionally exciting, and should encourage everyone involved in IT to investigate its uses. What also emerges is that, contrary to reports, automation does not simply lead to job cuts, but can create a new range of data science roles.
7 steps to become an AI-enabled enterprise
While platform companies – firms like Google, Amazon, and Alibaba – continue to collect tons of data and train their systems using information on consumers’ lifestyles, most enterprise leaders seem to underestimate the effect this will have on their businesses. A general AI strategy is needed for old world companies to keep up with this competition.
Challenges for the established economy
Apart from their own inability to change effectively and the intra-industry competition, large companies also face an underestimated challenge in the form of stiff competition from platform companies in three main ways. These are:
The ability to "burn" money
Platform companies have more financial resources they can leverage to invest with no urgency to prioritize budgets. In contrast, companies from the old economy have very little money to play with since they are constantly subject to pressure from the capital market, shareholders, and customers. For example, if Pfizer or Bayer were to invest USD 500 mn in cancer research – which is related to their core business – and come up with no drug, their CEO would be fired immediately. Google or Alibaba, on the other hand, could burn money on the same cancer research – a non-core project – and move on. This is coupled with the mindset companies like Amazon have to take risks and invest massively in R&D, including on untested AI technology, which is lacking in old-world companies even when they do have the cash to spare.
The strategy to hijack direct customer relationships
Today, as customer engagement, advertising and sales have moved on to platforms like Facebook, Google, and Amazon, these companies can easily hijack the direct relationship between established companies and their customers. As an enterprise owner, you might get analytics and some data from platform companies, but they can then use the same data and AI to enhance product recommendations, customize shopping experiences and improve targeting to offer goods and services from a pool of brands that keep customers from switching.
The power to collect massive data for building general AI
The range of products and services offered by platform companies – think Amazon’s marketplace, Kindle, Echo, and more – helps them collect endless amounts of data and thus create their own general AIs. These general AIs can then be deployed in different areas like finance, shipping, entertainment, and healthcare, further cementing the companies’ connection with a customer. This disruptive approach is not possible for enterprises from the old economy..
AI is the strongest tool to overcome the threats
While a strong brand, outstanding services, and innovation can help in survival, turning exponential is the only way to successfully compete against companies that are already exponential and trying to touch billions of customers across industries. AI is possibly the only tool today that can help ward off the competition, making use of the strong side of established companies: their experience. Their own general AIs can help established companies run processes autonomously across their organizations while retaining their knowledge and monetizing their data and experience. But, given the relatively limited breadth of data with these companies, a pooled approach would work better. Here is a step by step strategy to execute this idea:
Established companies collect every piece of data within the company.
Give data to a secure and independent intermediary platform operating a shared data pool for an established economy.
In return, they get access to a shared pool of aggregated and semantically organized data.
They use this data and the necessary technology to build their own corporate general AI.
Outcome-based on corporate general AI - New business models, offerings, services, etc.
Give resulting data from new business models, offerings, services, etc. to the shared data pool.
Adapting this approach can allow companies to keep their intellectual property and create value from their experience.
AI-enabled enterprise: anything that is a process can be and will be run by AI
Any process that can be automated today can be run by AI, leading to massive savings in money and manpower. This also spares resources to innovate at a faster pace. The following seven steps can help your company become an “AI-enabled enterprise”:
Create a semantic map of your data - accept continuous data flow as a foundation for future strategy.
Automate your IT operations - automate IT operations to receive immediate value brought by AI and to collect data; also make them autonomous.
Rethink your strategy - think about a new (exponential) business model.
Retrain your entire organization top-down - prepare and train your organization for an AI-enabled enterprise and for accepting a new business model.
Expand autonomous operations to other business processes - use company knowledge gathered through IT automation to make more processes autonomous.
Embrace predictive analytics - use data from the semantic map to expedite, improve business processes and future business events.
Consider data-driven processes - use data and AI to generate outcome-based processes.
Strengthen your core, widen your horizons
All data collected in the entire company eventually ends up in the IT environment: stored in applications, databases or storage systems. Thus, it is recommended to start setting up an AI in the IT environment – the core of your enterprise. Most importantly, come out of your comfort zone – your existing business and the industry you operate in right now. It doesn’t help you to simply develop something for the media industry when you are a media company. You have to think across industries and develop something that adds an additional value for your customers and thus opens a new industry for you.