If we look at the history of AI, we can say that the field of artificial intelligence was born in 1956 on a conference in at Darmouth College. Organized by John McCarthy it involves people like Claude Shannon, Marvin Minsky, Arthur Samuel and others. Minsky wrote about neural networks with Seymour Papert. Ideas like expert systems – wherein computer contain deep stores of “knowledge” for specific domains, can be tracked to event. Arthur Samuel approach to that was developed in 1959 to asses probability of checkers moves was coined “machine learning”. But it took some time before machine learning started to develop, since AI development started to focus on symbols and relationships, that ultimately proved un-successful. In 1990 machine learning begin to flourish with practitioners integrating statistics and probability theory into their approaches. Now machine learning has finally broken out of research lab and it is fast becoming “the cornerstone of business disruption” according to Danny Lange, former head of machine learning at Uber.
New types of collaborative robots are equipped with the ability to sense their environment, comprehend, act, and learn, thanks to machine-learning software and other related AI technologies. Because of that working process is more flexible, self-adapting and teams of machines and humans can be set up more quickly and with great flexibility. Companies are now reaching cross-roads in their use of AI, they can use it as in the past as automation tool, that will bring smaller gains in productivity or they will use it in a way to create real breakthrough. New use of AI enables more adaptive human + machine teams that can improve work processes.
Machines are best at: performing repetitive tasks, analyzing huge data sets and handling routine cases. People are best at: resolving ambiguous information, exercising judgment in difficult cases and dealing with dissatisfied customers. This kind of cooperation can be called third way of business transformation. First two being standardization and automation. This third way involved adaptive processes. They are more flexible, faster, and adaptable to the behaviors, preference, and needs of company workers at a given moment. This adaptive capability is being driven by real-time data rather than by an a priori sequence of steps.
In human-machine relationship there is a missing middle field where humans and machine are cooperating. If we look at fields each party is taking care of, we can classify them into:
- Human-only activity: Lead, empathize, create, judge
- Human and machine hybrid activities:
- Humans complement machines: Train, explain, sustain
- AI gives humans superpowers: Amplify, interact, embody
- Machine-only activity: Transact, iterate, predict, adapt
Companies that are successful already today with use of AI and have moved beyond only automation are following five principles that are connected to their organizational mindset, experimentation, leadership, data and skills (MELDS).
- Mindset – assuming radically new approach with using missing middle, people improving AI and machines give humans more power.
- Experimentation – actively finding potential for testing AI and to actively use reimaged process with use of missing middle.
- Leadership – making a commitment for responsible use of AI.
- Data – building a data “supply chain” to fuel intelligent systems.
- Skills – develop new “fusion skills” in order to enable reimagining processes.
Factories have been strong area of second wave with automation taking jobs away from humans. But third wave has different perspective, it can improve working situation. AI is freeing up time, creativity and human capital. They are jobs and task that will continue to shift to robots based on their comparative advantages in handling repetition and data processing power. But before robots and people working together was a dangerous endeavor, since robots with their speed and power can hurt people. With new lines of robots that have sensors and are programed to detect and avoid people, their cooperation is much more reliable. Safety and efficiency are main drivers of automation in factories.
One of the best companies for building robots is owned by Rodney Brooks and names Rethink Robotics. He believes that instead of loading robots with predetermined symbols and relationships between symbols, we should use sensors to allow robots to learn about the world them self. This approach that follows “two AI winters” – periods from 1950 to 1970 and in 1980’s when money was invested into AI development but run out, since no major achievements were presented, is now bringing results and new investments that started after year 2000, are focusing on supporting it.
Use of AI in factory is connected with maintenance (GE use AI-enabled system called Predix), faster machine onboarding (Sight Machine from San Francisco develop solutions to help companies adding new machines to factory floor), unmanned vehicles (drones – BHP Billiton, Fortescue Metal Group, deep-sea robots – Boeing), reimagined product development (use of data to make improvements in developments), reimagined operations, for warehouse and logistic (Amazon acquired Kiva Robots in 2012, Symbotic robots can adapt to shapes and sizes and act accordingly).
Smarter warehouses are just beginning. AI technologies are now enabling entire supply chains to become increasingly intelligent. One task that AI can optimize and will bring enormous benefits to companies is demand planning.
AI technology will also play important part in food production industry. Precision agriculture – which leverages AI and fine-grain data about the state of crops – promise to improve yield and reduce wasting of other resources needed. Use of IoT devices and smart algorithms will enable precision agriculture.
AI can serve in many repetitive works. Backoffice processes include a lot of them. So instead of training humans to work like robots, we should aim for workers to be more human. In order to do that, a lot of processes will be redesign in light of use of new capabilities that AI have. If we look at the process of categorizing and resolving complaints, we can see that a lot of tasks are general and repeated, but in order to automate them, we should take care of structuring inputs from customers that are usually very individual and unstructured, but with machine-learning platforms that can work on natural-language processing, those inputs are adjusted to computer work. Every company has mass of behind-the-scenes activities. The introduction of AI can help offset the burden of repetitive, low-visibility tasks so that employees can focus on higher-value tasks. Companies like Goldman Sachs (influence on share prices), Woodside Petroleum (sharing knowledge), Huffington Post (flagging comments, span and abusive language) and Arizona State University (personalized tutor) use AI for back office processes.
RPA – robotic process automation, is software that performs digital office tasks that are administrative, repetitive and mostly transactional within a workflow. But in order to reimagine processes, companies need to use AI in order to merge computer actions with human supervision and final decision – fraud detection system in banks, Unilever’s hiring tool. Repetition, replication and redundancy. If these elements show up in your business operations, it’s a clue that tasks or processes are ready to be changed. Gigster is a company that uses AI to develop software based on best usage of AI principles to address above mentioned issues. They use AI to check standard features of software that they have catalogued based on previous projects, AI then create quote estimation based on previous experience that helps explain project complexity. Then if project is excepted task are delegated to team members based on AI estimation. Swedish bank use customer facing AI assistants only after they tested it as internal virtual IT assistant. One industry that will be affected by self-learning ability of AI is IT security since automatic systems become more and more advanced and move from static response tools to adaptive proactive tools.
If we look at AI technology, we can classify it in three layers:
- At core is machine learning, that can be deep or shallow and includes
- Reinforcement
- Supervised learning
- Unsupervised learning
- Semi-supervised learning
- Than we have AI capacities, what it can do:
- Knowledge representation
- Computer vision
- Natural-language understanding
- Speech to text
- Expert systems (inference)
- Predictive systems
- Speech and optimization
- Audio signal processing
- And top layer is AI applications:
- Intelligent agent
- Collaborative robotics
- Biometrics, facial and gesture recognition
- Intelligent automation
- Recommendation systems
- Intelligent products
- Personalization
- Text, speech, image and video analytics
- Extended reality
Good example of how AI moved into R&D processes is Tesla. They build in their AI platform for autonomous driving, but it will only learn from drivers until it will consistently perform better operations than driver and when that happen, AI will be ready to take over.
Science in age of AI needs to adjust its clear steps approach with observation, hypothesis, testing, generalization. AI bring on the fly learning and adaptivity. Human researchers are very good at creative insights, machines are unequivocally better at data organization and presentation, especially when data volume becomes unwieldy. One company Quid is using AI to reimagine find and search part of science process. GNS Healthcare is exploring options to generate hypothesis automatically. They are looking for adverse reactions between drug combination, using data from Medicare for seniors. Usually this kind of tests are hard to run, and results are based on intuition of researchers. System was called REFS (Reverse Engineering and Forward Simulation). In the design phase of R&D AI can help in compressing time for testing with ability to simulate different solutions. AI now enables even more rapid analysis of customer preferences, allowing for personalized and customizable experiences. Many trends – including the increasing availability of customer data – are leading to a new level of product customization and delivery. But even with all this AI potential of fast and painful simulations and testing, we need to be careful to follow ethics and protect privacy.
Sales and marketing are two strong areas where use of AI can have direct revenue and profitability impact. Coca-Cola is using Einstein – AI based on CRM from Salesforce to manage coolers around their partners. It uses picture recognition, weather forecast, promotional offers, inventory levels and historical data with seasonal affects and other factors to make proper order recommendation. Three main functions of customer interactions are sales, marketing and customer experience. And AI is changing all of them. New modes of human-machine interactions change the way companies deliver goods and services.
Using AI to optimize shops into customer-aware shops. Clothes retailers is using smart mirrors to bring different views in order for customer to see how they look, it has built-in function of stock and different colors of product and possibility to notify sales assistant to deliver it to dressing room. Retailer is on other hand receiving feedback from customers about length of trying, conversion rate, products tried. Another example of use of AI is staffing and sales assistant management. One of those systems is called Percolata. Almax, Italian company has developed mannequin with computer vision and facial-recognition technology. They use it to adjust offering and sales assistance to customer base – since they can track and categorize it with their technology. This kind of development are coupled with ethics and privacy concerns.
AI gives salespeople and marketers the time and insights to cut through the high volume and opacity of digital interactions and letting them be more human. Some of the biggest changes in the front office are happening through online tools and AI-enabled interfaces. Sometimes when AI is taking over interactions with customers it can become face of the brand. For this reason, AI interactions can have both very good and very bad influence on brand image. That is one of the reasons why creators of conversational bots need to be careful to set proper boundaries and evolvement mechanisms.
With changing landscape of economy, rise of assistant, new services standard business models are facing extreme pressure and companies need to adjust to that changes. In a hyper-networked world where mobile phones, speakers, thermostats, and even exercise clothes are connected to the internet and potentially each other, brands have to learn to play well with each other or give up a certain amount of control to those that own the most popular interfaces. Amazon is using this to sell more than five million of their speakers called Echo. It is moving into “zero-click commerce”.
Since conversational bots are getting into more and more human interactions, one challenge they are facing is to improve their empathy responsiveness. One company MIT startup called Koko is currently developing a kind of empathy engine that could plug into any AI. It is trained on human responses.
Best practices from digital giants like Amazon, eBay and Google are reused by traditional companies, so they can become advance players in digital approach and proper content creators using inputs from enormous data pools. One example is Google using AI on millions of signals to determine proper pricing for their marketing tools AdWords and DoubleClick Search.
If we would to categorize strengths of human and machines, we could say:
- Human
- Creativity
- Improvisation
- Dexterity
- Judging
- Social and leadership abilities
- Machine
- Speed
- Accuracy
- Repetition
- Predictive capabilities
- Scalability
In future biggest growth in jobs will be in the area of human-machine cooperation. The most successful companies think of AI as an investment in human talent first and technology second. They value workers that are adaptable, entrepreneurial, and open to retraining. Companies that want to tackle AI use in proper way need to develop missing middle, rethink process fundamentals and have management that can address the challenges of reimagining process with an awareness of responsible AI.
If companies want to set their human-machine structure properly then they need to address missing middle challenge properly. In order to do that, new skills are needed, and roles that human plays in developing and deploying responsible AI, need to be defined. Companies need to rear their algorithms right. Melissa Cefkin from Nissan has a role of organizing smooth collaboration between human and machine. She said: “You need to understand humans if you want to provide them with automated partner.”[1]Her job could be called “vehicle design anthropologist”. New jobs will be needed to train, explain and sustain AI behavior. Questions like: where in your organization might you find these new jobs or how do they fit into existing and reimagined process, are what companies need to address.
We have two sets of training for machines:
- Task performance training
- Clean data for upload
- Discover relevant data and data streams
- Have machine observe decision making
- Tag data for better use
- Work with HR to inform the design of workplace retraining initiatives
- Humanness attribute training
- Train language, gesture, empathy
- Have machine observe interactions
- Correct errors, reinforce successes
- Define and develop personality
As a first step in training physical and software-based systems, company should consider using expert employees who already work closely with AI or with the system that will integrate AI as initial trainers. Advanced AI will learn from human trainers’ empathy, they will train it on personality. Also, trainers need to make sure AI systems are globally and locally acceptable. Sometimes training AI can be outsourced or crowdsourced. One of this companies that can do such tasks is Mighty AI.
As AI systems will only be as good as data they work on, new job of data hygienist is crucial. Not only algorithms themselves need to be unbiased, but the data they are trained on must also be free from any slanted perspective.
Another category of jobs will be needed to bridge gap between technology and business leaders, since sometimes nature of black-box approach is questionable for business leaders, especially when recommendations are against conventional wisdom. Those explainers will test, observe and explain algorithms and update interfaces to add explainability, but they will also interpret machine outputs into insights, making sense of outputs and explain machine workings to stakeholders. GDPR is actually introducing right to explanation for consumers, allowing consumers to question and fight decisions against them based on algorithms. Companies will have to develop internal knowledge called algorithm forensics analyses. But even before any need to conduct autopsies, companies should have a transparency analyst responsible for classifying the reason for particular AI algorithms acts as black box. Two other roles that are important for general use of AI in companies are explainability strategist and sustainer, one being responsible for making important judgment calls about which AI technologies might be best deployed for specific applications and second one continuously working on ensuring that AI systems are functioning properly as intended. Another role that should take care of intended work of AI is context designers. And controlling proper use of AI is a job for an AI safety engineer. Ethics compliance manager will take care of upholding generally accepted norms of human values and morals. Automation ethicists will evaluate noneconomic impact of AI systems. Machine relations manager will take care of machine business development and monitor their roles and mobility inside companies.
AI tools are empowering workers in a range of fields, from design to medicine engineering to factory-floor operations. This augmentation comes in a variety of forms – from augmented reality and virtual reality to analytics engines to robot arms and chatbots. AI augmentation and its reshaping of business processes is happening right now, across the three categories of human-machine interaction:
- Amplification – AI gives human data-driven insights, often using real-time data.
- Interaction – AI agents employ advanced interfaces like voice-driven NLP-ing.
- Embodiment – physical realm, AI using sensors, motors and actuators to allow robots to share workplace with humans. Cobots (collaborative robots) are one example.
Amplification activities:
- Matching
- Match resources, Q&A tasks
- Automate repetitive or low-level tasks
- Recommending
- Rank or design alternatives
- Prioritize resources
- Automate process change
- Patterning
- Identify trends in real time
- Personalize offerings
- Identify anomalies
- Categorize and route data
- Augment strategic decisions
Interaction activities:
- Administering
- Enable human workers to focus on high-value interactions
- Coaching
- Nudge the next best action based on domain expertise or corporate policy
- Accelerate understanding of customer and process context to solve problem
- Conversing
- Allow for voice-powered access to services and analytics
- Allow natural-language querying, commands, and sophisticated improvisation during dialogue
Embodiment activities:
- Navigating and extending
- Self-navigate around humans and autonomous machines
- Extend sight, hearing or touch
- Collaborating in physical space
- Assist on very precise, arduous or routine physical work
AI is moving from task replacement to process change. In organization AI augmentation opens up the possibility to fundamentally rethink business processes to uncover hidden gains, embolden workers, and discover brand-new business models for this new age.
For management to get into AI game, they need to work on MELDS as authors call them. When we talk about mindset, management needs to completely change its mindset. Total break with world we take for granted.
In order to do that, management can use three-step method: discover and describe, co-create, and scale and sustain. One methodology executive can use is design thinking or emphatic design. It basically means translating consumer needs into products. Identifying opportunities for reimagination takes time – executives must capture the current business context, distill insights from various observations, and identify the potential value impact of the reimagined process. Pursuing opportunities for process reimagination requires ability to envision work in the missing middle. Audi introduce Audi Robotic Telepresence (ART). Robots that help partners solve customers problem quickly and educate technicians at partner. Once you find solution, you need to scale it and sustain it with continual improvements.
When we talk about culture of experimentation, Amazon is a prime example. Like Jeff Bezos said: “I’ve made billions of dollars of failures at Amazon.com.”[2]The only way to push state of the technology forward is to test its edges. Experimentation is key, because we move from an era of standard business processes and companies can no longer strive to copy best-of-class companies. Culture that support experimentation is a culture that allows mistakes. After all, in science, an experiment that doesn’t support the hypothesis isn’t called a failure, it’s called data.
Main leadership task in AI era is establishment of organizational culture that promotes responsible AI. Role of explainers, trainers and sustainers is important, but positive attitude to technology from leadership is also important, since people tend to be sceptic about it. People tend to forgive mistakes of humans easier than those of machines. Managers should encourage a thoughtful balance of skepticism and acceptance amid complex changes wrought by AI. Managers can use safeties in order to control usage of AI. Those safeties can be: installing guardrails, use humancheckpoints, minimize ”moral crumple zones”, consider legal, psychological and other issues. Managers discovered that, without being given an explanation, people are more likely to trust human judgements than an algorithm’s recommendation.
Data is main source of AI development should be seen as end-to-end supply chain. It should be seen as a dynamic, enterprise-wide activity for capturing, cleaning, integrating, curating and storing information. It should be rich and big. But when starting with AI projects, companies should focus on small scale and simple AI challenges. When accumulating data, companies should be aware of obstacles, either technical or social and adjust to them. Another important challenge is sorting relevance and necessity of data and adjust technical structures to it. It is important that anyone, especially less technical business users can take advantage of insights from data. Firms should appoint a data supply-chain officer that will take care of coordinated structure, sustainability of data usage and would work on resolving all potential issues.
New business models based on AI use will need new skills. The concept of fusion skills is: abilities for combining strengths of a human and machine to create a better outcome that either could alone. Author identify eight novel fusion skills:
- Rehumanizing time – the ability to increase the time available for distinctly human tasks: interactions, creativity and decision making.
- Responsible normalizing – the act of responsibly shaping purpose and perception of human-machine interaction.
- Judgment integration – the judgment-based ability to decide a course of action when a machine is uncertain about what to do.
- Intelligent interrogation – knowing how best to ask questions of AI, across levels of abstraction, to get insight you need.
- Bot-based empowerment – working well with AI agents to extend your capabilities and create superpowers in business processes and professional careers.
- Holistic melding – the ability to develop robust mental models of AI agents to improve process outcomes.
- Reciprocal apprenticing -performing tasks alongside AI agents so they can learn new skills and on-the-job training for people so they can work well within AI-enhanced processes.
- Relentless reimagining – the rigorous discipline of creating new processes and business models from scratch, rather than simply automating old processes.
AI use is polarizing economy today, because it uses approach of humans either machines. But authors believe in missing middle concept where machine and humans are cooperating and improving performance of each other. They believe in doing different things and doing things differently. The problem they see is a skill gap on the digital side. Jobs are disappearing because people are not re-trained quickly enough to take on new, digital ready tasks. Reimaging business processes with help of AI use, will have even more negative effect if companies and society as whole will not work on re-training of workforce and potential workforce with new skills.
[1]In the book on page 113
[2]In this book on page 161
