AI is adding value to every form of human endeavour, radically transforming work processes, building high degrees of operating efficiencies while human judgment steps in when AI reaches the limit of its capabilities.
Collaboration between Human Judgment and AI
In order to take full advantage of the collaboration between humans and Artificial Intelligence, humans need to develop capabilities which will help build the collaboration. Humans will need to learn how to train artificial intelligence to map decisions and decision-making processes of humans. Humans need to figure out ways to understand the logic of the decision made by the AI device and humans must make sure that the device does not act in a manner that will harm the humans or violate principles of fair play followed by humans that override considerations of ‘rational’ decision-making processes. In other words, Artificial Intelligence must be aligned to the human way of decision-making in order make its decisions workable in human society.
Making machines learn human decision-making processes
The first step to harnessing the power of a machine’s information processing ability to process information the way humans do is to train the machine. Machine learning algorithms that mimic human decision-making processes must be developed. A vast amount of data must be collated to mimic the database humans rely on while making decisions. For example, a language processing and translating application must be fed with expressions and idioms used by humans while communicating in order to avoid literal translations. Medical diagnostic applications must be fed data on symptoms of a disease, diagnostic and prognostic procedures. Financial decision applications must be fed data on financial indicators for decisionmaking, decision making logic and actual decisions made by humans in thousands of decision-making situations.
The Human Interface
AI devices must be taught human interfaces in order to discern information communicated by humans the way it is communicated to other humans and information to be communicated to humans the way one human would communicate with another. Interpreting stimuli generated by humans into instructions that must be processed by machines, and interpreting responses generated by machines into responses that can be comprehended by humans is a major aspect of training AI devices.
Creating human like responses
Koko is a startup incubated by MIT Medialab. Koko does not use standard phrases while responding to situations. If the user seems to be having a bad day, Koko does not respond by saying ‘I am sorry to hear that; It would seek out more information and is likely to respond by saying that the stress may be good and lead to stimulating actions with positive outcomes
Explaining decisions made by AI
AI decisions may not follow a logical decision-making process. AI decisions may be made mimicking human decisions made in a given situation which may be based on a logic of precedent rather than a logic of reasoning. Therefore, if the decision of the AI device is counter intuitive, it may need explanation. If an explanation is not available, the decision may be rejected as an erroneously made machine decision, in which case it would be overruled by a human decision based on reasoning. Hence machine decisions must be explained by a reasoning logic. A medical practitioner would not accept a decision on prognostics unless it is supported by data from diagnostics and a logic for arriving at a line of treatment. AI supporting a legal process will be required to explain the reasoning behind a sentence to ensure that the decision is fair and has been taken considering all aspects of the situation. A driverless car makes decisions which should be interpretable to ensure that there is no programming error which could lead to accidents. In order to protect the rights of consumers regulatory agencies are stepping in to mandate the requirement to explain decisions taken by businesses while making offers to customers. The European Union has legislated the General Data Protection Regulation (GDPR) which gives consumers the right to know why a certain offer has been made to a certain customer by a business. For example, an insurance premium, a credit card offer or the price of a department store purchase.
Aligning with Human Codes of Conduct: The role of ‘Ethics Sustainers’
AI system decisions must be monitored by humans to make sure that the decisions do not violate a human code of conduct. The decision may not be harmful for humans in any way even though they may be logically arrived at and are correct otherwise. The role of ‘Ethics Sustainers’ ranges from basic accident prevention to examining the outcome of an AI decision which may not be in the best interests of humans. The sustainers are also tasked with understanding the ethics of a decision. The AI algorithm may map human decision-making processes and make decisions which discriminate between people. The decisions may reflect bias against a group of people. The Ethics Sustainers would be tasked to prevent such biases. In order to make the right decisions, AI systems need to gather data from various sources. The collection of data for use by AI devices could infringe on the privacy of customers. Ethics Sustainers have a role in ensuring that data collected by the system to facilitate decision-making is not used in an unethical manner. Privacy of customers must be maintained.AI may learn about individual customers and customer groups in a statistical sense but it may never violate the privacy of customers by disclosing the data to third parties or use the data for any purpose other than the designated decision-making processes.
AI Assisted Manufacturing
Assembly lines are notorious for their inflexibility. The inventor of the moving assembly line, Henry Ford famously said about the Ford Model T- You can have it in any colour as long as it is black! Assembly lines have become much more adaptable since then. But the advent of the Cobot makes the Assembly Line truly flexible. Robots that are aware of their surroundings are called contextual robots or Cobots. In an industrial context work can be divided between human workers and Cobots according the skillets needed. Repetitive jobs requiring high effort, low dexterity and standard maneuvers are performed by Cobots. Tasks which require dexterity, judgment and are intricate in nature are performed by humans. The robot is incapable of executing the variations required in individual cars being assembled, when each car has a customized feature. The human assembly operator could execute the changes. But as robots take over assembly functions, this is no longer possible. But with the introduction of the ‘Cobots’, this has changed. At the M ercedes plant at Stuttgart, the assembly line is operated by Robots. The company has introduced a system of human pilots who direct the robot to execute customized tasks. The Robot executes the tasks that require hard labour and heavy lifting. The humans guide the robots to execute custom tasks. The human pilots are able to reprogram the robots on the fly using tablets. The demand for customization is fulfilled. At the Stuttgart plant of Mercedes, no two cars coming off the assembly line are the same. This is possible because the Robots are piloted and re programmed by human operators.
Wearable Robots (Excoskeletons)
Hyundai has created the concept of Excoskeletons. Excoskeletons are wearable robots which can perform repetitive tasks and tasks requiring extra ordinary physical effort. They are guided by the human wearing the device. The Excoskeleton interoperates with the human operator and performs the tasks demanding physical effort while the human operator takes of tasks requiring judgment and dexterity.
AI assisted Product Design
Artificial Intelligence can contribute to the field of creativity. Autodesk has an AI Application known as Deeamcatcher. The Dreamcatcher is an AI application used for designing furniture. Dreamcatcher is fed with hundreds of thousands of furniture designs of various furniture. If Dreamcatcher is given a certain set of specification such as the weight and size of the user and the preferred design type, the AI device can create hundreds of designs which even the most creative designer may not have thought of. Dreamcatcher not only does the computations needed to arrive at a design, it picks design types which would appeal to the customer.
AI assisted Customer Interface
Companies are experimenting with AI applications and looking for opportunities for scaling up.SEB, a major Swiss bank has an AI application called Aida. Aida is capable of handling natural language queries and can engage customers in conversation too. SEB created the AI application Aida initially as a help application for its employees numbering 15,000/- This was a low-risk application since a parallel human helpline war always available in the event of a failure of the AI entity. Once the application stabilized, SEB became confident of implementing it across its customer base running into millions of people. Aida has access to vast amounts of customer data and decision algorithms. Aida can ask follow through questions. It can detect tone of voice indicating emotions such as appreciation, frustration and concern. Aida is capable of understanding at what point the concern is not getting resolved or the customer is getting frustrated. At an appropriate point Aida transfers the call to a human representative.
AI for Administrative Process Automation
Consumer goods giant Unilever employs 170,000 people worldwide. It recruits thousands of people every year from a pool ten times as big. In order to shortlist candidates, the company has introduced an AI based application at the early stage of recruitment. The application invites candidates to play video games. The video games are programmed to discover traits such as risk aversion. Based on the candidate’s behaviour while playing the games, the application creates a shortlist of candidates. The shortlisted candidates are next interviewed by human selectors. The combination of the AI and human process significantly reduces human intervention and potential human error in recruitment. The application also serves to expand the pool of prospective hires. The process of applying for a job at Unilever is now vastly simplified. Applicants can download the recruitment application on their smartphones and navigate the first phase of the application online which requires no human intervention. The process has led to the number of Universities participating in the recruitment drive rise from 840 to 2,600. The number of applications doubled from 15,000 to 30,000. The time required for processing the application has dropped from sixteen weeks to four weeks. The time recruiters spent on the process has dropped by 75%.
AI for Rapid Response
In many businesses, speed is of the essence. In credit card use for example, the system needs to determine whether a transaction needs to be accepted or declined within seconds. If the transaction is fraudulent, the credit card company will have to absorb the loss. If it is genuine, declining will lead to loss of revenue and customer dissatisfaction. HSBC has created an AI enabled system which monitors credit card use patterns based on customer use pattern, location, IP address and other information that profiles the customer. Any deviation from the pattern signals possibility of a fraud. This is detected in nanoseconds.
AI in Agriculture
Crop and seed giant Monsanto and IBM have collaborated to identify optimal conditions of temperature, weather, soil conditions, precipitation, seed variants, fertilizer and pesticides to arrive at the right combination of inputs in order to maximisre crop yield and revenues. The analytical complexity of such optimality computations is very high and the application of conclusions so derived may be difficult to achieve. AI scientists have arrived at a different route to the situation. Farmers spell out the variables they need to optimize in order to maximise the gains from their efforts. The information requested may be limited to a few options of crops vis a vis inputs such as nitrogen content of fertilizer and soil conditions. The analytics could be restricted to a limited number of variables, making the process faster and far less complex. This is an instance of human knowledge and decision-making process being complemented by AI and analytical tools.
AI assisted Office Procedures
Cortana is a voice-enabled virtual assistant developed by Microsoft to help Windows 10 users initiate requests, complete tasks and anticipate future needs by surfacing relevant data in a personal context. Cortana can facilitate interaction between two people. Cortana can transcribe a meeting and distribute minutes that can be searched by voice. Those who could not attend the meeting can find out what discussions took place on a specific topic they were interested in.
AI in maintenance of sophisticated machinery
General Electric produces complex equipment such as jet engines, wind turbines and power turbines. These equipment are subject to preventive maintenance which means that the machines are opened up periodically to verify that their worn-out parts are replaced, and malfunctions are avoided. GE has created a vast database on aberrant and normal performance indicators for all these machines. The operating system is fed with data on indications of potential wear and tear of parts and symptoms that are likely to indicate malfunction. The Artificial Intelligence System is able to predict whether a malfunction is likely. Based on this prediction engineers are able to decide if the preventive maintenance routine is required or not. The application is able to avoid unnecessary downtime due to preventive maintenance routine not required and avert breakdown by interpreting the signals that predict malfunction. Maintenance crew have information on potential malfunction in real time, vastly improving response, reducing downtime, averting breakdown and reducing costs.
AI for Personalized Customer Experience
Customising individual experiences is a coveted goal of any marketeer. Artificial Intelligence is able to discern customer preferences based on past purchases. The customer can be recognized from the phone number, computer IP Number or even by face recognition software. Starbucks uses AI to recognize customers from their phone numbers and baristas are able to make recommendations based on the past orders of the customers. The AI application sifts vast quantities of data and throws up a set of options. The human operator picks the recommendations that are likely to be best suited to the customer at that point of time.
Brand specific response of interactive AI devices
Advanced AI devices are trained to display human emotions and communication styles. Some of the most commonly used AI applications are Microsoft’s Cortana, Apple’s Siri and Amazon’s Alexa. Each application is trained by humans to use language to interact with users which reflects the personality of their respective brands. Cortana comes across as caring and helpful, Siri comes across as just that bit irreverent and Alexa comes across as friendly.
AI has the potential to transform every form of human endeavour dramatically.
AI has not realized its potential. The possibilities it can create are not well appreciated and the vast majority of organisations are completely unaware of the role it can play in every form of human endeavour. What is needed is evangelists who will help disseminate knowledge of capabilities of AI and technology application professionals to build the bridge between prospective users and developers of AI applications.
Authored By:
Dr. Amit Bhadra,
Vice Chancellor,
Woxsen University