The above graphic is from a Deloitte report on how humans might work with automation in government - but what’s interesting about it are the seven relationships it suggests we could have with smart machines, which are subtler than the usual “they’ll completely replace us” scare-stories. The authors outline what they mean below, with work examples taken from the perspective of US government and public services

Shepherd: A human manages a group of machines, amplifying their productivity

Here are some ways in which humans can serve as shepherds to machines in the workplace:

A human manages a fleet of autonomous buses. The buses behave as a swarm, attempting to maintain separation between vehicles and keep themselves on schedule, while the human monitors the buses to identify problems or issues they need to step in and resolve.
A nurse manager oversees a group of hospital robots. Several hospitals have piloted the use of robots for tasks such as transporting medication, supplies, and test samples within the hospital. A human can supervise and manage the robots, changing or reassigning tasks and schedules as required.

Extend: A machine augments human work

Humans and computers combine their strengths to achieve faster and better results, often doing what humans simply couldn’t do before, for example:

A department of human services uses cognitive technology to help predict which child welfare cases are likely to lead to child fatalities. The department uses machine learning to predict which cases carry the highest risk. Once high-risk cases get flagged, they are carefully reviewed and the results are shared with frontline staff, who then choose remedies designed to lower risk and improve outcomes. This process helps field staff target investigations based on risk rather than on random samplings.4
Cognitive technologies can be used to scan through vast amounts of medical data to help doctors make faster and more accurate diagnoses. For example, Google’s Deepmind algorithm analyzes 3D retinal scans to detect signs of eye disease. And IBM’s Watson for Oncology is designed to help physicians make more fully informed decisions by recommending individualized cancer treatments, citing evidence and a confidence score for each recommendation.5

Guide: A machine prompts a human to help them adopt knowledge

Machines help humans learn new knowledge and skills, or adopt desirable attitudes and behaviors. Here are two scenarios:

Adaptive learning. The US Air Force, working with a startup called Senseye, is redesigning its pilot training program, using VR simulation. The system tracks factors such as cognitive load levels, stress levels, and a pilot’s ability to plan ahead and strategize. As Senseye’s Founder, David Zakariaie, explains, “The AI will build a custom syllabus for each pilot based on what’s going on in their mind.”6
Digital research assistant. A researcher can set up a custom assistant that not only knows what current research a person is doing (based on their writing and speaking on their technology) but can also crawl the web for old and new research relevant to the topic that the researcher might not be aware of. This allows researchers to conduct literature reviews and stay up to date on the most recent advances much more quickly, accelerating their learning.

Collaborate: A problem is identified, defined, and solved by human-machine collaboration

Humans and machines collaborate to create solutions superior to those that either could create on their own, for example:

An AI-enabled chatbot or interactive voice response (IVR) system could work together with a human services caseworker to provide services to a client in need. As the employee engages with the client on the phone, the AI system could transcribe the conversation, automatically flagging relevant information as it comes up. This would allow the human to focus on the conversation and contextual clues, while the AI would simultaneously suggest tactical solutions.
A mobility manager who oversees a city’s transportation operating system. Most of the time, the system would operate autonomously, while the manager monitors the activity to make sure all is going well. But when problems arise, the manager would collaborate with the system to identify the problem and find an appropriate solution. Based on the situation and available information, the algorithm might recommend specific mitigating action, but it’s up to the mobility manager to accept that suggestion, or reject it and use a different tactic.7

Split up: Work is broken up and parts are automated

When a job is broken into steps or pieces, automating as many as possible, humans are left to do the rest and, when needed, supervise the automated work. Here are some current examples:

Ride-sharing. The machine assigns a driver to a trip, monitors their progress, collects a rating at the end of the trip (which is factored into the worker performance calculation), and handles payment. The human steps in to drive the car (at least for now).
Chatbots. A number of government agencies around the world, from the Australian Tax Office to US Citizenship and Immigration Services, use chatbots to answer basic questions. This frees up time for employees to respond to more complicated inquiries.9

Relieve: Machines take over routine, manual tasks

This is when lower-value, manual, tedious work is automated, creating opportunities to reduce cost and redeploy staff time to more valuable activities. Examples include:

The Food and Drug Administration’s Center for Drug Evaluation and Research (CDER) uses robotic process automation (RPA) in its application intake process. When CDER automated a part of the drug application intake process, it slashed application processing time by 93 percent, eliminated 5,200 hours of manual labor, and saved US$500,000 annually.10
San Diego County uses RPA to verify the eligibility of low-income applicants who claim benefits from government assistance programs. The software looks at the open forms on a caseworker’s screen, sifts through the verification fields, identifies relevant documents, and then pulls up those documents from another system, replacing a once-manual task with the stroke of a hot key. As a result, the county decreased the time it takes to approve a Supplemental Nutrition Assistance Program (SNAP) application from 60 days to less than a week.11

Replace: Machines completely perform a task once done by humans

Once staffed by humans, here are examples of entire jobs that are being fully automated:

Toll collection. Once performed manually by workers, collecting tolls has been automated in most states. In 2016, Massachusetts replaced 32 toll booth locations on the Massachusetts Turnpike with automated tolling technology.12
Mail sorting. The US Postal Service uses handwriting recognition to sort mail by ZIP code; some machines can process 18,000 pieces of mail an hour.13

More here.

We often worry that automation will just "replace" us. Here's a much subtler map of the many way we could "pair" with our machines