We humans have been dreaming of artificially intelligent beings for quite a while. Enter Talon/Talos, a giant robot protecting Crete from pirates by circling her shores three times daily. But the relationships between “us” and “them” has mostly been one of conflict and dominance – for example in 1920s play by Karel Čapek ‘Rossum’s Universal Robots’ humanity is taken to extinction by the robots created by us.
k
Roll forward 100 years or so and the reality is slightly different. Our robots today are not “universal” but specialised in a small number of functions, where they excel but are largely incapable of sorting problems beyond the routine events at the workplace. And they are quite far away from being self-conscious never mind rebelling.
In this context, it seems the optimal arrangement is the one where we use robots to augment the human intelligence. AI can complement human analytical and decision-making abilities, with humans adding creativity to the collaboration. According to Wilson and Daugherty writing for HBR[1], smart machines can expand human capabilities in three ways; by amplifying our cognitive strength, by interacting with our customers using chatbots, and by embedding our skills in a cobot. Autodesk’s Dreamcatcher is an example of AI enhancing the creative imagination of exceptional designers. Cortana is a good example of human and AI interaction to facilitate communication between people. Cobots are AI machines that can work alongside humans in factories hence embedding human skills.
The last scenario is our focus at present – it would be great to have robots and humans work side-by-side, each playing to their strengths. This is also AI but now expanded into “augmented intelligence”, and the robots collaborating with humans are called “cobots”, or collaborative robots.
This shape of human-AI collaboration increases productivity by assisting workers in routine tasks, ensuring consistency and quality improvement. Human workers are there when there are problems to be solved.
There are some challenges in this arrangement however. If humans and cobots interact in shared spaces, we should ensure they don’t get in each other’s way. Cobots are mostly fixed, resembling a robotic arm, a technician is involved in the initial setup but at a later stage, the robot starts learning from human actions
Another challenge is ensuring that the human understands what the cobot is doing, and that they can control the behaviour to optimise results according to the changing context of work. This requires special effort in designing the control interface of the cobot, driven by UX principles but going beyond them.
Human-cobot teaming scenarios are rapidly explored nowadays with the aim to have a safe and seamless environment for collaborative tasks. One appealing application would be to use robots in environments which are not fully suited for humans, for example in very dusty circumstances or within a limited width and height tunnels, and have humans watch “over the robot’s shoulder”, coming to the rescue if something gets stuck. This will be robot in routine work and cobot when problems need to be resolved.
Such an application is envisioned in the RoBétArmé EU-funded project under grant agreement No 101058731. A movable robot will be executing the routine task of shotcreting – applying high-pressure mixture of cement, air and water to tunnel walls and ceilings to create even concrete covering. The environment is full of dust and the tunnels could be quite small in size, making the application of shotcreting nearly impossible for a human worker. At the same time, training a shotcrete operator takes 10 years and this is a very skillful job, with expert knowledge about how to lay the shotcrete but also how to resolve problems such as when one of the pipes bringing the mix to the nozzle gets clogged. In the remaining 36 months of the project we will endevour to prototype a collaborative human-robot system which can capture the best each party can offer, with the robot operating the shotcreting nozzle and the human operator helping out when problems arise (as they do).
More research needs to be done to understand mechanisms for human actions, if we are to have successful collaboration with AI systems.
[1] Wilson, H. J., & Daugherty, P. R. (2018). Collaborative intelligence: Humans and AI are joining forces. Harvard Business Review, 96(4), 114-123.