We have been using tools for as long as we defined ourselves “human”, actually the use of tools pre-dates home sapiens, going back 2.3 million years ago at the time of homo habilis (we, sapiens, started to walk the world some 500,000 years ago). The Human-Tool interface was straightforward you don’t need to be taught how to use a hammer …. More recently, our tools have become more complex and versatile and we have developed a Human-Machine Interface -HMI that in many cases needs to be learnt. A lot of design work has been done to keep this learning to a minimum but in order to operate a feature rich machine (including a washing machine) some learning is required.
One approach to simplification of the interaction has followed the path of transferring functionality to the machine and using meta-functionality in the human machine interaction. When you drive a car you are using this paradigm. What you do is actually converted by on-board computers in commands to the engine, to the injection, to the suspensions and soon. Actually, we won’t stand a chance if we were to command directly the variety of systems at work in our car.
This has opened to door to an increased automation of systems, to the point that we are losing control on what is actually going on (see the example of MCAS on the Boeing 737 MAX).
The availability of sensors, data, processing power and actuators made possibile this increased automation. Actually, we are reaching (have reached in a few sectors) the point where the deluge of data that need to be taken into account is exceeding our capability of doing so. At the same time, this huge availability of data makes machines smarter and smarter. Deep Learning is leveraging on this abundance to increase the machine intelligence. This, in turns can evolve into into increased automation and decreased need for the HMI.
In these last two years this has led to a growing concern that:
- we are losing control (and understanding) on what is going on >> generating a claim for AI transparency, easier to say than done
- we are becoming obsolete, no longer needed and replaceable by machine (software) >> generating the fear of losing our job
- we no longer know who is to be held accountable >> fuzzier boundaries with new, yet unsolved, ethical issues
- we, as lay citizens, abdicate our role and power to data oligarchies >> data concentration in the hands of very few global companies
Now a new wave is rising, that claiming for Humans In The Loop, HITL, an evolution of the HMI paradigm that consider the coexistence, and aims to leverage, human and machines as component of the whole system.
An example in healthcare is Unanimous AI, a company that is weaving together the human intelligence with machine intelligence. Their tag line is “We amplify Intelligence”. An interesting article on NPJ is describing this new paradigm, a “Human Machine partnership with AI for chest radiograph diagnoses“.
Another interesting article “How managers should prepare for Deep Learning: New Paradigms” is applying the HITL to enterprise management.
The recent Viking Cruise ship that was buffeted by high waves in the Nordic Sea as result of faulty analyses of sensors data (on low oil level) is an example of HOTL – Humans Out of The Loop- and questioned the design approach to automating crucial components of the ship cutting out the humans from the decision.
Automotive companies working on self-driving cars (starting at level 3, so the ones that are already on the road today) are faced with the decision of moving under a HITL or HOTL paradigm. I already mentioned the Boeing 737MAX issues with MCAS. Here is an interesting article going in depth into design decisions and what did not work.
The problem is that if we look at accidents reports, both in automotive and aeronautics, we see that by far these are the result of human error. Hence, by removing the human factor from the equation we would have safer transportation systems!
The HITL however aims at overcoming this dichotomy between humans and machines (software) enabling a participation of both where the machine helps human in taking more informed decisions and the humans help machines in understanding the global context. Clearly this is a shift from an interaction based on syntax to an interaction based on semantics. We will be seeing this shift more and more, including in the way we are interacting with the cyberspace. Answers will be based not on what we ask, rather on why we ask…