Introduction
In 2014, Geoffrey Hinton, Yoshua Bengio and others behind Google succeeded in combining neural networks, large quantities of data, and powerful computers to make breakthroughs in the field of AI.
AI or Artificial intelligence is the ability of machines to perform tasks commonly associated with intelligent humans. . In practice, AI is not one thing but a set of techniques that can automate tasks such as forecasting, recognizing images, recognizing speech, understanding text, detecting anomalies and other tasks previously reserved for human intelligence.
In real life though, human work is complex. It combines multiple tasks at different levels. A medical doctor making a diagnosis will gather information using multiple senses and techniques. He may talk with the patient, touch the patient’s skin, listen to the sound of his breath, ask about his age and weight, measure his blood pressure, look at the results of his blood test, look at his x-ray results, among other things. Each of these subtasks comprises diverse tasks elements.
To successfully automate some of the tasks, machines will use a set of techniques that may include a combination of the latest AI systems and classic software engineer methods.
Making AI projects work
Instead of starting with a set of AI techniques and trying to find problems to solve, cognitive engineers perform cognitive analysis first and then allocate tasks between humans and machines. Cognitive task analysis helps understand the mental task. A mental task is a set of goals and subgoals that invoke cognitive operations such as detecting, perceiving, analyzing, judging, recollecting, recognizing associating, to name a few.
Some of the task locations such as image recognition might invoke the latest AI techniques while others might require classical ruled-base programming or database management.
The success of the project lies in the design of solutions that involve a combination of techniques available.
Proven and tested human-machine design principles such as task compatibility, flexibility visibility, and control, are essential to the success of any solutions design to remain vital to its success.
For example, when automating tasks such as forecasting, the quality of the outcome depends on the quality of the models which correspondingly depends on the quality of the training data, and the quality of the inputs.
In real life, input data might be corrupted, new situations may make the model irrelevant. Naturally, the result will be wrong.
The design must factor in when things go wrong. Design principles such as visibility and controls become essential when automating.
Conclusion
Instead of an AI project, make the project an automation project. It is possible you might be automating tasks without using any of the latest AI techniques.
Prior to crafting a solution design, perform the cognitive tasks and process analysis.
Use all necessary techniques when performing tasks allocation between humans and machines including traditional techniques.
When automating tasks, provide visibility and user control using human-machine design techniques principles.