Artificial Intelligence in Aviation. What is it and when is it coming? (Teste seu Inglês)
There are currently more than 1700 Artificial Intelligence (AI) start-ups with over $14.6 billion in total funding from 70 different countries. AI revenues are expected to reach $47 billion by 2020, from $8.0 billion in 2016. Let's take a rationale approach to demystify AI in the aviation context. What is it and when is it coming? Is it a force for good? Are the 63 million aviation jobs at risk?
AI refers to computer programs that exhibit human-like intelligence such as logical reasoning, problem solving and learning. It comes in embodied (e.g. Robots) and disembodied (e.g. Apple Siri, Google Now) form.
Evolution of AI
1st Generation AI: Rule Driven Reasoning
It started when humans succeeded in incorporating knowledge into computer programs through the definition of static rules. For example in the early days; chess playing computer programs worked through a defined set of rules, and were able to perceive a very small portion of the outside world (the chessboard) in a limited domain; With no ability to learn and abstract knowledge.
GPS navigation software is another form of 1st Gen AI. It's able to perceive the location, analyze and contextualize it on a map, and conclude with directions. It has very limited ability to cope with deviations from the routine. 1st Gen AI computer programs are still alive and kicking; delivering value to many industries.
Some existing aviation use cases include:
Flight instruments used in an aircraft cockpit to inform the pilot of the state of the aircraft, e.g. relative to the horizon.
Autopilot feature to steer the aircraft along a predefined trajectory without the need of pilot intervention.
Automatic aircraft cabin pressurization to ensure the cabin environment is safe and comfortable for passengers.
2nd Generation AI: Learning through Big Data
Having access to lots of data, otherwise known as Big Data, has its challenges, however technological advancements in data management practices have turned the challenges into opportunities. 2nd Gen AI programs have leveraged this to learn from Big Data; Applying transformative algorithms (e.g. Deep Learning architectures such as Neural Networks) to identify clusters, and patterns in datasets. Neural networks is essentially forcing the computer to find patterns through trial and error.
2nd Gen AI is characterized by learning through analysis of data. but it lacks the ability of logical reasoning, understanding the context and abstracting knowledge to different domains. The learning is far from how humans learn. It's rather learning through well-defined statistical models that simulate the problem domain, and use data to constantly train the model. 2nd Gen AI can even be very unintelligent. Microsoft's twitter bot is a good example of how distortion in the training data can lead to unwanted results.
Some potential aviation use cases include:
Analysis and prediction of passenger behavior/demand.
Seamless airport security processes, through facial recognition and biometrics.
AI providing support to the optimization of revenue management, route network, fleet management, and pricing strategies.
3rd Generation AI: Analytical Awareness
2nd Gen AI needs large amounts of training data to be able to learn, and draw conclusions. If questioned about a conclusion, the answer is almost always: "because the data says so". On the other hand 3rd Gen AI is aware of the analytical path and to a very limited degree the context of the analysis.
3rd Gen AI doesn't need to see thousand example images of different aircraft damage, to be able to detect one on an aircraft.
4th Generation AI: Contextual Awareness
Understanding context and being able to go from one domain to another; abstracting knowledge and learning without the need for large amounts of training data are typical characteristics that separate 4th Gen AI from the previous generations.
Some potential aviation use cases include:
Autonomous airport processes, e.g ground handling, loading, fueling, cleaning, and aircraft safety checks.
Autonomous aircraft and in-flight services.
AI driving the optimization of revenue management, route network, fleet management, and pricing strategies.
Impact on Aviation Jobs
The aviation use cases are very promising in terms of more efficient operations across the value chain. However if this kind of automation takes place, it's almost certain that some jobs will be impacted. Considering that the aviation industry supports around 63 millions jobs worldwide; it's important to understand the changes that are expected and be ready.
The net change (jobs replaced vs. created) in the number of jobs is very difficult to predict. It's a common misconception to think that we know the timelines of when AI will supersede human intelligence. Realistically we only know the different possible scenarios for which we can try to be ready.
In the previous industrial revolutions jobs were indeed impacted and replaced with other types of jobs. In some cases technological advancements have resulted in the creation of new markets with significant increase in the net no. of jobs supported. Commercial aviation is a good example.
Whilst technology is unlikely to put the net number of jobs at risk. It can cause distortion in the demand for certain job types. For example in the years to come as AI is rolled out by airlines, airports and ground handlers, the following types of jobs are at risk:
Physical jobs that are repetitive in nature.
Data collecting and processing jobs.
Less at risk further down the line are:
Physical unpredictable jobs.
Jobs that require application of expertise.
And the most difficult ones to replace are:
Jobs that involve emotional interaction with people.
Convergence of Humans and AI
Often forgotten is that human intelligence is also evolving; even in modern times our intelligence has risen by 20 IQ points since 1950; and we are using technology and tools to enhance our capabilities. Elon Musk recently announced ambitions to use technology to enhance the human brain; starting with wireless connectivity.
The future is likely to go towards a convergence scenario between humans and AI to the point that one is indistinguishable from another. Imagine a human enhanced with robotic and AI features. Is that very different than a robot enhanced with human features?
Aligning AI Goals, Strategies and Values with Ours
AI can have a goal. Think of the aircraft autopilot. Once the goal of the pilot is set and aligned with the autopilot system; the machine takes over and fulfills the task towards that specific goal.
For more advanced AI systems it's imperative to define:
Goals that are aligned with those of humans to ensure that AI does things that humans want.
Strategies that are aligned with ours to ensure that the right method is used to achieve a goal.
Values aligned with ours to ensure that the right ethical framework (principles, standards, behaviors) is applied when devising a strategy and a plan of execution.
Bringing it all together
While it is inevitable that advanced AI is on the horizon; there seems to be consensus that the timeline is not as aggressive as a few years, nor is it as known as argued by some. And just because something can be replaced by AI, doesn't necessarily mean that it will be before there is a positive business case and the willingness of humans. When 4th Gen and even more sophisticated AI is around the corner; the aviation landscape and the role of humans is likely to be very different than today. If at some point in the future AI is more intelligent than humans; it will be interesting to see how the alignment with our goals, strategies and values will unfold.
Questions (Answer the questions below in the comments section.):
1) In your opinion what do you think will be the impact of A.I. on aviation jobs?
2) What is you solution to these problems?
3) What other ways do you think we can better implement A.I. in the future?