Indian Defence Review

Aviation is one of the most heavily regulated industries in the world due to safety reasons. Strict regulations have helped the aviation industry provide the safest way of transport per mile travelled. Aviation incidents are few and far between, and are getting rarer every year. Some degrees of automation have indeed helped get aviation to where it currently is. But human control and intervention have always been at the heart of it, from pilots to Air Traffic Controllers. This is about to change.

The development of computer systems to be able to perform tasks normally requiring human intelligence, such as visual perception, speech recognition, decision-making and translation between languages is termed as Artificial Intelligence (AI). Intelligent machine systems that can interpret complex data; perceive the environment and take appropriate actions using learning and problem solving techniques are termed to have Artificial Intelligence. Like humans, AI process includes perception, reasoning, knowledge, planning, learning, statistical analysis, computation, and finally manipulates output. It has evolved using expertise in fields such as computer sciences, mathematics, psychology, neuroscience, among many others.

AI applications already exist in industrial machines, automotive industry, surgery and aviation among others. On the other hand, AI is considered a great threat to human jobs and may result in high unemployment. Some people also consider AI a danger to humanity if it progresses unabatedly, and may one day start threatening human existence. There are also ethical and morality issues. For years, discussions have been binary – human intelligence vs artificial intelligence. Good vs evil. Hopefully this will change. Another school of thought is to use the phrase “extended intelligence” to signify how AI is used to augment human decision-making rather than to replace it.

Evolution

For centuries we have heard fictional stories about thought-capable artificial beings. Even if they did not exist, they were certainly thought about conceptually. The first calculating machine came up in the early 1620s which performed based on concepts rather than numbers. Since the 19th century we have seen research on Robotics. In the 1940s, Alan Turing’s theory of computation suggested that a machine could simulate any conceivable act of mathematical deduction. Soon digital computers began simulating any process of formal reasoning. Also concurrent discoveries in neurology, information theory and cybernetics led to consider the possibility of building an electronic brain.

The first work that is now is recognised as AI was when McCullough and Pitts’ in 1943 designed ‘artificial neurons’. Computers then began winning at chess and checkers, solving word problems in algebra, proving logical theorems and speaking English. Like in most other technologies, US DoD allotted huge funds in 1960s for AI. In the early 1980s, AI got a flip with the commercial success of expert systems. In the late 1990s and early 21st century, AI began to be used for logistics, data mining and medical diagnosis.

Deep Blue became the first computer chess-playing system to beat a reigning world chess champion, Gary Kasparov in May 1997. Advanced statistical techniques, access to large amount of data and faster computers enabled machine learning. By mid-2010s, machine learning applications were in use all across the world. IBM’s Question Answering System defeated the two greatest quiz champions by a significant margin. The Kinect which provides a 3-D body–motion interface for the Xbox One uses algorithms that emerged from AI research.

In 2015, the number of software projects that use AI within Google increased from just a few in 2012 to more than 2,700. In March 2016, AlphaGo won four out of five games of Go in a match with a Go champion. Increase in affordable cloud computing infrastructure has greatly helped. Microsoft’s development of a Skype system that can automatically translate from one language to another and Facebook’s system that can describe images to blind people have become possible due to AI.

Machine learning is a hot topic today in AI research. It is already used for tasks as diverse as decoding human speech, image recognition or deciding which adverts to show web users. The programmes work by using Artificial Neural Networks (ANNs), which are loosely inspired by biological brains, to crunch huge quantities of data, looking for patterns and extracting rules that make them more efficient at whatever task they have been set. This allows the computers to teach themselves rules of thumb that human programmers would otherwise have to try to write explicitly in computer code, a fairly time consuming task.

Training the AI Systems

A University College London (UCL) team has written what it calls an Intelligent Autopilot System that uses ten separate ANNs. Each is tasked with learning the best settings for different controls (the throttle, ailerons, elevators and so on) in a variety of different conditions. Hundreds of ANNs would probably be needed to cope with a real aircraft, says Dr Bentley of UCL. To train the autopilot, ANNs observe humans using a flight simulator. As the plane is flown – taking off, cruising, landing and coping with severe weather and aircraft faults that can strike at any point – the networks teach themselves how each specific element of powered flight relates to all the others.

AI in Aviation

AI is, today, a part of research by every think-tank and the subject of seminars and cocktail party conversations. Notwithstanding sceptics, aircrew can find reasons to be enthusiastic about AI which will enable airplanes to become proactive and make decisions, thanks to machine learning and neural networks. HAL 9000 (Heuristically programmed ALgorithmic computer) was a fictional character of Space Odyssey series with a computer controlling Discovery One spacecraft that interacted with the ship’s astronauts. The key part of HAL’s hardware was the camera lens located strategically. HAL speaks in a soft and calm voice and is capable of speech and facial recognition, language processing, lip reading, art appreciation, interpreting emotional behaviours and automated reasoning.

Aviation was an early adopter of AI. Most pilots have been flying with primitive forms of AI for years with autopilots, FADEC and load-shedding electrical systems all using computer power to make intelligent decisions. These came much before the autonomous cars. Research is on to apply real AI to an autopilot, beyond just programming it to fly pre-planned profiles. Machine learning is being used to make a more resilient autopilot that can adapt to changing conditions. Monitoring and storing hundreds of hours of detailed data from real flights, a library of scenarios can be created. A traditional auto-pilot may fail when an engine fails or the turbulence crosses a threshold, new designs will allow it to continue to be in operation. AI-powered autopilots will soon mature to absolutely new levels. A DARPA project called Aircrew Labour In-cockpit Automation System (ALIAS) aims to create a full replacement for a human co-pilot. It is a complete mechanical system that manipulates the controls of an airplane making it an autonomous airplane. Till now, all aircraft systems were to help pilot exercise control over aircraft and systems. The next step is real decision-making tasks. This will require deep machine learning and neural networks to create powerful algorithms that attempt to ‘think’ like a human.

AI is already being used to interpret MRIs in hospitals and next step could be to read an aircraft engine monitor. AI could mean comparing that engine signature to a database of millions of hours of engine data. It could also mean ability predicting and warning of impending failure based on digital engine monitor data. Weather forecasting is another place where AI will be of interest to aviation. Pilots need much more than just weather picture and diversionary airfield data. AI could decipher the radar image of weather and other data to make constructive suggestions, being predictive. It can analyse fuel consumption rate and winds and suggest flight route. Garmin’s Telligence is a product that uses voice commands to complete hundreds of common tasks in the cockpit. This is certified and available to install today and it will surely improve in years ahead.

Unmanned Aerial Systems (UAS) is a good place for developing practical, airborne AI. Quadcopters are already delivering goods and inspecting pipelines with no pilot to make decisions. Drones are already doing self-diagnosis of mechanical problem. Regulation and certification of AI will also require finding means to catch up on such technological breakthroughs. The FAA is expected to be cautious in approving bold new capabilities. Aviation is actually ahead of cars in many ways as they are regulating and training on automation for decades.

After the Air France crash in the Atlantic Ocean that killed all 228 people onboard, investigators concluded that as cockpits become ever more computerised, pilots need to be re-skilled to cater for different kind of failures. This has also led AI experts to develop special kind of ‘trained’ autopilots that use a ‘machine learning’ system.

Aircraft Simulators and Diagnostics

Airplane simulators are already using AI in order to process the data taken from simulated flights. Other than simulated flying, there is also simulated aircraft warfare. Computers are able to come up with the best success scenarios in these situations as these machines can also create strategies based on the placement, size, speed and strength of the forces and counter forces. Pilots may be given assistance in the air during combat by computers. The AI programmes can sort the information and provide the pilot with the best possible manoeuvres, not to mention getting rid of certain manoeuvres that would be impossible for a human being to perform.

The Interactive Fault Diagnosis and Isolation System (IFDIS) uses a rule-based expert system made using collecting information from documents and expert advice from technicians. The performance system will also replace specialised workers. The system allows the regular workers to communicate with the system and avoid mistakes, miscalculations or having to speak to one of the specialised workers. The Air Traffic Controllers train by giving directions to the artificial pilots and similarly for pilots to respond to the ATC. The programmes incorporate the speech software created by using neural networks.

Design and Safety

The AI-supported Design of Aircraft (AIDA) is used to help designers in the process of creating conceptual designs of aircraft. One thing technology has done in the past and will do even more in the future, is improve flight safety. Bad planning is responsible for 60 per cent of the accidents in leisure aviation. To help plan flights more effectively, more and more sophisticated autopilot programmes have emerged. NASA’s Dryden Flight Centre and many other companies have created software that could enable a damaged aircraft to continue flight until a safe landing zone can be reached. The software compensates for all the damaged components by relying on the undamaged components. The neural network used in the software proved to be effective and marked a triumph for AI.

The Integrated Vehicle Health Management system, also used by NASA, onboard an aircraft, must process and interpret data taken from the various sensors on the aircraft. The system needs to be able to determine the structural integrity of the aircraft as well as to implement protocols in case of any damage on the vehicle. Over the last few years, a number of aircraft accidents have been caused by pilot fatigue. Automation failure combined with pilots’ lack of self-flying skills and tiredness has caused fatal crashes. Boeing and Airbus have forecast the need for as many as 617,000 new pilots by 2035. Will some of them be replaced due to AI?

AI machine-learning algorithm would make it possible for machines to learn from how human pilots cope with unfamiliar situations such as serious emergencies, sudden turbulence, engine failures or loss of critical flight data, reducing pilot workload and fatigue. The machines are proving capable both for airliners and fighter jets. Predictive AI will probably involve voice alerts and speech recognition, too. The Garmin’s Electronic Stability and Protection system is a passive safety system that continuously monitors the airplane’s attitude and uses the autopilot to put the airplane back to safety if it is either too low or at a wrong angle. Garmin’s Intelligence system uses voice commands to complete hundreds of common tasks in the cockpit. This is certified and available to install today.

Billions of dollars are being spent to develop drone technology that avoids terrain, obstacles, traffic and weather or self-diagnoses a mechanical problem and returns to base using AI. A lot of these scenarios and technology will be tested on drones before coming to passenger aircraft. An AI system could alert you about your fuel level, systems status and weather or even remind you to do a checklist before you have to ask or think about it. Beyond the alerts, the autopilot could also advise on new flight plans dynamically generated with weather data, fuel consumption rate and other parameters it needs to take into account. As prevention is better than cure, day to day maintenance of the aircraft itself is crucial. AI can automatically and periodically run maintenance scenarios and check status of systems. By using huge amounts of flight data processed thanks to machine learning, it can detect failure or inconsistent patterns and trigger repair at the first signs of weakness before something is broken.

Filling the AI Gaps

We need to have a system to know when AI is falling short. AI is well equipped for about 80 – 95 per cent of the task and the remaining 5 – 20 per cent ends up being the real headache. What AI leaders at Google, Amazon and the like have figured out is that when it comes to mission critical applications, you need a combination of AI and human judgment or ‘IQ’ in order to close the gap. Google Maps were built by using Google’s AI to find the streets and intersections in the imagery, but then Google’s ‘Team Ground Truth’ (Human IQ) had to fill in the gap on tricky one-way streets and construction zones. Amazon embraces the AI plus IQ formula even more. When it comes to AI systems for complex environment, think about utilising systems where human judgment gets you the last mile, so that operators on the frontlines do not have to deal with the false alarms of AI-only systems.

Counter View

On the other hand, some operators are questioning the ‘if-then-else’ approach. That is what led to an Airbus autopilot throwing up its hands and handing over the aircraft to those Air France pilots over the Atlantic. The autopilot disengaging in the AF 447 crash was in the absence of airspeed data – that was exactly what the autopilot was specifically designed to do by its human designers. Dumb programming could be a disaster. But AI is meant to cater to design and input failures by simultaneously taking alternative inputs from other data sources and sensors. What are the job prospects when all of those co-pilots get pushed out of the cockpits after having been replaced by AI?

Similarly, all of the railroad crew members, truck drivers, maritime crew, surgeons, diagnosticians, to say nothing of vast segments of general management, the medical professions, lawyers, accountants and even government employees who will have to ‘retrain’ and will perhaps require very high IQ even to be considered for any sort of job? Many such as Stephen Hawking regard AI as possibly the greatest threat to humanity. Automation in the cockpit has already reduced the pilot to a Flight Systems Manager. Some of us old retired lucky pilots are glad that we were in a time when flying was something humans could do.

The Future

AI is about ‘man-and-machine’; not ‘man-vs-machine’. The field of robotics is closely related to AI. Intelligence is required for robots to be able to handle such tasks as object manipulation and navigation. Problems to solve include own localisation, mapping what is around and motion or path planning. Effective computing is the study and development of systems and devices that can recognise, interpret, process and simulate human affects. A motivation for the research is the ability to simulate empathy. The machine should interpret the emotional state of humans and adapt its behaviour to them, giving an appropriate response for those emotions. Many researchers think that it will eventually be incorporated into a machine with general AI combining all the skills and exceeding human abilities at most or all of them. A few believe that anthropomorphic features such as artificial consciousness or an artificial brain may be required.

Aviation is one of the most heavily regulated industries in the world due to safety reasons. Strict regulations have helped the aviation industry provide the safest way of transport per mile travelled. Aviation incidents are few and far between and are getting rarer every year. Some degrees of automation have indeed helped get aviation to where it currently is. But human control and intervention have always been at the heart of it, from pilots to Air Traffic Controllers. This is about to change. Air freight seems to be the obvious entry point for pilotless planes, just as driverless trucks are about to disrupt the ground transportation industry. The foundations for AI being in control of ground movements are already in place. When a plane lands, a human alone does not decide how and which gate it should go to. AI is already in the airport  and helping reduced human resources and optimum use of infrastructure and making flying safer, faster and more efficient.

The ever increasing processing power of silicon electronics and the ability to harvest massive amounts of data for computers to process, analyse and categorise is the single reason for AI to be possible. Amazon, Google, Facebook, IBM and Microsoft have established a non-profit partnership to formulate best practices on AI technologies, advance the public’s understanding and to serve as a platform.

Click to Buy: IDR Apr-Jun 2017

Airbus, a big European maker of jets, says it is investigating neural networks, too. But such systems are unlikely to be flying passenger jets just yet. Neural networks learn by modifying the strength of the connections between their simulated neurons. That means that ANNs cannot yet be validated by aviation authorities. DARPA has presented a project this year called Aircrew Labour In-cockpit Automation System (ALIAS), which is a physical autopilot that manipulates all the controls of an airplane. It is primarily designed for the military. Militaries can just use this drop-in option on any airplane.

The AI autopilots will probably find its first uses in drones. The system’s ability to keep control in challenging weather might see it used in scientific investigations of things such as hurricanes and tornadoes. The US DoD is looking into ‘autonomous wingmen’, the Russians and Chinese are looking into fully automated battlefields. A fully autonomous civilian aircraft has already been suggested. Give it a little over decade for it to be a reality. Historically, regulation moves slower than technology because ensuring safety requires lots of tests and certifications. However, aviation is ahead of the automotive sector in many areas, thanks to the dynamism of this industry. Even if this feels uncomfortable, the moment we will allow machines to be entirely autonomous in flight, we will have the safer, faster and more efficient flights we dream of.