Today we can design systems to respond to unplanned yet known events, but humans struggle to cope with the challenges provided by completely unexpected situations
OEMs need to find the right problems to solve and the right types of AI to solve them
Understanding how data and AI can be used will help you to differentiate your business
Set expectations: be prepared to fail - and to succeed - in unexpected ways
Humanity has been fascinated by artificial intelligence since it has been able to conceive of the idea. An Ancient Greek myth imagined “Talos”, a giant bronze automaton, built to protect the Cretan town of Europa from pirates. Wolfgang von Kempelen built “The Turk” in the 1700s, which he claimed was a mechanical man able to play chess. It could indeed play chess – and win – but there was a real man hiding inside.
Today we have chat bots we can talk to, robots that are taught by engineers to build cars, and cars that can drive themselves (to some degree). But while you can ask Alexa, Siri, Cortana, Google Assistant, and other digital agents questions - or tell them to control your lights or heating - try to hold a conversation with them and most autonomous chat bots are lost for words.
Still, while we are a long way from building anything that would be mistaken for human intelligence, we do have versatile automated machines, not least in the automotive sector. Navigation systems can learn your preferred route. Cars can automatically respond to the environment through mechanisms such as adaptive cruise control, where automobiles travel at the required speed but slow down if the vehicle in front gets too close; some will even overtake to maintain the desired speed.
But can we build machines that will deal with unexpected circumstances? You can design a system to respond to a number of unplanned yet known events that might arise, but to cope with the completely unexpected is a challenge many humans struggle with. Fully autonomous cars will come in time, but much work has to be done before they can safely be put into situations that cannot be controlled.
Advances in computer power and the development of machine learning offer huge potential for automation, prediction, and generation of insights from patterns in the data that humans fail to see.
We have come on in leaps and bounds, but we must be able to trust autonomous vehicles with our lives. So, while to some it is magical and mystical that machines can do such things, we need to understand the technology we are dealing with: there is no room for admiring the magic – this is engineering, pure and simple.
Advances range from a leap forward in computer power to the development of machine learning, in which big data can be used to train neural nets and deeplearning, cognitive algorithms. These offer huge potential for automation, prediction, and generation of insights from patterns in the data that humans fail to see.
Applications for the automotive sector are exciting. Market research specialist Gartner now expects there to be 250 million connected cars on the road by 2020. Those connected cars will be generating a high volume of data that will be used to generate insights into customer behavior; insights that can inform new product design, as well as improvements and add-ons that can be provided as part of Software Over The Air (SOTA) updates. Forrester Research predicts “Artificial intelligence will drive the insights revolution” (note 2) and that “those that are truly insights-driven businesses will steal $1.2 trillion per annum from their less-informed peers by 2020”.
As well as generating insights for what products to create, at least one attempt has been made to “grow the actual product” by creating an autonomous car based on a neural net that learned how to drive by “observing” human driving behavior (note 3).
Jobs at all levels in society presently undertaken by humans are at risk of being reassigned to robots or AI, and the legislation once in place to ‘protect the rights of human workers’ may be no longer fit for purpose.
Gerlind Wisskirchen, vice-chair for multinationals at the International Bar Association Global Employment Institute
In fact, AI in automotive is just one part of a technology revolution with the potential to transform the world in which we live, changing the nature of our societies. As Gerlind Wisskirchen, vice-chair for multinationals at the International Bar Association Global Employment Institute has explained, “Jobs at all levels in society presently undertaken by humans are at risk of being reassigned to robots or AI, and the legislation once in place to ‘protect the rights of human workers’ may be no longer fit for purpose”.
It may even be that legislation is required to ensure human beings get their “fair share of jobs”, alongside the robots. Automation and AI can reduce costs and improve productivity, boosting the competitiveness of industries and economies, but there will be downsides to manage, including the prospect of widespread unemployment.
Moreover, amid the excitement, pragmatism is required.
As one might expect, there is fear relating to AI, from both safety and work perspectives. The reality of AI capability today, however, is nowhere near that described in sci-fi novels. The potential for human-machine partnerships far outweighs the likelihood that people’s jobs will be taken by machines. In the automotive sector, the challenge is to find AI applications that work, both technically and commercially. It’s easy to get carried away by the potential for AI in connected cars, but imagining the future is the stuff of science fiction – in realizing this vision, automotive manufacturers face engineering challenges.
The automotive sector is already making extensive use of AI technologies, with 7 million systems installed in vehicles by 2015 (according to IHS Technology). As communication networks improve, this number will increase rapidly, as extra bandwidth becomes available from 4G and 5G technologies as well as advances in Software Over The Air (SOTA).
Many of these systems are dedicated to delivering infotainment and driver interaction services, from smarter traffic and mapping services to voice and gesture recognition, built on natural language processing capabilities and pre-trained neural-nets. GM, for example, is working with IBM to incorporate its Watson AI technology, made famous in 2011 by beating two former winners of the US TV show Jeopardy in a head-to-head version of the quiz (and winning $1m as a reward). IBM Watson will be built into GM’s in-car “cognitive mobility platforms”; these will essentially be digital assistants capable of following drivers’ instructions and anticipating their needs.
As one might expect, there is fear relating to AI, from both safety and work perspectives.
A second application of AI is the creation of new insights from big data: automotive manufactures source huge volumes of data from vehicles on the road that it is difficult for humans to comprehend. The connected car can provide a constant stream of information about how and where it is being driven, and how it is performing. Machine learning provides a powerful tool in the ongoing struggle to see the patterns within the data and to make sense of it.
Manufacturers such as BMW have already begun using this data in new ways. One early application is predictive maintenance – the ability to anticipate faults at very early stages so they can be corrected before breakdowns occur, reducing cost and inconvenience.
Increasingly, manufacturers are also incorporating data and analytics tools into their production technologies, using the insights gleaned from in-car data to power next-generation design and engineering. For example, the Los Angeles-based start-up Hack Rod aims to create the first ever car engineered with AI that has been designed in a virtual environment.
Then there is the essential role of AI in powering the advanced driverassistance systems (ADAS) that will eventually turn fully autonomous vehicles into a mainstream reality. ADAS incorporates tools such as camera-based machine vision systems and radarbased detection units, but is ultimately powered by intelligent software capable of making instantaneous decisions on the basis of the complex information set with which it is presented.
Increasingly, manufacturers are also incorporating data and analytics tools into their production technologies, using the insights gleaned from in-car data to power next-generation design and engineering.
Opinions are split on when fully autonomous vehicles will be routinely using public roads, but manufacturers including Tesla, BMW, Mercedes, Nissan, Ford and Volvo have all developed selfdriving cars, while entrants from other sectors - including Google and Uber - have proved that the technology works in a series of public trials. Mobileye, acquired by Intel for $15bn in March 2017, claims its partnership with BMW will begin mass production of fully autonomous vehicles in 2021.
The race is on, in other words. The engineering challenges now are to make the smartest, most convenient, adaptive, intelligent self-driving electric vehicle with fully personalized infotainment, and to get as much meaningful data from such vehicles to inform a cycle of continuous improvement.
For all the potential of AI in automotive, substantial challenges remain and disappointments are inevitable. Indeed, the latest Gartner Hype Cycle for Emerging Technologies puts autonomous vehicles right at the top of the curve, squarely in the “Peak of Inflated Expectations” zone where early success stores are accompanied by many failures.
The reality is that automotive autonomy is at a relatively early stage, with human drivers still required to monitor journeys and take control of vehicles in certain circumstances. Uber, for example, admits its cars struggle to navigate bridges, where an absence of buildings makes it hard for the vehicles to find reference points. Also, Ford and Tesla point to extreme weather problems such as snow that settles on lane markings, covering them from view.
This creates further problems. The Toyota Research Institute has pointed to the challenge of giving drivers enough warning to take control of their vehicles when they need to intervene.
One issue is that the business of replicating human judgement is a developing science. AI developers have made huge progress with ideas (such as neural nets) and they are experimenting increasingly with deep-learning techniques and concepts such as natural selection, where the right solution is arrived at through complex trial-and error processes.
There are many types of AI and each is suited to different kinds of problems with different forms of data and decision-making requirements.
Fully autonomous driving requires a complex skillset. The system must have powerful perception, to understand exactly what is happening in the realtime environment, but also the ability to model intention – to anticipate how that environment is likely to change when, for example, other vehicles change position.
Moreover, the judgements made by humans are often very difficult to explain; building algorithms to replicate the processes by which certain judgements are arrived at is therefore very difficult. It may be that neural nets, say, will also deliver answers that do not appear logical. That will require a high degree of trust among users.
A related problem is that connected cars generate increasingly vast data volumes, for which automotive manufacturers will need new capabilities to collect, store, organize and analyze. Many may find their existing data management abilities are not fit for purpose. In these early days of autonomous driving, collecting and analyzing these large volumes of data will be essential for improving AI training, as well as improving the competency of AI engineers.
It’s imperative for manufacturers to think in terms of the problems to be solved– and which tools are best placed to help – rather than to start with the technology.
Then there is the question of safety. Autonomous vehicle manufactures will be tested on every journey as to their claims to be producing safe vehicles, with painful reputational and adoption consequences following setbacks. Delivering perfect safety records will be impossible, particularly given the presence of non-autonomous vehicles on the roads. Agreement is also necessary on difficult ethical questions: where some form of crash is inevitable, which moral choice should be made – for example, should the car swerve to avoid another vehicle if doing so will mean hitting a pedestrian?
There are many types of AI and each is suited to different kinds of problems with different forms of data and decision-making requirements. We are still learning how to apply these technologies as each type of AI has its own limitations. Pattern recognition is only as good as the example patterns with which computers have been trained. Inference engines work only as well as the rules and the variables set for them. Genetically evolved designs depend on the chosen criteria for natural selection and the granularity of the learning algorithm.
There are many ways to combine different kinds of AI to compensate for their weaknesses and to make best use of their strengths. The use of different layers of processing is also important. More manually crafted layers of “intelligence” have been combined: for example, image-recognition algorithms often demonstrate improved performance when images are prefiltered to accentuate key features, reduce noise, or remove irrelevant parts of the data (sometimes a black-and white image is sufficient, as color merely distracts).
Building an intelligent machine that can transport people along the obstacle courses of our highways requires engineers who understand the problems involved and how the different forms of AI can be combined effectively, to interpret the wealth of problem-data being spewed out every millisecond by the myriad of sensors on the car to make driving decisions.
How will manufacturers move forward with AI? It’s imperative to think in terms of the problems to be solved – and which tools are best placed to help – rather than to start with the technology. In other words, think of AI and the connected car as just another engineering challenge.
In practice, that means beginning with an AI opportunity assessment – an audit of the business’s most pressing challenges and opportunities, combined with a scoping exercise to identify where AI technologies will deliver the greatest leap forward in understanding and value. In some cases, the answer may derive from customer-facing use cases, but it will also be important not to neglect the potential for AI to deliver more internal benefits, such as process efficiency, product design optimization, and cost control.
Automotive manufacturers also need to consider whether they possess the skills and resources to exploit these opportunities. In some areas, services have become commoditized – voice recognition is a good example – and the simplest solution may be to buy them in from third-party suppliers (or buy the OEMs outright). In other cases, closing the expertise gap may be more difficult; this has been one factor influencing M&A activity in the automotive sector, with manufacturers buying niche providers to acquire specialist knowledge. Ford’s recent $1bn investment in the AI business Argo is one example.
Successfully exploiting AI in automotive will depend on manufacturers’ ability to incorporate key concepts from multiple disciplines.
Successfully exploiting AI in automotive will depend on manufacturers’ ability to incorporate key concepts from multiple disciplines – demand pressure in automotive, but also emerging trends in the application of AI, including chat bots, intelligent agents, voice recognition and adaptive behavior, as well as big data and analytics capabilities. It’s important to learn from industry lessons. For example, Microsoft’s Tay Twitter chatbot aimed to showcase advances in AI and natural language processing capabilities, but it very quickly learned from other Twitter users to disparage women and ethnic minorities. Microsoft was forced to decommission Tay, making it one of the year’s biggest machine learning and AI busts. (note 4) It may also be necessary to develop new ways of working and even new business models. Concepts such as “fail fast” and “agile” can help manufacturers deploy new technologies rapidly, testing and iterating them in the marketplace, and then discarding the less successful ideas. The automotive sector is used to working on extended production and development cycles, but for AI and related technologies, these will need to accelerate.
Nor should automotive companies ignore the potential for new collaborations, including with academic partners. Such links are now strengthening, with new ideas from research scientists finding commercial applications. Toyota, for example, already has a research partnership with Stanford University and Massachusetts Institute of Technology (MIT) that is working on systems to assist vehicles in interpreting real-time challenges in urban environments.
While it’s easy to slip into the realms of science fiction, AI actually represents just another business challenge and opportunity. Other industries recognize exactly this point. For example, in financial services, the hedge fund Bridgewater Associates is working on an AI project that will automate decision making and strip out human emotion from its management processes.
The automotive supply chain is already being disrupted, as technology firms get ever larger slices of the automotive pie. Frost & Sullivan points out: “Technology companies are expecting to be a new tier 1 for OEMs… 13 OEMs will be investing over $7.0 billion in the development of various AI use cases. Hyundai, Toyota, and GM will account for 53.4% of the total investment share.” (note 5)
For the automotive sector, the challenge is to understand what the connected car could achieve for the customer base – and how to secure those benefits.
For the automotive sector, the challenge is to understand what the connected car could achieve for the customer base – and how to secure those benefits. The key to meeting that challenge will lie in engineering, innovation, and product and service development. There is urgency if OEMs don’t want to be left behind, but there is also a place for traditional engineering discipline, and investment portfolio and risk management.
by Professor Mark Skilton, Professor of Practice, Information Systems Management & Innovation, Warwick Univeristy
Mark Burnett: Mark, can you give us your perspective on the impact AI is having and will be having in the future?
Mark Skilton states: “We are in a race towards a new artificial intelligent automotive ecosystem that will impact how transport will work across all industries and countries. AI is the new competitive advantage that is seeing existing brands and new ones rapidly redefine their business model and vision from a product to a service enabler as the mobile platform of the future. The automobile will be the proxy for everything we do on the move, defining how vehicles are designed and assembled in intelligence factories and connected supply chains, to redefining the role of travel in the smart city, connected home, automated retail and personal lifestyles. Vehicles will change human employment through self-driving capabilities within the next decade. It will change car ownership combined with a massive change to clean energy and new possibilities of onboard connected living and work experience. All these automotive innovations will reinvent the car and automotive companies place in society as we move into the fourth industrial revolution.”
Mark Skilton has 30+ years’ experience as a professional consultant with a track record in top 1000 Companies in over 20 countries and across multiple public, private and start-up sectors. He has direct Industrial experience of commercial practice leadership, boardroom and investor strategy to program team and transformation management at scale. Mark is a recognized International thought leader in digital, company strategy, telecoms, digital markets and M&A strategies, CxO practices and is the author of many books and international papers, including Building the Digital Enterprise Building the Digital Enterprise and Building Digital Ecosystem Architectures. His work and views have been published in the Financial Times, New York Times, Wall Street Journal, Washington Post, Bloomberg, Associated Press, Mail, New Scientist, Nature, Scientific American and broadcast via many television and radio channels around the world, including BBC, Sky, ITV, Al Jazeera.
BearingPoint can help in a variety of different ways: we are working with various kinds of AI in different business areas and as part of Digital Ecosystem Management, we are generating business insight from big data, and we are applying AI and Robotic Process Automation (RPA) to various real-time system, process and decision-support problems.