In this blog I am going to discuss about the AI in automobile. The crispy concepts that
What are general cyber threats that are happened in car?
Comparable cybersecurity scenarios that are listed below are:
|Threats||Cyber Attack||Learning Outcomes|
|Unauthorized access to network||Came into existence in 2011, remote attacks on cars using internet have recognized in press due to the work of Charlie Miller and Chris Valasek. The attack is mainly concentrated on CAN bus by attempting to modify the messages such that the behaviour of the vehicle can be taken into control.||
|The access gained remotely can be used for misuse, for example force the geo-fencing of the vehicle in irrespective location.|
|Malicious software, Arbitrary code execution||Researchers compromised libraries used by garages to control diagnostic tools, in order to allow the installation of malicious firmware on cars||
|Manipulation of hardware, Man in the middle, replay of messages||Researchers with a physical access to the vehicle performed a man-in-the-middle by inserting an unauthorized component directly on the CAN bus, then proceeded to drop/alter/replay messages.||Direct CAN access is easier than many manufacturers might think. Lack of protections in the CAN protocol allow to perform a man-in-the-middle, even if timing constraints makes the exploitation non-trivial in practice|
|Man in the middle, Inadequate design and planning or lack of adaption||Researchers recently presented a correlation-based attack on remote keyless entry systems concerning millions of cars. In this case, the researchers claim that the attack could explain theft cases found in the wild. Attackers exploit vulnerable cryptography on these systems.||Vulnerable cryptography|
|Information leakage||Researchers devised an experimental setup to validate their cost analysis estimation of a surveillance attack performed by a mid-range attacker using dedicated hardware. The attack uses ITS communication interfaces.||Surveillance is possible in practice for a mid-range attacker, interfaces lack the pseudonymise measures allowing to mitigate the attack|
Table 1: Common Threats in Automobile
What is AI and how securely should be implanted?
Artificial intelligence (AI) is a technology which deals with the functioning of an activity without interference of humans. The implementation of AI in a complex and safety critical system is very difficult. The threats to this system will be more. When the hacker finds vulnerabilities in the system then he can get into the vehicle network and controls the functionality of vehicle. Cybersecurity attacks will be more if system design is not up to the mark. As mentioned by Ralph Nader the safety is given a high priority when comes to manufacturing, and he also suggests that the car industry should follow the proper norms when designing a car. The car should be properly tested before releasing to outside world. Engineers who are developing the system must take care the complexity of system, and design should be more user friendly to the drivers. AI system should be compatible with the respective country conditions.
Suggestion’s in development and testing phase of software needs to be done?
Designed AI System should have cognitive capabilities like learning and reacting to different situations. Research in cognitive systems should be increased multifold. Cognitive capabilities systems solve the problems like human does. An International Data Corporation (IDC) cognitive artificial intelligence AI contributes machine learning from data and making decisions suggestions. Cognitive AI consists of machines that strive for human capacity digital intelligence. Learning from human created data and reasoning to make decisions with the mind set to emulate human thought process and produce similar results.
- Supervised Cognitive AI is manually updated and customized by a team of human data scientist. This method is time intensive and not scalable.
- Unsupervised Cognitive AI creates unique models that are client specific and no humans are needed for learning phase. Technology is language agnostic that data just comes in like humans learn the data. Learns and reason based on observing the ingesting the best human created data consciously learning more. Scalable and automatically updates with new information with no human intervention.
For an instance we can take few examples like google predictive text keyboard which suggests the words while typing. In which it searches the word from the data base and predicts using machine learning techniques, and add-on feature it provides when the word is not in the database it provides addition of word to data base such that when we type the word next time it automatically updates with user required word. This is how learning and reacting system needs to be injected to automobile systems for getting effective results in AI.
Human analogy of understanding the situation and take the instant decision by considering following:
- Observing the evidence and things that are visible.
- Figure out what we know about the incident and formulate the hypothesis about the incident.
- Evaluate proper hypothesis that is suitable for incident.
- Decision must be taken which is best for that instance and act accordingly.
IBM Watson, which is a Cognitive system, does decision making in massive speed and scale when compared to decisions made by humans. A hypothesis is generated after identifying the parts of speech in a question and evidence to support or refute this hypothesis is gathered by the system based on weighted evidence scores which nothing is but statistical models related to the evidence. During evidence scoring and ranking, the confidence and the response is estimated. From the interactions with humans, it keeps on learning, gaining value and continues to get smarter from its own successes and failures.
Test AI systems in the car as we test for the system for outer space. Number of miles should not be some testing criteria. Number of live scenarios and past data should be inherited to the AI system as reference. Accident investigation should be taken into consideration and increase the effectiveness by taking past drawbacks as an input. Stress testing needs to implement in AI around the boundary limits of the all system. AI should be tested in a multiple combination of negative scenarios occurring simultaneously. OEMs should collaborate and encourage this type of future systems and figure out the possible conditions and deliver it to the world. Collecting of various scenarios and testing each car for all scenarios shall be taken into consideration.
Major challenges facing in the AI while implementing in the automobile are
- Convincing the OEMs for the effective working reference from Ralph Nader article as it took 30 years for car manufacturing companies to realize.
- AI system response to un seen scenarios is unpredictable.
- They are allowing us to overcome many limitations rapidly but tremendous issues also there for another side, like when we are training the system with a wrong data then system ends up with wrong learning and it might cause distraction too.
- The system is used to surrounding conditions, but we cannot predict the system behaviour in new conditions.