Friday, 8 September 2017
Artificial Intelligence and the Internet of Things – a Legal Scenario
By Adam D.A. Manning LLB, LLM
Artificial Intelligence (AI) already plays a surprisingly large role in our society and economy, from the way mobile phones route calls to financial software predicting or playing the stock market, and will become even more prevalent in the future. A term used to refer to computer systems that mimic aspects of human intelligence such as taking decision or even learning, AI is here to stay.
Another related, revolutionary advance is the Internet of Things (IoT). Put simply, IoT is interconnecting all sorts of devices, including everyday objects, over the Internet so that they can exchange data. This can include medical equipment such as pacemakers, beds in hospitals able to detect if someone is lying in them and even smart watches worn to track exercise. It can include devices monitoring infrastructure such as roads, bridges and railway tracks. It is of potentially enormous application and there are estimates of 20 to 30 billion IoT enabled devices by 2020. One consequence is a huge increase in the amount of data available for access.
In considering the practical implications of these developments, one illustrative scenario could be that unfortunately regular occurrence, the road traffic accident. Cars are a perfect example of “devices” that could be on the IoT and if so, this might entail a car transmitting constant data about its position, speed, orientation and even whether it has a fault or is damaged and in need of repair.
Imagine also that the roads themselves have sensors embedded in them that can detect similar information about the vehicles travelling on them, such as their position and heading. This might be linked to a CCTV system through fixed position cameras or even drones. Data from the transport infrastructure system might also include information about the conditions of the road, such as whether it is dry or not, and the general prevailing conditions (time of day, bright sunshine, visibility, fog, cloud, wind and so forth).
For the time being, ignore the issue of driverless cars to keep things simpler. Now, on this road of the future, imagine that there has been a nasty accident involving several motor vehicles, in which sadly some of the occupants have been injured.
The data from vehicles and the road network could be transferred to the police force’s computer to determine whether the emergency services were needed to attend at the scene and, if so, in what number. As this is a serious incident, a substantial contingent might be appropriate. As a result, the emergency services might be alerted sooner than any of the humans at the scene or observing might have been able to contact them.
Once the emergency personnel arrive, their data would be added to the set of information already available from the vehicle and road sensors, including an assessment of the injuries suffered by victims at the scene and more details about the damage to the vehicles, including photo and video imagery. Witness evidence from people at the scene (or later taken at the police station or a victim’s home) would add further data.
The AI system used by the police would then analyse all of this data and consider issues concerning criminal liability. This might involve, in part, the use of an expert system, which takes data and applies decision making processes to it. Expert systems are an established field of AI with a long history and do not represent a radical new advance, apart from perhaps the huge amount of data that a true IoT would provide.
Whether anyone involved was driving over the speed limit might be a relatively straightforward issue for the system to analyse. The software might also analyse whether any of the drivers were driving in such a way that might be considered dangerous or caused the incident. This analysis might evolve over time as the software begins to recognise features of a situation that to a police officer suggest dangerous driving; true machine learning might be required.
The data set could then be further analysed by the police force’s AI to determine if any of the drivers had breached the criminal law and warranted prosecution as a result. This would involve analysing the witness evidence for weaknesses such as contradictions with the objective data obtained from the vehicles and road sensors. Part of this system would entail natural language comprehension on the part of the AI; that is the system would have to “understand” the text of the witness evidence.
If the police force’s AI determined that the case might warrant prosecution, a recommendation could be made for the case to be forwarded to the Crown Prosecution Service (CPS) for a CPS lawyer to review, possibly with the assistance of the CPS’ AI system. The CPS’s AI could analyse the data relevant to this incident and compare it to its library of similar incidents, reviewing their outcomes and predicting what could happen concerning this particular incident. The conclusion of this analysis would be a recommendation regarding prosecution and the possible offence involved.
With regard to the civil law aspects of the situation, the data would be transferred to the AI system for the insurers for those involved in the incident. The insurer’s AI would analyse all of this data to determine if any of the parties involved had a personal injury or motor vehicle damage claim against any of the others.
This would, as the case progressed, include data from the hospitals, GPs and other medical practitioners involved concerning the nature and extent of the injuries suffered. At an appropriate point in time, when prognoses were as clear as they could be, the insurers’ AI systems could recommend offers to make to settle the claims and even negotiate settlements based on precedents for the amount of compensation involved.
Data from garages, employers, physiotherapists and others could be included in the claim to ensure full compensation was recovered, including loss of earnings. Insurance companies already use software such as Colossus to value claims; this just takes the process to its ultimate conclusion.
Of course, in reality this system might have regular human supervision to ensure it wasn’t making aberrant decisions, but in principle this could operate quite independently of human intervention. Lawyers, that is human lawyers, might only be involved in very difficult cases where the data, even in this amount, was unclear as to whether there was any issue of liability.
The widespread use of driverless vehicles in the future implies that a lot of the data about the vehicle’s position, speed and heading will be available in any event as it will be needed by the driverless navigation system. In the event of an incident, the issue of who, if anyone, is prosecuted or liable in a civil sense becomes rather more abstract and it may be that the appropriate tort is one of product liability rather than negligence.
This thought experiment illustrates an example of how an even more data driven digital society might function. It relies on an Internet of Things generating the huge amount of data necessary, the internet as a medium for the exchange of this data and AI systems within our institutions and companies processing it and taking sometimes very serious decisions as a result.