Automated Vehicles Act. A precedent for drowning detection systems

OPINION

On 26 January 2022, the Law Commission (2022) published its recommendations for the safe and responsible introduction of self-driving vehicles to UK roads. In this article, we ask the question, who might be responsible were a swimming pool drowning detection system (DDS) failing to detect a swimmer in distress if the Law Commission precedent were extended to DDS? 

 

What the Law Commission said about automated vehicle regulation

The Law Commission proposed that the new Automated Vehicles Act should distinguish "self-driving" and "assisted driving" features when examining the question of liability. 

Assisted driving features include those such as adaptive cruise control and parking sensors. The Law Commission identified that when using such features, the driver is still in active control of the vehicle. It, therefore, makes sense that the driver should be responsible should an incident result in harm or damage. 

The Law Commission saw self-driving features as reducing the vehicle's driver simply to a "user-in-charge". A user-in-charge cannot be prosecuted for offences such as speeding, dangerous driving, or running a red light. A user-in-charge would retain liability for holding insurance and enforcing seatbelt usage. 

So who is responsible when a self-driving vehicle runs a red light and hits a pedestrian? In the Law Commission's view, it is the Authorised Self-Driving Entity (ASDE). The licenced operator (individual or company) who obtained an ASDE licence to sell or operate the vehicle would be the responsible person in such instances. This would include instances where there was no user-in-charge within the vehicle at the time of the accident (such as is foreseen in the case of self-driving taxis on their way to a pick-up). 

The new Automated Vehicles Act would improve the position created by the Automated and Electric Vehicles Act 2018, which removed the requirement to prove that anyone was at fault, with the insurer compensating the victim directly; a position clearly unpopular with insurers. 

 

How does this fit in with drowning detection systems?

If you replace "self-driving" with "detection" and "assisted driving" with "blended lifeguarding", parallels can be drawn. 

It is important to start by saying that no drowning detection system manufacturer, to my knowledge, has claimed their system can be used without constant or remote supervision being in place. Such a system might be said to be "self-detecting". Current uses of DDS technology follow the "blended lifeguarding" approach, which is akin to our "assisted driving" analogy. However, it is foreseeable that the technology may, in the future, afford a pool operator to rely solely on the technology to make a detection. 

In a commercial swimming pool context, we can envisage five scenarios in which DDS might be deployed: 

1. constant poolside supervision is required, and lifeguards have 100% visibility of the pool with or without the technology. 
2. constant poolside supervision is required, and lifeguards do not have 100% visibility of the pool without the technology. 
3. remote supervision is permitted, and emergency responders have 100% visibility of the pool with or without the technology. 
4. remote supervision is permitted, and emergency responders do not have 100% visibility of the pool without the technology.  
5. no supervision is provided, and detection is dependent on the technology. 

Scenarios 1 and 3 present the safest options for operators seeking to install DDS technology. Operators need to be more careful when implementing scenarios 2 and 4, as additional scrutiny in an investigation will fall on the quality of the lifeguard view provided by the CCTV monitors where the DDS relies on object-recognition technology. Additional scrutiny may fall on the operator's information and signage where wearable DDS technology is relied on.

To my knowledge, no commercial pool is operating scenario 5 in the UK. No DDS designer has provided a warranty that their system can meet the requirements without being accompanied by human surveillance. A DDS designer who does make that warranty has moved from DDS usage from a blended lifeguarding arrangement to a self-detection one. I see little distinction to be drawn between a self-driving car that fails to detect a pedestrian and a DDS that fails to detect a pool user in danger. Both can result in potential loss of life or serious bodily or psychiatric injury. 

 

Are we likely to see regulation of drowning detection systems? 

Indirectly, the short answer is yes. The type of system described in scenario five above is already recognized in the law surrounding AI as a "life system". Life systems, where the algorithm's performance can affect whether someone dies or is injured, are already heavily regulated. 

For DDS technology operating in the arrangement described by scenarios 1-4 above, we are likely to see litigation arising out of harm that accrued due to late detections, perhaps where discrimination arises due to algorithmic bias. It may be difficult to defend the use of a system where a DDS better protects female pool users than male pool users. Or white pool users are better protected than black, Asian or ethnic minority pool users. Some form of independent 3rd party certification scheme (against ISO 20380 or otherwise) will need to be established to provide operators purchasing DDS technology with additional assurance as to the robustness of the original dataset, modelling, and deviations caused by any machine learning algorithm.  

Claims for psychiatric injury may also present challenges for DDS technology. Limitations of some object recognition DDS include the algorithm's inability to make a successful detection until a body is at the bottom of the pool. Others cannot detect danger where the body continues to move following cardiac arrest. Some systems cannot detect danger where the pool user is wearing a certain swimming costume or in a particular position. In all such cases, detection might result in additional bodily or psychiatric injury, particularly in swimming pools where additional lifeguard cover would likely have reduced the time taken to perform a rescue (see scenarios 2 and 4).     

We already have a raft of product safety legislation that includes drowning detection systems, and the detail of that legislation is beyond the scope of this article. It suffices to say that designers, retailers, installers, maintenance technicians and operators all have duties in respect of drowning detection systems. These can be summarised as follows: 

• a designer may be responsible when the designed algorithm fails to detect a drowning. 
• a retailer may be responsible for misrepresenting the product's capability or recommending the wrong product for a pool of this type or foreseeable usage. 
• an installer may be responsible for inadequate system testing or calibration before the conclusion of the design stage. 
• a maintenance technician may be responsible for failing to detect the deviation of a machine learning algorithm that fails to detect a drowning. 
• a company (operator) may be responsible for failing to provide appropriate information, instruction and training to employees. 
• an employee may be responsible for silencing an alarm that resulted in greater harm to a pool user in distress. 

In summary, we are unlikely to see additional regulation of DDS options used in the blended lifeguarding format. Systems operating in the scenario 5 format above would likely be included in existing regulations that exist outside the sector. Health and safety and personal injury law will in this decade begin to grapple with how DDS technology blends with more traditional operator controls, and this may increase the likelihood of judges and coroner's making specific remarks about the limitations of when and how DDS technology can be used in swimming pools. The industry should recognise its inexperience in this space, embrace outside expertise and develop guidance to manage the technology's adoption safely and successfully.  

 

Looking to the future

Drowning detection systems provide opportunities to increase the number of commercial pools in the UK estate, the length of opening times, and the number of people who can become physically active through accessing aquatic-based exercise and wellbeing facilities. This is good for councils, operators and communities. These systems should, given sufficient time, facilitate the return of leisure and freeform pools, which remain hampered by high staffing costs due to challenging visibility conditions. Greater diversity in aquatic facility design offers greater value for consumers, and reduced operating costs allow operators to reinvest savings into new or improved services.  

Introducing these systems is not all doom and gloom for lifeguards either. Drowning detection systems can only identify, perhaps in the future detect, incidents. Lifeguards will still be needed to perform a rescue for the foreseeable future, and the pool will still need to be supervised, whether by constant poolside supervision or remotely. When fully-automated DDS technology does arrive, we are likely to see lifeguarding as an add-on to another leisure centre role, similar to that of an emergency responder. The focus of the training will likely retain the rescue skills and place less emphasis on providing surveillance. I wouldn't worry about jobs just yet. A leisure estate is slow to build and upgrade and will likely need substantial investment over several decades. 

Object recognition technology (ORT), on which DDS is based, is not exclusively used in swimming pools. We will likely see all leisure centre security cameras across the facility equipped with incident recognition technology in the coming decades. First aiders will be alerted to a slip in the changing room by ORT. Duty managers may be alerted to users causing damage to centre property by ORT. Fitness instructors may be alerted to gym users' poor technique and allowed to add value using ORT. These possibilities seem remote to many, but the technology to enable institutions across the globe is already trialling all this. ORT is the technology responsible for self-driving cars and self-landing planes.  

In summary, the future of DDS technology is one of opportunity. Managing its safe implementation is key if operators are to turn opportunity into value for themselves, their clients, and communities.  

 

References

Law Commission. (2022). Legal reforms to allow safe introduction of automated vehicles announced. (26th January). Available at: https://www.lawcom.gov.uk/legal-reforms-to-allow-safe-introduction-of-automated-vehicles-announced/ accessed 29th January 2022. 

 

Read more about drowning detection systems and their use in swimming pools

HSG179:2018. Health and safety in swimming pools. 

ISO 20380:2017. Public swimming pools. Computer vision systems for the detection of drowning accidents in swimming pools. Safety requirements and test methods. 

RLSS UK. (2022). Technology in swimming pools. Available at: https://www.rlss.org.uk/tejhgdgjghdchnology-in-swimming-pools accessed 29th January 2022. 

UK Active. (2021). Drowning detection systems. Available at:  https://www.ukactive.com/projects/drowning-detection-systems-dds/ accessed 29th January 2022. 

 

Citation. Jacklin, D. 2022. Automated Vehicles Act. A precedent for drowning detection systems. Water Incident Research Hub, 30 January.