One of the fundamental
concerns in relation to AI is the absence of a body of law which applies
specifically to AI. This is hardly a novel problem – even before the industrial
revolution, the law has had to develop and adapt to deal with new technology.
However, never before have we been faced with a situation where a particular
form of technology may (one day) be truly autonomous.
While AI has not yet
developed its own body of law, it does not sit in a complete vacuum and many existing
areas of law have an impact on AI, most notably intellectual property which is
relevant for the purposes of identifying and regulating the ownership and
output of AI. There is no overarching legal framework which applies
specifically to AI and, in particular, which regulates who has responsibility
and liability for artificial intelligence.
The autonomy and
inscrutability of more advanced AI raises liability issues. When mistakes are
made, who has ultimate liability? Where AI is employed to mimic the process
carried out by a human, fault is likely to be easier to identify. The party who
has provided the wrong data, the wrong pre-determined result or the wrong
process is likely to be liable but when AI writes its own code and uses its own
algorithms, who is at fault then? This is a question which will affect every
sector and area of business.
The possibility of holding
AI itself liable seems to be out of the question. AI does not have legal
personality and from a natural justice perspective it would seem perverse to provide
that an entity which has no assets and no capacity for punishment is
responsible for acts or omissions that cause damage to third parties.
That then brings us back to
the issue of liability for the owners or controllers of the AI. Identifying the
responsible party may not be an issue where the result is effectively pre-determined
by an entity with legal personality, but where machine learning is involved
this is often not the case.
Given the current state of
the law, it is difficult to envisage how liability may be attributed in
litigation where there is an unintended and unexpected consequence as a result
of machine learning. How is causation demonstrated in a situation where no one
could have anticipated the results? Will it simply be the case that no
liability arises or will the law have to adapt to deal with these situations?
If the former, it is imperative that the class of those potentially affected by
AI is made aware of the fact that AI is being used and that they effectively
agree to waive any claims insofar as its use is concerned. It seems highly unlikely
that consumers or businesses would agree to that.
An alternative is simply to
accept that if AI is used, the possible consequences will be potentially very
far-reaching and while not foreseeable to those employing the AI, it is
accepted as a risk of use of the technology. In this scenario, insurance is
likely to play a key role in meeting the costs of claims arising out of the use
of artificial technology. This is certainly the path envisaged in the
government proposals for the regulation of automated vehicles, with the insurer
assuming liability for any accident caused by the driverless vehicle. Where no
insurance is in place, the owner of the vehicle will be liable. Interestingly,
the Vehicle Technology and Aviation Bill also allows the insurer to exclude
liability for failures to update software and operating systems. This adds
another layer of complexity to the problem, where failure to run updates or the
installation of defective updates may amount to an act which breaks the chain
of causation.
There has been some
discussion as to whether the application of vicarious liability is appropriate
in the context of AI. While that proposition does have a certain attraction
(the human creator will have ultimate responsibility for its AI creation), the
concept of vicarious liability as we currently understand it is ill-suited. It may
be true that AI has a certain degree of autonomy, but before true vicarious
liability can apply, AI would still need to have separate legal personality.
Agency would also fall short on the same principles. Perhaps a model of
liability based on the Consumer Protection Act 1987 would be appropriate,
whereby the producer or the supplier may be found liable depending on the
circumstances. Were such a model adopted, the end-user would have some comfort
that a remedy may be pursued against the supplier if it proves impossible to
identify the ‘producer’ or creator of the technology. This model would address
some of the potential difficulties the claimant would face where several
different parties are involved in the creation of the AI technology. From a
public policy perspective, this is an attractive solution. However, supplier
liability is generally viewed as a remedy of last resort. Any legislation
regulating liability in this area must be sufficiently robust to avoid a
situation where parties using AI to deliver a service simply offer up all other
parties involved without providing any greater clarity for the consumer.
One outcome which does seem
certain is the increased use of contractual frameworks to regulate liability
where AI is used. A growth in insurance cover and claims also seems likely and
it is inevitable that parties using and developing AI will try to make
provision for liability and responsibility for damages claims and potential
fines levied as a result of the use of AI.
AI, Liability and Lawyers
Lawyers are not well known
for embracing modern technology. We are seen as a staid profession that has not
quite caught up with modern ways of working and our obsession with detail is
often perceived as an obstacle to business rather than a facilitator. How many
lawyers have inwardly groaned when a client has insisted that a contract be one
side of A4, when most of the contracts we deal with fail to encapsulate
anything more than the defined terms on the first page? Despite this,
artificial intelligence is now making inroads into the legal profession,
particularly in litigation in disputes ranging from road traffic accidents to
PPI claims. But are lawyers ready for these changes and, more importantly, is
the law?
Liability for artificial
intelligence is of particular relevance to lawyers not only because lawyers
will no doubt be involved in more and more disputes regarding the liability of
AI for loss and damage but also because of the increasing use of technology in
the legal sector.
In dealing with negligence
claims, one of the first tasks of any lawyer is to identify the responsible
party, whether that party is a natural or corporate body. The difficulty with
AI is that it has no legal personality and cannot be held accountable for legal
wrongs.
More basic forms of AI
should be easier to regulate. Where AI is designed to reach a pre-determined
outcome, it is easier to conceive that identifying the “responsible party” will
simply be a case of looking to the users, owners or programmers depending on
the particular facts of the case. But as we move to a form of AI that is truly
autonomous, does that need to change?
While use of AI can
certainly assist lawyers in processing claims, the real use of AI technology at
this stage tends to be limited to just that – a process. AI can be used to
predict certain outcomes based on a set of facts as presented but many current
technologies would struggle with anything more complex, for example identifying
the court which has jurisdiction based on anything more than the address of the
defender, identifying whether a claim may be time-barred and applying any
recent changes in the law, particularly case law which may affect the prospects
of success. One of the potential problems with machine learning is that the
program learns from previous problems which it has processed. The risks then of
providing “negligent” advice during that learning process seem relatively high.
Perhaps no higher than engaging a human lawyer, some would argue, but the scope
for using machine learning in situations which turn entirely on their own facts
and circumstances seems some way off. For now the old maxim seems to apply – garbage
in, garbage out.
If court proceedings are
unsuccessful, because a solicitor fails to take into account recent precedent
which has subtly, but importantly, changed the law, that solicitor will, more
than likely, be held to account for that omission. However, when the merits of
the case are assessed by AI how does a recent change in case law factor in any
liability for a decision made by AI? A different type of learning will need to
be undertaken by the system before the correct legal tests are applied.
Even if court proceedings
are successful, consider the scenario where the client has incurred costs of,
say, £5,000 in successfully pursuing a claim for £500. The rate of recovery of
judicial expenses means that the client is still substantially out of pocket.
In order to replicate the task of the solicitor, AI must operate in such a way
that it considers not only the merits of the claim itself but the wider
economic consequences. It is certainly conceivable that AI will be able to
factor in variable such as expenses in litigation but monitoring the
cost/benefit analysis involved in a litigation is an ongoing process. What
looks like a worthwhile investment at the outset of a case may look like a very
different prospect two years down the line when substantial costs have already
been incurred.
It seems very unlikely that
the legal profession’s regulators would absolve a solicitor from blame in situations
where they have failed to give the standard warnings about the perils of
litigation but where AI is used in processing a claim, are solicitors now also
under a duty to advise of the risks of using AI as a basis for informing or
making decisions?
Conclusion
Artificial Intelligence tends
to polarise opinions with many seeing it as a panacea and others as the source
of deep-rooted concerns or a form of technology that will never really catch on
to the extent some claim. The truth more likely lies somewhere in the middle. While
AI is does not yet have the freedom of thought that one might associate with the
novels of Philip K Dick, the extent to which AI, in some form, already plays a
part in our everyday lives is striking.
The Privacy and Electronic
Communications Regulations 2003 are due to be replaced over the coming year.
Technology has developed apace since 2003 and an update is well overdue to
reflect modern digital interaction. This aptly demonstrates the long lead-in
time that is usually required for the legal system to address new challenges. On
that basis a review of the law in relation to AI would be welcome sooner rather
than later.
Lynn
Richmond is an Associate at BTO Solicitors, Edinburgh
