Robots Rise to Data Challenge
30 January 2018
Full article available here https://www.nrtimes.co.uk/single-post/2018/01/26/Robots-rise-to-data-challenge
Since a computer defeated world champion Garry Kasparov in 1997, chess has been a continual marker of AI’s progress.
In December, the world marvelled at
the latest machine-driven victory on the chequered board, which underlined just how far the technology has come in the last 20 years.
AlphaZero - Google Deep Mind’s gaming AI – taught itself how to play chess in under four hours with no human intervention.
Within a day it had become a master of the game. Alpha was able to process 80,000 positions per second taking what Deep Mind called a “humanlike” learning approach, if at unfathomable speed.
AI’s ability to rapidly process, learn from and objectively interpret vast amounts of data presents amazing possibilities far beyond the battle for pawns, knights and queens.
Neuro-rehab’s chessboard could be the towering piles of printed data which shape the future of patients in personal injury and negligence cases.
For each carefully planned move in pursuit of the king, read the many potential rehab interventions and outcomes informed by research and past experience.
“The use of AI in this field is inevitable. It is a question of when, not if,” says Edmund Bonikowski, whose expertise straddles neuro-rehab and AI.
The honorary consultant in neuro-rehab at Taunton and Somerset NHS Foundation Trust also produces medico-legal reports on people with brain and spinal injuries via the National Neurological Rehabilitation Chambers.
Currently, he is also investigating ways of applying AI to neuro-rehab data in medico-legal reporting, as part of a PhD programme at Cambridge University’s Information Engineering department.
"Medico-legal reports in neuro-rehab require the expert to assimilate a very wide range of inputs from a substantial body of medical records, some of which will be handwritten, with some structured and some unstructured.
"There are often other expert reports to factor in, as well as witness statements from families, friends and employers. The expert must consider all of this information, reference it against their own experience and say, 'this is what I think'.
“For a serious personal injury or negligence case it is not uncommon to be sent a pile of paperwork that would stand two or three feet tall. On some cases I have received nine or 10 boxes of notes, or thousands of pages on an electronic file, which can be nauseating to scroll through on screen."
While his specific focus is on revolutionising medico-legal reporting, Bonikowski believes AI could also
solve neuro-rehab's wider problematic relationship with data.
"Data in neuro-rehab is very poor and di cult to source. It is extremely hard to identify numbers of people with particular levels of disability resulting from traumatic brain injury, for example.
"Incidence and prevalence levels related to hospital admissions are available, but it is much more diffcult to get at the numbers of people with mild brain injuries who may not go to hospital, or whose injury wasn’t identified by healthcare professionals.
“Superimposed on top of that is the fact that neuro-rehab is generally poorly funded compared to other healthcare specialities. With limited resources in terms of clinical and support staff, the processing of data comes fairly low down the priority list."
The acceleration of AI could help neuro-rehab professionals face up to such issues. And, thanks to easy-to-access networks – which are increasingly cheap, while offering higher speed and capacity – this is theoretically possible now.
“We could easily use document analysis techniques to scan through huge volumes of data and extract meaningful and related content.
"If we were giving an estimate of the risk of epilepsy occurring in a particular case, for example, it needn't just be based on your own experience or one or two papers.
"It could be based on how many cases have been put through an algorithm and what the experience was. What were the outcomes? The data doesn't necessarily all have to come from the personal injury or negligence world.
"It could come from other patients in the general population of brain injury sufferers. This whole process could be powered by AI.”
Whether or not AI is the definitive answer to data problems in neuro-rehab, a new and improved approach is certainly needed, says Bonikowski.
"We lack really good objective markers as so much of this field is subjective. Unlike haematology, for example, in which blood counts can be measured and plotted, we just don't have enough of this type of information.
“Also, the field we are working in is so multidisciplinary and multifaceted. Brain injury covers physical impairment and disability, as well as cognitive and behavioural problems. An injury impacts so many areas. Representing this complex situation within poorly resourced services is virtually impossible.”
Medico-legal experts must swim through oceans of complex data in search of relevant and valuable insights.
Ultimately they are charged with delivering professionally correct advice, while giving the court three vital elements: brevity, clarity and reasoning. The challenge is intensified by an ongoing influx of additional data which must be factored into each case.
"New information comes into the process, perhaps related to the deterioration of the client, the impact of new interventions or perspectives previously unseen.
"The expert must incorporate these into their evolving viewpoint.
Views must also be based on research evidence. Because medical science is very fast-moving, it's not always
easy to keep up with the very latest developments.
"Furthermore, research evidence is often viewed differently from one expert to the next, causing further difficulties."
The adversarial nature of personal injury and negligence cases can also cloud the process.
"Our opinion is for a lay judge to assimilate a complicated medical process. The difficulty for the judge is when they see two opinions and have no way of knowing which one is right.
"Through AI, there is an objectification of the opinion – views become much more objective than subjective.
"This is a fundamental principle of AI; to be able to arbitrate between views and say, with a degree of certainty, what’s right and what’s wrong."
AI also has the potential to challenge the lack of consistency which inevitably arises in tasks carried out exclusively by humans, says Bonikowski.
“AI is simply a way of incorporating human experience into mathematical machines or algorithms."
An algorithm is a process or set of rules to be followed in calculations or other problem-solving operations.
The more information processed by an algorithm, or system driven by algorithms, the more reliable its results.
"If thousands of brain injury cases were put through an algorithm, then the certainty of the outcome, and consistency, increases.
“In contrast, a human expert may recollect their experiences differently tomorrow from how they did today, if their mind is distracted.”
Crucially AI is able to analyse data spanning a far greater number of dimensions than is humanly possible.
While humans may be able to comprehend data on three or four dimensions, AI can make sense of it on thousands, within seconds.
It is therefore possible to find correlations between seemingly unrelated factors which may be relevant to the patient’s case.
Despite such possibilities, neuro- rehab has been slow to take advantage of AI, especially compared to more commercially minded fields.
The modern concept of AI dates back to Alan Turing's days at Bletchley Park. Its acceleration in recent years is the result of a perfect storm of factors.
Computing capacity that can perform billions of iterations in the blink of an eye, the emergence of quality data on a massive scale and surging levels of investment are all contributing to its rapid development.
Research, administration and data- heavy tasks like due diligence are among a range of legal sector duties being influenced by AI.
In life sciences, Google is pioneering AI on several fronts. It recently launched Deep Variant, an open source tool which uses AI to draw a picture of a person's genetic blueprint.
In 2016, it teamed up with Moorfields Eye Hospital to test whether “machine learning” technology applied to eye scan data can speed detection and treatment of eye diseases.
Machine learning is also being used extensively in the insurance industry to enable operators to better understand risk, claims and customer experience.
In financial services, fintech products and so-called 'robo-advice' services are increasingly fuelled by AI.
The UK government is also on board the AI bandwagon. Last year it announced plans to build a new data ethics and innovation centre aimed at ensuring ethical innovations in AI and data-based technologies.
The government is also spending £75m on progressing recommendations made in a recent independent review on AI.
Professions already embracing AI initially did so tentatively, as trust in the technology built up. This will be mirrored in neuro-rehab when the field finally catches up, says Bonikowski.
"We implicitly trust people and their opinions. If you put an expert on a personal injury case in court, there is an immense amount of trust invested in them by the patient's family, the court and society.
"It is this degree of trust which AI machines would need to build up in order to become a fundamental source of support to expert opinion.”
With other areas of healthcare now waking up to the power of AI, neuro-rehab’s widespread adoption of the technology looks to be close.
"I would project that five years from now it will be commonplace for AI systems to be forming a fundamental platform for expert analysis and opinion; with human experts essentially validating what the AI is saying on the basis of interaction with the client and their environment.
"The analysis of all the historical and concurrent data coming in could easily be done by machine in five years.
"It could produce an analysis which states, ‘on the basis of everything here, this is the extent of the injury sustained, here is a sense of the impairment and all the other disruptions that have occurred in terms of quality of life and here are the interventions that have and haven’t been effective’.
"It could then show how an ideal healthcare system would handle the individual and the outcomes one could expect.
"All of that could come into fruition within a five-year horizon.”
An artificially intelligent army is not plotting an imminent invasion of neuro-rehab, however. Bonikowski sees AI’s presence as remaining complementary to - rather than threatening – professional roles in the sector.
"We are seeing tedious and grinding processes being replaced by machines, because they are much better at them than we are.
"The subtler, softer and empathetic human elements almost certainly will never be replaced.
"The idea that machines would be able to interface with somebody with a multifaceted, severe set of disabilities is extraordinarily far-fetched.
"But within five years we should have a really solid expert support system available, with an expert still required to validate findings and add some of the subtleties which may not come out of machine analysis.
"Often in medico-legal reporting we are asked to give a view on what the patient requires to improve functioning. I think that will always require some human interface.
"If you look in someone's eyes and tell them they are going to need a piece of surgery or a very long period of rehab, they are much more likely to engage with a human demonstrating genuine empathy than a machine which simply says ‘from the analysis, this is what you require’.