Blockchain, bots, and machine learning: Demystified18.05.2018
Simplifying intimidating technology words and their applications for healthcare research.
The trouble with emerging technologies is that they usually come with new terminology that can make them intimidating to approach. This can mean their adoption is delayed or rejected because it’s easier to procrastinate until they’re commonplace than to grapple with things that are perceptually complex. This ‘terminology intimidation’ is particularly true for three technologies which have potential value in the healthcare market research environment: blockchain, machine learning, and bots.
At HRW, as part of our never-ending quest for better approaches, we’ve been watching these technologies for a while and are keen to demystify them so all of us across the industry are more informed and ready to discuss the best applications. What’s more, these technologies are already being successfully applied in the healthcare space, so whilst we of course need proof of concept and in a market research/business intelligence context, we are already seeing their application in adjunct industries and need to be informed as this technology may already be in your home or workplace.
In simplest terms: A secure yet transparent record-keeping method that is permanent and un-editable.
What it is: Blockchain was invented in 1991 but was largely unused until 2008 when it was adopted for the online currency bitcoin by the enigmatic Satoshi Nakamoto. In the bitcoin application blockchain functions as a bank and transaction ledger; storing secure records of buying and selling transactions for the currency. Blockchain works well for bank records because it is a timestamped and un-editable series of data. It is called blockchain because data are captured in blocks (packets of data). Each block also includes a cryptographic ‘hash’ (a unique identifier that functions like a fingerprint) and each hash also includes the unique identifier of the block that precedes it. This approach is called ‘one-way hash’ and deepens the security because it is randomly created, so not predictable to imitate or hack. If somebody edits a block, it is corrupted, so its hash changes which corrupts the entire block, as the preceding hash cannot read/match the next one anymore, meaning you can’t go in to a single record and alter it because the whole data set would be threatened. The hash-validated relationship of each block to the preceding block is shared with a secure global network of peer-to-peer computers (which all can see the block’s hash, but not its contents). This network each need to see the same hash relationship before the block is validated and added to the chain, giving a delay to the process that ensures no rogue actors are tampering with lots of data at once. Furthermore, the decentralised nature means that even if a single computer was hacked, it would be too difficult to identify the others and hack them as well, so the network would recognise the anomaly and block access, protecting the data. Therefore, data are always stored in sequence, securely, and cannot be edited.
Case studies of applications in healthcare: In the US, Medicalchain – a blockchain based method for storing medical records securely is being piloted.
Why people are excited: It’s extremely secure and yet transparent, because it is not only encrypted but also decentralised (the network of related blocks of records are validated by secure networks of individual computers); so the security of the network is very transparent, yet the contents of the data blocks are secure. Furthermore, this constant validation means the whole set of information (chain) is less susceptible to hacking or manipulation.
Possible research applications: Protecting raw data, legal documentation, or storing research to ensure it’s not modified or manipulated.
Considerations and limitations: GDPR. With the implementation of General Data Protection Regulation in May of this year, the potential for blockchain as a method of transferring data is both improved and limited. The need for transparency and security in who views/holds data means that this method could hold value over other file transfer/data storage methods. However, the decentralised nature could pose a problem if the network is split across multiple geographic areas and would need to be structured clearly to explain where/how their data will be stored to participants. Furthermore, data in blocks cannot be deleted, so with the requirement from GDPR to delete personal data after use and not store it would require the creation and deletion of entire blockchain networks.
Our verdict: From our experience, there aren’t significant unmet needs in data integrity in market research. The contract between clients and agencies involves trust that the data will be appropriately represented, so this format for data storage would only really hold value versus current data storage methods in cases of extreme confidentiality, sensitivity, or lack of trust.
2) MACHINE LEARNING
In simplest terms: Using computer power to adaptively ‘learn’ patterns from data sets.
What it is: Machine learning is an application of a bigger (and perhaps more intimidating) category of technology, Artificial Intelligence (AI), but all it means in this context is that the computer power is not limited to performing pre-defined functions or human control; but has flexibility to act on feedback from its performance, learn from new datasets, and continue to improve. It is an evolution of the computer power we already use to analyse and recognise patterns in data sets, but with the added advantage of being adaptable to changing conditions – with the goal to get computers to think more like humans, in terms of recognising and classifying data. For example, machine learning starts with some target (this can be a fact, a decision, or a prediction) – e.g. survey data – and we can ask it to predict an endpoint; the computer will troll through to identify which fields appear to be related; much like CHAID or correlation analysis but less limited to the pre-defined input variables. Machine learning creates and augments an algorithm to explain the relationships between the variables within a degree of certainty; the unique benefit is the use of neural networks (feedback loops) – which relay if the decision outcome was right or wrong and thereby refines the accuracy of the algorithm on future runs. Therefore, if you collect a large enough data set and an effective enough programme, you can evolve an algorithm that is ‘smart enough’ to predict one variable based on the other(s).
Case studies of applications in healthcare: Many companies are using machine learning to analyse data collected from wearable devices to link together time and location stamps with a variety of biometric measures to see what has an impact on health and activity.
Why people are excited: It is just making life simpler for people the world over, finding accurate suggestions on Netflix without having to trawl through whole libraries, application in modern smartphone cameras which capture and reference a growing database of images daily to better understand how to adapt to and capture them more precisely, post processing our photos automatically for balance, contrast, brightness in seconds. It allows smarter analytics to throw up patterns wherever they lie because it is self-directed rather than being guided based on what you put in. It is also exciting because it looks at entire data sets and can combine variables now and again to see whether that has a role and will adapt to growing/changing data sets.
Research applications: Machine learning already powers many automated analytics programmes and will only grow in potential as it becomes more commonplace. It has particular value in conjunction with therapy area/ behavioural psychology understanding as it can sometimes uncover patterns that may seem counterintuitive that psychological phenomena may explain. For us at HRW, it has a lot of value in diverse aggregated data sets like normative databases for common tracking parameters that we use like Net Promoter Score; to allow us to further distil what characteristics really relate to generating brand advocacy within and across therapy areas.
Considerations and limitations: If it’s not a person, who has access to this data, are there any GDPR implications of analysing data in this way? Also, machine learning may come up with crazy things – it will seek efficiency, but these may not be commercially viable or socially acceptable (e.g. it might tell you that experiencing a particular side effect drives satisfaction).
Our verdict: Whilst we already regularly work with providers who use machine learning in creating their analytical tools, most of the data sets we collect in healthcare are too small to use machine learning effectively at the moment. But we’re so excited about the potential for this approach to help our clients leveraging the data they’re already sitting on and are also keeping a close eye on partners delivering self-evolving coding and code frames via machine learning to improve turnaround time on quantitative surveys.
In simplest terms: Simple computer programmes that respond to cues in free text fields.
What it is: Although the title might have you imagining an android or R2-D2 character, a bot is a simple set of ‘if/then’ programmes that allow a computer to have a ‘conversation’ with a person in an email, phone call, or chat thread. Bots use word detection to get a sense of audio or free text entered by the human and then deliver their scripted response depending on which trigger word is present. Many large consumer companies (like banks and phone companies) are using bots for customer service.
Case studies of applications in healthcare: Most applications use the technology to aid ‘virtual’ diagnosis conversations, but excitingly charities have been using bots to help adolescents and teens with depression through positive reinforcement. The young people are able to chat with the virtual partner who provides encouragement and positivity.
Why people are excited: Bots allow for a more engaging exchange of simple information as opposed to data stores; saving money and making information entry and retrieval more conversational and easy-flowing.
Possible research applications: The main application is to make surveys more conversational, rather than clunky routing, the participant would feel as though they are having a conversation and the programming would facilitate the correct filtering. It could also be applied in qualitative ongoing/longitudinal research – install on your phone and talk to it about experiences, and each time you bring up something of relevance, it asks specific probes. Thinking bigger, we see a future where we could work with clients to create a bot that works with a dynamic research library; you could ask the bot questions and the computer would search the database to point out relevant project reports.
Considerations and limitations: It is relatively simplistic and constrained. Bots are most effective for simple surveys because the technology is quite rudimentary and relies on recognitions of correct themes based on specific words rather than their synonyms (which is why many consumer customer service bots ask questions like “If you would like to speak to a representative, say ‘representative’”).
Our verdict: Bots hold a lot of potential for replacing or augmenting the survey interface to make the process more enjoyable and natural for participants; possibly reducing respondent fatigue, and even allowing participants to feel as though they are having a conversation, but one where the ‘moderator’ is incapable of judging them because they’re ‘just’ a robot. And hey, who wouldn’t want to tell C3PO a secret?
So, as we follow these technologies unfolding, we are excited about their potential. And the future also holds some intersections of their potential: bots can make obtaining information simpler, but the best will rely on some machine learning in order to get to a level which is effective for the diversity of respondents, yet transparency and trust could be a large barrier to our participants conversing with a ‘robot’, which the security of blockchain could help to reassure participants with.
To share your thoughts on the use of this technology or for more information on the digital innovation we’re applying at HRW, please get in touch.