A 21st century Taylorism?

Whether it’s Amazon’s infamous efficiency metric ’the rate’ to set the desired productivity of its workforce, or the continuous 360 feedback app used by NextJump employees – firms are increasingly using data generated about their own business to surveil the productivity and effectiveness of their workforce. Some regard this as the promise of management information and business intelligence finally making it into the real world, others see it as fundamentally inhumane and unethical. Where one person may be motivated by relentless feedback and short-term goal setting (regardless of the progenitor of that goal) others may see it as smothering and repressive. With the investment going into data science and artificial intelligence we can expect to see such working practices get adopted at an increasingly exponential rate.

A timely question seems to be – is this an acceptable form of managing employees? Or are we about to embark on a journey where we demonstrate that we have learned nothing from the de-humanisation of our workforce that was pioneered by Frederick Taylor’s time-and-motion studies of the early 20th century look like a side project, and from which – as an advanced society – we have still not yet recovered (ask anyone who works at McDonalds). Is data-driven decision-making, most of it automated through the use of AI, going to introduce a new management/employee paradigm that creates a deeply unfair society? Are we on the brink of a 21st century Taylorism?

Amazon’s ‘the rate’ algorithm

Amazon is probably one of the better known examples of a company that is using machine learning and data analytics to automatically assign performance targets for its human workforce working in their huge distribution centres. The now-notorious efficiency metric ‘the rate’ is determined and set by an algorithm. 

“Every task in an Amazon fulfillment centre has an efficiency rate. The two most demanding tasks are ’stow’ and ‘pick’. They represent the input to a fulfillment centre (deliveries) and the output from the fulfilment centre (deliveries going out to customers). Stowing requires removing products from their boxes and then stacking them on shelves. Picking requires checking and labelling up products before they are sent out to customers. Robots assist both of these processes. As a box is handled a bar-code is scanned. This creates a data point that Amazon’s automated system then monitors. An employee is informed how well they are doing by a visual graph shown on the monitor on their workstation. The graph changes colour (green, yellow, red) depending on how well they are doing. Task performance is determined by an algorithm. If a worker falls behind the target efficiency rate then they receive a warning. After the fourth warning, a worker is fired. Apparently, supervisors rarely question the targets set by the algorithm and, hence, those workers who are affected do not have a ‘human in the loop’ ensuring that ‘the rate’ is fair. As to be expected, this simple management tool has come to feared by Amazon workers.” – ‘the rate’ (extracts from The Verge article)

The issue here is not that Amazon are using quantitative metrics to objectively assess employee performance – most manufacturing industries use similar methods; the issue is that an algorithm is automatically determining what the rate should be and appears to do so with one aim in mind: continually increase the rate over time. For example, Mohamed began her job at an Amazon fulfillment centre she had a target stow rate of 120 items per hour. In three years that has increased to 280 items per hour. The type of packages and parcels that Mohamed has to deal with hasn’t changed. 
The constant uncertainty over what ’the rate’ will be from one day to the next is causing anxiety. 

Of course, Amazon leadership are loath to change anything as they are achieving remarkable results. The Institute for Self-Reliance found that Amazon only required half of the employees of a traditional retailer for every $10 million in goods sold. Furthermore, the Intitute also note that the general trend is downwards: Amazon are requiring fewer and fewer employees over time to fulfil the same amount of business. Of course, this all affects profitability. Automation (robots and machine learning algorithms) and efficiency (’the rate’) are credited as the main enablers of such performance. 

“Amazon essentially has developed factory-line technology for retail,”, Spencer Cox, PhD candidate and ex-Amazon worker.

NextJump‘s 360 feedback app

NextJump is an e-commerce company that generated $2 billion in revenue in 2018 from linking employees in other people’s companies to perks and rewards from other companies in the form of discounts on things like cinema tickets, eating out at the local restaurant, and so on. But that is not the most interesting thing about them. Rather it is there relentless focus on changing their own culture on, it seems, a daily basis for which the company has gained attention. The company has invested in so-called HR technology that does two things: it provides an app on every employee’s phone that allows them to provide feedback on their colleagues constantly.

Ordinarily, feedback about your performance is something done between a member of staff and their line manager, probably once or twice a year in line with setting personal objectives for the next 6 to 12 months. Not so at NextJump. Here you can expect to receive, and are expected to give, constant feedback after every meeting or day in the office. The person giving you feedback implies a happy, neutral or sad emotion along with a text entry: ‘your enthusiasm is always inspiring to me’, ‘you forgot to mention next week’s meeting’ to ‘I thought you came across as a bit negative’.

Now, I personally don’t see this as a bad thing. After all, it can be difficult to accurately interpret someone else’s feelings at a conscious level unless the other party feels that they can be honest and explicit. Whether I would want constant, 360 degrees feedback every day, I am not sure that my personality type – I have a tendency towards introversion – would find this comfortable. Nevertheless, I think that any technology that can improve the quality of human dialogue and interactions is valuable if used carefully.

Where NextJump appear to stray over the moral/optimal performance line is in how their managers analyse and respond to the feedback trends of each employee. Constantly getting negative feedback? You can expect a visit from the ‘happy police’. Constantly getting positive feedback? That’s great, but are you – perhaps – getting too comfortable? Apparently, getting some negative feedback is seen by NextJump managers as indicative that you are not in your comfort zone – this is deemed good: they don’t want you getting too comfortable otherwise you are not learning. As a colleague of mine once put it: NextJump employee’s appear to never be ‘in flow’. Now, I am firmly signed up to the belief that for any organisation – whether private or public sector – to remain viable and valuable, they need to be able to adapt appropriately to their environment, but a relentless focus on staying out of your comfort zone? There’s just too much constant anxiety going on there for it to be a healthy, long-term mode of working.

Too much change and too often runs the risk that the organisation never actually gets sufficiently good at doing one job well enough to be efficient and cost effective at it. Likewise, too little change, and an organisation runs the real risk of becoming disconnected from its environment and therefore irrelevant. There is a pretty sizeable middle ground that, to take a term from Ulanowicz (a researcher in the field of ecology), is called ’the window of viability’, and I advocate that an organisation should strive to understand where such a window exists within their ecosystem, and how they need to position themselves strategically within such a window. I will talk more about this ‘window of viability’ in a future article.

For NextJump, the ‘feedback app’ is an important part of how they work and this is exemplified in their company slogan ‘dedicated to changing workplace culture’. They are aware that their approach to workplace culture is not ideal for everyone and so they recruit for individuals with certain demonstrative characteristics: a willingness to be coached, and a healthy dose of humility. To understand how NextJump reached this point, we need to consider their history.

NextJump was founded in 1994 by Charlie Kim in his ‘dorm room’. He successfully grew the company to 150 people during the .com boom, and subsequently saw the business scaled back to just four people as a result of the subsequent .com bust. As can be imagined, a lot of soul searching and lessons learned was done, and Charlie and his surviving team, came to the conclusion that the type of people that they had employed – technologists – were not suited to a business that required constant innovation and change to external drivers. After all, the characteristics of a good technologist – expertise, depth of knowledge – are hard earned and, some may argue, should not be given up easily. However, that was not necessarily in line with the need to be open-minded and humble about your own capabilities compared to your competitors, nor the quality of your product compared to what your customers actually need. Charlie Kim and his team subsequently identifed the more desirable characteristics for NextJump employees are an open, growth mindset. 

I believe that NextJump’s efforts to create a culture of continuous change is innovative and admirable. I do, however, have concern over their use of data insights from their ‘feedback app’ to perform constant, micro-adjustments of people’s tasks and behaviours to keep them out of their comfort zones. Now, it may be that for a company of NextJump’s size (as of November 2019, this was 200 people), this is not a problem and that the people who work there find such constant change and challenge as exhilarating. After all, NextJump specifically recruit people who are going to be up for such a way of working. However, where I do take issue is with the almost cult-like zeal with which NextJump wish to share their cultural practices with other companies across the world. I don’t believe it is sensible to assume that NextJump have solved the problem of cultural change. I believe they have developed an approach to organisation that appears to work well for them within their environment.

What do I mean by that? The world of technology moved on from the .com bust of the early to mid-2000’s. What once required staff with scarce technical skills – writing web pages from scratch in HTML – to where we are now – content management systems – means that, as an e-commerce company, NextJump operate in an ecosystem where they don’t need to have people with deep technical skills. Instead, NextJump’s value comes from being a platform in the ‘perks for employees’ ecosystem. Linking employees to businesses requires primarily empathy, creativity and a willingness to move quickly to constantly identify new offerings for customers. That is admirable. However, there is no one size fits all when it comes to culture. Some organisations generate value from being deep technologists, inventors and innovators. Recruiting people who have invested heavily in becoming deep scientific or technical experts may not be what NextJump want because those types of individuals tend to be more reflective, thoughtful, and introverted. Neurodiversity is also a factor that needs to be considered here. NextJump’s culture is deeply social. Neurodiverse people may struggle to feel included in a culture where they are expected to express their thoughts and feelings, constantly, about their colleagues. The leadership imperative should be one of diversity and inclusion rather than only wanting people who can efficiently fit into a mould.

The case for caution over the use of data insights to drive employee performance is clear with Amazon. Lawsuits, employee strikes, bad news coverage, and so on are a clear indicator that Amazon has crossed a line from efficiency to exploitation. Human Rights legislation will have a word to say about that.

The case with NextJump is more nuanced and, in some ways, potentially more concerning. Whilst their immersion courses tend to excite and repel participants in equal measure, I have heard now on several occasions from people who were the former, that ‘we need to be like NextJump’. I often acknowledge the success that NextJump have had with adapting their culture to their environment, however, that is quickly followed by a caveat that the keywords there are ’their culture’ and ’their environment’. By all means let’s be inspired by the success of NextJump, but let’s not fool ourselves into believing that what is good for the goose is also good for the gander. I think this advice is most strongly targeted at leaders who can be attracted by NextJump’s impressive growth trajectory. However, perhaps I worry too much; how many members of the C-suite would be up for wholeheartedly adopting NextJump’s ‘feedback app’ within their own organisation, and where they themselves are on the receiving end too?

Research into causal AI has grown a bit in the last 3 years…. if NeurIPS is anything to go by

I recently wrote a paper for a client on how AI could be engineered to maximise the cognitive performance of a human who routinely needs to make complex decisions with uncertain, incomplete and ambiguous information (aka. no mean feat). I made the point that investing in equipping machines with a causal understanding of some world of interest would be key.

Now, whilst causal inference is not new to me – its my main area of academic research – I did recall that I’d been seeing more published papers on the subject in recent times. This piqued my curiosity so, after logging in to the NeurIPS 2020 site, I did a (very) quick search on all papers that had ever been presented at NeurIPS that had ‘causal’ or ‘causation’ or ‘causality’ in their title. Here’s what I found:

Now, as I said, this was a quick and dirty piece of analysis (i.e. there may have been any number of papers that covered topics related to causal inference such as probabilistic graphical models, and so on, that I have not included here).

But I think the results are somewhat illuminating and, if I was Gary Marcus, I’d be quietly encouraged that the AI community is beginning to enter the 3rd wave of AI where robust, reliable and trustworthy AI will reign.

Accessing the hidden structure of complex systems using information theory

One of the more useful tools in the complexity scientist’s toolbox is information theory. Now, don’t worry, I’m not going to dive into this much, but I do want to talk about the central concept to information theory called Shannon entropy (or information entropy as it is also known). 

In 1952, Claude Shannon – a research engineer at Bell Laboratories – was tasked to invent a method for improving the transmission of information between a transmitter and a receiver. His invention – which he called ‘a mathematical theory of communication’ – was based on a very simple idea: surprising events carry more information than routine events. In other words, if you wake up in the morning and the sun has turned green, then that is going to jolt you into a hyper-aware state of mind where your brain is working overtime to try and make sense of what is going on. When our interactions with friends or our environment reveals information that we were not expecting, then we seek to make sense of it. We process that information with a heightened sense of consciously doing so. 

This response to surprise is no different whether we are individuals (), in a team (discovering that a colleague is also a part-time taxidermist), an organisation (the sacking of a well-respected CEO) or an entire country (the death of Princess Diana). We seek to understand why and in seeking to answer this, we traverse Judea Pearl’s ladder of causation. However, there is one key difference. When we are dealing with a complex system, or situation, then there is uncertainty over cause-and-effect. This uncertainty is the result of a structural motif of a complex system – feedback loops which I will discuss in a future post – that leads to what is called non-linear behaviour.

Information as a level of surprise is measured in binary digits (bits). The more unlikely an event is to occur, the higher the information that is generated if it should occur. Let me illustrate this with the example of flipping a coin. 

When you flip an unbiased coin there is a 50/50 chance of it landing on heads or tails. Because both events are possible – it was heads, or it was tails – then our uncertainty of the result is at its peak. We cannot have more certainty that the coin will land heads up. Here, the Shannon Entropy of flipping an unbiased coin is 1 bit which is the maximum information that can be obtained from a system (a coin flip) that can only generate two outcomes (heads or tails). 

Now, let’s assume that we’ve been given a biased coin that always lands on tails. We know that the coin is biased and so there is no surprise for us when the coin always lands on tails. If there is no surprise, then there is no information. The chances of the coin landing on tails is 100%. In this case, the Shannon entropy is 0 bits. Certainty does not yield new information.

Now, we don’t need to be too concerned with whether something is 1 bit, or 0.5 bit, or 0 bits or whatever. The point I am making here is that the greater the uncertainty we have about something, the greater the information we can gain from that situation. Likewise, if we have total certainty then there is no information, no knowledge, to be gained. Intuitively this makes sense – if I am observing something that is not changing then I am not learning anything new about it. However, if I perturb that system – add a component, remove a component – then I may be cajoling the system into a different state. This new state may yield new information, especially if I have managed to move the system into an improbable state. (Incidentally, this is why the modes of creativity – breaking, bending, blending – are fundamental to discovering new knowledge).

For Shannon entropy to be used in more practical ways, a probabilistic model of a system would need to be constructed. This simply means that we have identified the different states that a system can occupy, and we have estimated the likelihood of the system being in that state at a moment in time. We can construct a probabilistic model through observing and recording the frequency with which different states are observed. The more frequently we observe the system in a given state, over time we may infer that the system is more likely to be found in that state at a future point. Ordinarily we need to capture enough of the history of the system for us to have sufficient confidence in the probabilistic model we are building. This learning takes time and requires continual sampling of the environment; and there are some challenges to solve – like how to represent the environment – but the idea is to invest time in building a probability distribution, a probabilistic model, of our environment. Novelty is a previously unseen state and so that too should trigger a response, not least requiring an update of our probabilistic model.

As we build our probabilistic model we are forming a hypothesis, an untested belief, about how the environment behaves. Every time we observe and capture the state that the system is in, we are testing that hypothesis. The Law of Large Numbers is relevant here. We expect to see a system move in and out of different states. It may spend more time in one state than we have observed before, or the opposite. We would need to see a persistent, recurring change in the frequency with which each state of the system is observed before we begin to suspect that our hypothesis of the system may need to be re-visited.

Now that we have constructed a probabilistic model of our environment (or, indeed, any system of interest), we can calculate its Shannon Entropy. If we have a good degree of confidence that our probabilistic is sufficiently correct, then we can baseline these measures. We can then set a sampling rate of how often we re-calculate the Shannon Entropy of the probabilistic model (we may use machine learning techniques to optimise the sampling rate). If the Shannon Entropy measurement begins to diverge from the baseline value – by some pre-determined tolerance of +/- x bits – then we could infer from this that the system may be changing in some way. This out-of-tolerance measurement could flag the need for further investigation – either by an intelligent agent or a human.

What I am describing here is an idea. I am not aware of any existing technique, or concept, that achieves this. Neither do I know if there is much utility in what I have described. I believe it is technically feasible – the computational complexity of updating a probabilistic model and calculating its Shannon entropy can be achieved in polynomial time (i.e. very efficiently). As such, you should interpret this for what it is; an idea that I hope interests people enough to pursue it further. 

I believe the utility of this technique – of parsing the environment and comparing it against a probabilistic model – could be a very efficient way to manage a vast amount of automated monitoring of an environment for changes that may warrant further investigation. Of course, this ‘further investigation’ would call into play more expensive resources such as AI and/or humans.

My motivation for conceiving of this idea comes back to the need for any organisation to become highly proficient at anticipating change. When the organisation’s environment (internal or external) may be changing in unexpected ways, then we want to be observing the change as it is happening in real-time, rather than analysing after the event. Why is this important? If we are observing the genesis of an enduring change in our operating environment, then we have the opportunity to gain insights to the causes that led to that change.

Applying Shannon Entropy as an early warning system signals an alarm that our knowledge of our environment may no longer be accurate. We can respond to these warning signals by expending effort to understand the changes that may be occurring. From this we may create new knowledge and, therefore, update our semantic graph to represent that new understanding. The semantic graph is critical, because all of our collective intelligence draws on it to make good decisions. If that semantic graph is erroneous or significantly out of date, the quality of our decisions are impacted. As an organisation harnesses AI to the fullest – where we are talking about millions, if not billions, of decisions being taken every second – then the accuracy of the semantic graph becomes a critical and protected asset.

Anticipation gives us time to prepare; yet to accurately anticipate our environment we need to be sufficiently open to detecting changes that suggest that our understanding of the environment may no longer be up-to-date.

I’d like to finish this discussion by making one final point. The use of information theory to measure the behaviour of a dynamic system is not a new concept. Indeed, information theory is one of the most promising tools in the complexity scientist’s toolbox for unravelling the mysteries of a complex system. One of the biggest challenges for the complexity scientist is having access to information about the system of interest. Most of the time we simply cannot access a complex system with the tools we have. To give just a few examples: the brain, the weather, genetics. It is neither practical, nor feasible, for a complexity scientist to have access to every element or aspect of systems of this kind. Yet we are not without hope. As long as we can capture the signals, the data, the transmissions, from these systems then we can begin to understand the system, even though it is hidden from us. Of course, as we gain more knowledge of these systems, we can then devise precise interventions that may yield crucial insights that either confirms our hypotheses or takes us completely by surprise. 

Up until recently, I had been researching the use of information theory to infer the causal architecture of a system. Techniques such as Feldman & Crutchfield’s causal state reconstruction, or Schreiber’s Transfer Entropy, or Tononi’s Integrated Information Theory were all part of my toolkit. They are all valuable as they can tell us something interesting about a complex system. However, they do not have the explanatory power of causality, especially Judea Pearl’s do-calculus. I pass on this observation here to those readers who may be more familiar with these subjects.