Interpreting Deep Neural Networks using Cognitive Psychology

If it’s hard to explain how the outputs of complex systems relate to the inputs in terms of sequential processes steps, because of the complexity of the model, then perhaps it makes sense to come at the problem the other way round. Neural networks are very crude representations of human minds, and the way we understand human minds is through cognitive psychology – so what happens if we apply approaches developed to understand the cognitive development of children to understanding black box systems?

That’s both powerful and worrying. Powerful because the approach seems to have some explanatory value and might be the dawn of a new discipline of artificial cognitive psychology. Worrying because if our most powerful neural networks learn and develop in ways which are usefully comparable with the ways humans learn and develop, then they may mirror elements of human frailties as well as our strengths.

Sam Ritter and David G.T. Barrett – DeepMind

One Team Government: What digital and policy can learn from each other

Following his post explaining policy people and processes to their digital equivalents, Paul Maltby has now written a deeply sympathetic but rightly challenging post about the frustrations digital has of policy and policy of digital – and what each side needs to learn from the other.

There is a core insight in each direction. Policy can learn much from the data driven, delivery focused model of digital service design and should be no less comfortable starting with the user need. Digital can gain from appreciating the need to understand and reconcile conflicting goals and interests and the basic principle that politics is our basic method for making public choices – and that digital is political too.

Paul Maltby – Medium

One Team Government: a short guide to policy for government digital professionals

Published to coincide with this week’s One Team Government event, this is an excellent short guide to policy making in government. My only quibble is with its title: it’s addressed to government digital professionals, but that shouldn’t suggest to anybody else – including the policy professionals who are its subject – that they have nothing to gain from reading it.

Perhaps the most important insight it contains is that policy isn’t a single specific thing: good policy (and good policymakers) bring together a wide range of skills and disciplines to address some very different kinds of problems. The synthesis of all that is what we call policy – but the boundary between that and other approaches (not least, in this context, digital) is an artefact of language as much as it is a division of substance.

Paul Maltby – Medium

Building a bank together

A sketchnote on building a bank, in this case Monzo, which is one of a number of new contenders starting in a very different place from traditional banks. The same basic approach could apply much more widely to service design – and indeed to organisation design.

The Critical Difference Between Complex and Complicated

Complicated problems can be solved; complex problems can only be managed. Complicated problems can be addressed algorithmically; complex problems require creativity and adaptability. This article is more concerned with describing that distinction then addressing it, though that may just reflect the fact that it is an extract from a book, It’s Not Complicated: The Art and Science of Complexity in Business, which no doubt provides a fuller account.

In a world of complex problems, plans and strategies do not align tidily with results, which is a reason for approaching them differently, rather than not having them. And that in turn requires organisations which can be comfortable with uncertainty and ambiguity.

Rick Nason – MIT Sloan Management Review

Rise of the machines: who is the ‘internet of things’ good for?

Technology is never neutral. What gets developed, how it gets developed, and how it gets used are all driven by social, economic and political factors. People who build services are never neutral either and can certainly never be normal users of their own services. This article looks behind the internet of things to reflect on how completely frictionless transactions move power from consumer to provider, how what is normal for designers of such services is very different from what is normal for many of those who will find themselves using them, and how technology – and the data it moves and organises – is always about power.

Adam Greenfield – The Guardian

Grenfell Tower, predictable surprises and slow violence

This is a post about black elephants: events widely predicted by those in a position to know, but found totally surprising when they actually happen, or elephants in the room retrospectively declared to be black swans. The Grenfell Tower fire was surprising and shocking – and at the same time, predictable. That puts it in a category of things which human institutions seem particularly bad at dealing with, where a problem builds up slowly and almost asymptomatically until suddenly a tipping point is reached, by which time addressing it has become massively more difficult. At one level, the solution to that is obvious – but that doesn’t seem to make it any easier to do in practice.

And it’s perhaps worth saying that this quite abstract way of thinking about disasters is not an attempt to distract from the human tragedy, but on the contrary is a way of recognising and understanding that we need to deal with structural as well as particular issues if we are to see fewer black elephants in future.

Andrew Curry – thenextwave

Creating the Conditions for Serendipitous Conversations

This is a great post on two entirely different levels. It’s a reading (and listening and watching) list of material on the future of work, with a dozen or so interesting annotated links to follow.

But it’s also an approach to improving the quality of conversations, creating the space to think differently and more creatively, using the shared material to support a richer conversation, based on the insight that “a library of inspiration develops through a lifetime of experiences”. That’s an approach which it feels well worth borrowing – whether on the future of work or any other subject.

Bromford Lab

Why Do We Still Need Managers?

Organisations are essentially a solution to communication problems. In a market economy, firms exist when and to the extent that the costs of communicating and co-ordinating internally are lower than the costs of using market signals – that’s the classic make or buy decision. But the way we communicate has changed fundamentally since the era in which large integrated firms flourished (and since the heyday of Weberian bureaucracies), which suggests at the very least that we should ask whether organisations in the form we know them are still the solution – or are now themselves part of the problem.

This post argues that managers are an expensive and unnecessary overhead, if certain conditions hold (the fact that they generally don’t is presumably an argument for applying them, rather than a counter to the conclusion that management is waste). As ever with Paul Taylor’s posts, the insight is powerful and the writing persuasive. But there is also an element of sleight of hand. Bad kinds of co-ordination are management and to be decried (but if the specific example is approving annual leave, that’s pretty low level), good kinds of co-ordination are leadership, which it seems we need more of, not less.

Paul Taylor

 

Lab+: Potential future states for government service delivery 

This is a great example of what might be called macro user research, investing in understanding people’s own framing of and understanding of their situation and reflecting on how government service design should respond to that. Three broad themes emerge, the first of which is ‘conversational services’. Conversation is a powerful concept, and one all too easily overlooked in the design of high volume services. And it is the starting point for the Cluetrain Manifesto, one of the defining texts of the web, made of up of 95 theses, of which the first three are:

  1. Markets are conversations.
  2. Markets consist of human beings, not demographic sectors.
  3. Conversations among human beings sound human. They are conducted in a human voice

And of course ‘markets’ don’t have to be limited to markets.

Pia Waugh – New Zealand Government Web Toolkit

Reading list for government policy people interested in digital

Paul Maltby asked on Twitter

The collected answers – crowd sourced in short order – make up an impressive list. It’s inevitably a bit uneven, but there is a lot of good stuff there, and it’s well worth dipping in to.

Paul Maltby (assisted by the crowd)

Cars and second order consequences

Predicting the future is hard. Predicting the second and third order consequences of your first prediction is much harder – but it is those consequential effects which really drive the wider social, economic and other impact. This post is about what happens when vehicles are electric and autonomous, and teases out potential changes ranging from reduced tobacco consumption (because in the US most tobacco is bought at petrol stations) to changing patterns of land use.

It’s a characteristically interesting read – but the reason for including it here is less to do with the cars and more to do with its being an example of a way of thinking about the future. It uses a challenging assumption as a starting point – in this case that autonomous vehicles will change cities as much as cars have done. It focuses less on the initial change and more on the ripples that causes. And it recognises that this can only be a way of exploring possible futures, not of predicting a specific one.

Ben Evans

Policy for the future

Good policy comes from good policy making. There is plenty of evidence that good policy making is based not just on the rigorous analysis and evidence assessment which is the best of the traditional approach, but also on effective implementation and deep understanding of the needs and behaviour of those who will be affected by the policy. This post argues for a more broadly based approach to policy making, drawing on The Blunders of Our Governments (which remains compelling reading) to make the case.

The problem with this is not – as the author supposes – that it sounds fanciful – it is that it sounds obvious. The problem is not in recognising that policy were better done differently, it is in the doing of it. The post introduces a more substantial paper which has some useful material, but ends up describing barriers to change without offering much about how to overcome them.

Jake Thorold – RSA

NYU Law’s Algorithms and Explanations

This is a neat summary of questions and issues around the explicability of algorithms, in the form of an account of a recent academic conference. The author sums up his own contribution to the debate pithily and slightly alarmingly:

Modern machine learning: We train the wrong models on the wrong data to solve the wrong problems & feed the results into the wrong software

There is a positive conclusion that there is growing recognition of the need to study the social impacts of machine learning – which is clearly essential from a public policy perspective – but with concern expressed that multidisciplinary research in this area lacks a clear home.

Zachary Lipton – Approximately Correct

Stormzy 1: The Sun 0 — Three Reasons Why #GE2017 Was the Real Social Media Election

Social media have been playing a part in election campaigns for quite a while now, but this year’s general election may mark a tipping point where for an important part of the electorate traditional media are essentially irrelevant. There’s some proper academic caution in this post – it really is too early to tell – but – but if the trend is confirmed, the implications go very much wider than election campaigns.

Helen Margetts – Oxford Internet Institute

Is Social Media Killing Democracy? Computational Propaganda, Algorithms, Automation and Public Life

Social media first played a significant role in electoral politics during the US presidential election in 2000. The first part of this inaugural lecture traces its development since then, through the Arab Spring, to more recent US and European elections, with some interesting insights into ‘computational propaganda’, the role of bots in moving and reinforcing public opinion, and the fake (or junk) news which is often its subject.

The second part of the lecture turns to the rapidly developing connections between big data, behaviour, and the internet of things. It is increasingly possible to derive political inferences from behaviour, such as purchasing patterns, as well as from overt speech – in the internet we have, privacy has essentially been lost. That could be countered, at least in part, by measures to improve the power balance between large organisations and civic society, but there is little current prospect of those proposed getting any traction.

Philip Howard – Oxford Internet Institute

21st Century Automation: Is This Time Truly Different?

Neatly sidestepping the question of whether technology will only destroy jobs or, as every past cycle of technical innovation has done, will also create new (and perhaps currently unimaginable) ones, this post focuses instead on when such a shift can be expected to occur. The period of disruption between the old and the new is important even if it were possible to be confident that the new would be a better place. For much of the nineteenth century, the industrial revolution brought poverty and reduced life expectancy for many – only towards its end did the subsequent century of rising living standards begin. Are we at similar risk of facing serious, and potentially long drawn out, disruption to social order?

Ryan Hagemann and Nicholas Ciuffo – Niskanen Center

Cutting the Gordian Knot of Technological Unemployment with Unconditional Basic Income

This is a powerfully argued article on the strategic case for a UBI – in this case, with a very strong stress on being unconditional. The problem of technological unemployment can be solved only by breaking the strong assumption that income should be linked to work. Doing that is hard because we have been so strongly conditioned to seeing them as inexorably linked – but breaking that link creates new social and economic possibilities and increases overall welfare levels, as well as creating the conditions in which people are far more strongly empowered to exercise their basic human rights. The conclusion is quite a radical one, but the real question is less whether it is right or wrong, and much more whether the starting analysis is well founded.

Scott Santens – Medium