Data and AI

AI (Deep Learning) explained simply

A straightforward and very useful post which does exactly what the title says – a step by step explanation of what AI can now do, what it might be able to do and what there is currently no prospect of its doing (for economic as well as technical reasons).

Fabio Ciucci

Data and AI

How to Call B.S. on Big Data: A Practical Guide

It is – or should be – well known that 82.63% of statistics are made up. Apparent precision gives an authority to numbers which is sometimes earned, but sometimes completely spurious. More generally, this short article argues that humans have long experiences of detecting verbal nonsense, but are much less adept at spotting nonsense conveyed through numbers – and suggests a few rules of thumb for reducing the risk of being caught out.

Much of the advice offered isn’t specific to the big data of the title – but it does as an aside offer a neat encapsulation of one of the very real risks of processes based on algorithms, that of assuming that conclusions founded on data are objective and neutral, “machines are as fallible as the people who program them—and they can’t be shamed into better behaviour”.

Michelle Nijhuis – the New Yorker

Data and AI

Will artificial intelligence ever understand human taste?

Were the Beatles average?  This is Matthew Taylor in good knockabout form on a spectacular failure to use data analysis to understand what takes a song to the top of the charts and, even more bravely, to construct a chart topping song. The fact that such an endeavour should fail is not surprising (though there are related questions where such analysis has been much more successful, so it’s not taste as such which is beyond the penetration of machines), but does again raise the question of whether too much attention is being given to what might be possible at the expense of making full use of what is already possible. Or as Taylor puts it, “We are currently too alarmist in thinking about technology but too timid in actually taking it up.”

Data and AI

Ethics Should Precede Action in Machine Intelligence

This is an extract from a new book, The Mathematical Corporation: Where Machine Intelligence and Human Ingenuity Achieve the Impossible (out last month as an ebook, but not available on paper until September). The focus in the extract focuses on the ethics of data, with a simple explanation of differential privacy and some equally simply philosophical starting points for thinking about ethical questions.

There is nothing very remarkable in this extract, but perhaps worth a look for two reasons. The first is that the book from which it comes has a lot of promise; the second is a trenchant call to arms in its final line: ethical reasoning is about improving strategic decision making.

Josh Sullivan and Angela Zutavern – Sloan Review

Data and AI

Three very different sources of bias in AI, and how to fix them

This is a short sharp summary of how biases affect AI design and what to do about them, reaching the conclusion that government oversight is essential (though not, of course, sufficient). There are interesting parallels with Google’s in house rules for working on AI, so worth reading the two together.

Joanna Bryson – Adventures in NI

Data and AI

You Say Data, I Say System

What’s the best way to arrange the nearly 3,000 names on a memorial to the victims of 9/11 to maximise the representation of real world connectedness?

Starting with that arresting example, this intriguing essay argues that collection, computation and representation of data all form part of a system, and that it is easy for things to go wrong when the parts of that system are not well integrated. Focus on algorithms and the distortions they can introduce is important – but so is understanding the weaknesses and limitations of the underlying data and the ways in which the consequences can be misunderstood and misrepresented.

Jer Thorp – Medium

Data and AI

Google’s Rules For Designers Working With AI

If machine learning is not the same as human learning, and if machine learning can end encoding the weaknesses of human decision making as much as its strengths, perhaps we need some smarter ways of doing AI. That’s the premise for a new Google initiative on what they are calling human centred machine learning, which seems to involve bringing in more of the insights and approaches of human-centred design together with a more sophisticated understanding of what counts as a well-functioning AI system – including recognising the importance of both Type I and Type II errors.

Katharine Schwab – Co.Design

Data and AI

Your brain does not process information and it is not a computer

Artificial intelligence is more artificial than we like to think. The idea that computers are like very simple human brains has been dominant pretty much since the dawn of computing. But it is critically important not to be trapped by the metaphors we use: the ways in which general purpose computers are not like human brains are far more significant than the ways in which they are. It follows that machine learning is not like human learning; and we should not mistake the things such a system does as simply a faster and cheaper version of what a human would do.

Robert Epstein – Aeon

Data and AI Social and economic change

Marketing Data Predictive service: Are you a butler or a stalker?

Does the power of big data combined with location awareness result in our being supported by butlers or harassed by stalkers? There’s a fine line (or perhaps not such a fine line) between being helpful and being intrusive.  Quite where it will be drawn is a function of commercial incentives, consumer responses and legal constraints (not least the new GDPR). In the public sector, the balance of those forces may well be different, but versions of the same factors will be in play. All of that, of course, is ultimately based on how we answer the question of whose data it is in the first place and whether we will switch much more to sharing state information rather than the underlying data.

Nicola Millard – mycustomer

Data and AI

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

Data and AI

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

Data and AI

What does it mean to ask for an “explainable” algorithm?

It’s often said that decisions generated by algorithms are inexplicable or lack transparency. This post asks what that challenge really means, and argues that there is nothing distinctive about decisions made by algorithms which makes them intrinsically less explicable than decisions made by human brains. Of the four meanings Felten considers, the argument from complexity seems the most prevalent and relevant – and where his counter argument is weakest. But this is an important and useful way of clarifying and exploring the problem.

Ed Felten – Freedom to Tinker

Data and AI Organisational change

8 tribes of digital and the rise of the robot army

‘Digital’ risks becoming ever more shapeless as a word – as it increasingly means everything, it necessarily means nothing. In a post three years ago, Catherine Howe brought some rigour to at least one aspect of the issue, identifying seven tribes of digital. Now she has felt the need to add an eighth – the robot army, reflecting the shift she see from large scale automation being an interesting theory to becoming a practical reality.

Catherine Howe – Curious?

Data and AI

To err is algorithm: Algorithmic fallibility and economic organisation

An interesting take on the problem of algorithmic reliability: treat it as an economic problem and apply economic analysis tools to it. Algorithms are adopted because they are expected to create value; a rational algorithm adopter will choose algorithms which maximise value. One dimension of that is that if the consequential cost of errors is low, the value of an improved algorithm will also be low (setting an inconvenient appointment time matters less than making an incorrect entitlement decision). More generally, decision making value is maximised when the marginal value of an extra decision equals its marginal cost.

One consequence of taking an economics-based approach which this article doesn’t cover is the importance of externalities: the decision maker about an algorithm (typically an organisation) may pay insufficient weight to the costs and benefits experienced by the subject of a decision (often an individual), so producing a socially sub-optimal outcome.

Juan Mateos-Garcia – NESTA

Data and AI

When algorithms are racist

Joy Buolamwini speaking at TEDx

This article on the biases of algorithmic decision making is notable for two reasons. The first is that it comes from  a national newspaper, suggesting that the issue is becoming more visible to less specialised audiences. The second is that it includes a superbly pithy statement of the problem:

Our past dwells within our algorithms.

There is also an eight minute TEDx talk, mostly covering the same ground, but putting a strong emphasis in the final minute on the need for diversity and inclusion in coding and introducing the Algorithmic Justice League, a collective which aims to highlight and address algorithmic bias.

Joy Buolamwini – The Observer

Data and AI Organisational change

How artificial intelligence will change the future of government

There is no reason to think that governments as organisations are any less vulnerable to the disruptive effects of automation than other kinds of organisations. As process delivery organisations, they are not fundamentally different from other process delivery organisations, and are certainly not immune to the pressures which are reshaping them (though they may be slower to respond to changing expectations). How far, though, might AI take over the policy development functions of government? More than you might think, is the argument here, asserting that governments have a moral obligation to make the best use of AI.

Danny Buerkli – Centre for Public Impact

Data and AI

The Blockchain Immutability Myth

This is a more technical post than most which appear here, on the apparently arcane point of whether blockchains actually are (or even should be) as immutable as is sometimes claimed for them. That matters to those interested in the use of blockchains, rather than their technology, for two reasons. The first is that although the distributed design of a blockchain – particularly a public blockchain such as bitcoin – makes it hard to compromise, hard is not the same as impossible, and understanding where motive and opportunity might overlap is important to decisions about how and when it is sensible to use them. The second is that the pattern of interests and opportunities may be different for large institutions such as government. Public blockchains are kept honest by large, computationally intensive processes which, by design, are expensive and inefficient, but which allow participants in exchanges to be trusted without needing to prove that they are trustworthy. If other forms of trust are available, the overheads of bitcoin-like blockchains can be avoided, bringing a slightly different mix of risks and opportunities.

The article also contains a neat and succinct description of how blockchain actually works (though it might not be the ideal starting point for people completely new to the subject).

Gideon Greenspan – Multichain

Data and AI

Australia, we need to talk about data ethics

If Weapons of Math Destruction shows how data models can be bad, that still leaves us with the question of how to tell the good from the bad, and how to make judgements about what might count as good. This post sets out some potential ethical criteria to help data scientists – and even more importantly, those who commission and use data science – to make good decisions. There are some strong parallels with Jeni Tennison’s talk on countering bias in data, perhaps not surprisingly given a common background in ODI.

Ellen Broad – The Ethics Centre

Data and AI

Predicting the tides based on purposefully false models

As a neat little example of the general proposition that all models are wrong, but some are useful, David Weinberger unearths a mechanical tide prediction engine from 1914, which refined the accuracy of its forecasts by using an increasingly unrealistic model of the solar system, with planets and moons being invented with gay abandon to make the results come out right. That’s a beautiful example of one of the points Weinberger is making in his article about machines, knowledge and artificial intelligence – that there will be complex consequences of systems which increasingly appear to have useful outputs without it being easy – or even possible – to work out how they were derived.

David Weinberger – JOHO

Data and AI

The AI Cargo Cult: The Myth of a Superhuman AI

A long and thoughtful essay challenging the idea that artificial intelligence will take over everything and displace humans. This isn’t the argument that as machines do more things, humans will move on to do other things, it’s a more a fundamental assertion that intelligence simply doesn’t work like that, and that ultimately reproducing the set of things humans do well is most easily done by reproducing humans – which we already have a rather effective way of doing. That’s not to say that some machines won’t be better at solving particular sorts of problems than humans are – self-evidently, that is already the case and has been for a very long time – but that won’t make them ‘superintelligent’, any more than my calculator is superintelligent because it does arithmetic better than I do.

Kevin Kelly – Backchannel