Recently on LinkedIn I saw a post from elsewhere in the data community, quoting the famous claim made by the late American economist and statistician, William Deming, that: ‘without data you’re just another person with an opinion’. The quote was, I assume, posted for the benefit of colleagues and other professionals working in data, broadly conceived. The post featuring the quote served as a lead into an article unpacking the various ways in which data literacy is a vital business skill, and it attracted nearly 2000 ‘likes’ and other endorsements, including from people working in or interested in various ways in data ethics.
Of course, the meaning of the Deming’s claim is intelligible, and at first sight it could even appear to be true, useful, valuable, informative, insightful, and so on. However, when subjected to even light critical scrutiny, it becomes clear that the claim is, unfortunately, none of those things. Rather, and as I aim to show here, the claim is specious: despite any superficial ring of plausibility, under examination, the claim turns out to be misleading or false. So, I hope that the analysis here makes it clear why the implications of the claim are at best incoherent, and at worst dangerous.
Fortunately, not all of the responses to the post endorsed the quote; several people pointed out some of the shortcomings that I’ll cover here. Similarly, even prior to this quote being posted on LinkedIn, I’m not the first person to point out the troubling implications of Deming’s claim. Indeed, as the writers of this piece show, it might be more accurate to say that the inverse of the claim is true; namely, that without an opinion, you’re just another person with data.
It’s important and a good thing that the problems with the quote are pointed out and explained, because in the interests of ensuring professional ethical practice in data handling, avoiding being taken in by specious claims or arguments, and knowing how to spot them, is vital. Earlier this week, in a discussion about data and AI ethics at MKAI’s excellent Global AI Ethics and Safety Summit, being held just next door to the government-backed event in Bletchley, I gave a potted version of the analysis that I’m going to lay out in full in this article. From this analysis, as well as showing the problems with the quote, I hope to demonstrate why, if you or your organisation face ethical challenges in any aspect of data handling and you credible, coherent advice, assistance or guidance in navigating these, you should come to us at IGS.
So, to begin, there is a shorter way and a longer way to show why the claim is false. In the interests of being as clear as possible, I’m going to do both – in presenting the longer version too, I’m going to show my working, so to speak. But first, here is the short version, which should quickly point to where the argument will go.
The short version
Imagine I say to you ‘Sexism, racism, homophobia, ableism, capital punishment, murder, rape, and torture are wrong’. It would be odd, to say the least, if your response to this assertion was ‘That’s just your opinion: where’s your evidence? Can you show me the data to back it up?’
Why would that be an odd, and indeed, troubling, response? Well, if it were true that I needed data to be able to successfully defend the claim that ‘Sexism, racism, homophobia, ableism, capital punishment, murder, rape, and torture are wrong’, then the moral wrongness of those acts would indeed be merely an opinion, that is to say, something which is open to doubt in the absence of evidence being presented to show why the opinion is true.
I am doubtful that many of us would be comfortable with this conclusion, nor find it remotely plausible. And this is because, when it comes to serious moral wrongs such as those listed, it is important that we can all commit to the view that their wrongness is not just a matter of opinion, but is in some sense true independently of what any particular person happens to think.
The long version (and showing the working)
It should be clear from these two paragraphs alone why the claim in question is specious; which is to say, appealing but false. Nevertheless, in case it isn’t completely clear, and, indeed, in the interests of good analytic practice, in this section I’ll go into more detail to show why it’s false and why it’s therefore not to be taken seriously.
Giving the benefit of the doubt
In general, it’s a good habit in philosophy to give the benefit of the doubt to the position or argument that you’re trying to debunk. This is because if you want to show why your analysis is correct, it strengthens your case to assume the plausibility of the claim or argument that you’re opposing, and then go on to show why that isn’t the case, based on what else would need to be true for it to be plausible. So, let’s do that here.
What would giving the benefit of the doubt consist of in this case? Well, first we would accept as plausible until shown otherwise the idea that data is needed to convert something from being a ‘mere’ opinion into a – more or less – objective fact. So, let’s do this and imagine again how the exchange sketched in the short version might play out. It could go as follows.
I say to you ‘Sexism, racism, homophobia, ableism, capital punishment, murder, rape, and torture are wrong’, and in response you say to me ‘That’s just your opinion: where’s your evidence? Can you show me the data to back it up?’ I grant you the benefit of the doubt that I need to provide you with some data to back up my assertion, so I point to the numerous harms and profound suffering caused to people by sexism, racism, homophobia, ableism, capital punishment, murder, rape, and torture, as evidence that would show my ‘opinion’ to be true in some more objective sense. Being a fair and ethically-minded person, you accept these harms as sufficient evidence for the truth of my assertion. You are satisfied that because I can point to data that you find persuasive, I’m no longer ‘just another person with an opinion’, but a person saying something that you can regard as true in some more objective sense.
Unfortunately, however, when we subject this next step to even similarly brief critical scrutiny, the scenario still produces problematic implications. Fortunately, in the scenario, you are, as I have said, a fair and ethically-minded person; that is to say, someone who would regard it as immoral to: cause people harm and suffering by being sexist, racist, homophobic, ableist; or to endorse the state putting individuals to death as punishment; or to murder, rape, or torture someone. Here, it is your opinion that these practices are morally wrong because they cause harm and suffering, which leads you to judge the data negatively. Can you see where this is going now?
To make it explicit, let’s imagine that you aren’t a fair and ethically-minded person, but some kind of sadist or prejudiced individual, who approves of the harm and suffering caused by sexism, racism, homophobia, ableism, who is a supporter of the death penalty, or who finds murder, rape or torture appealing because of the harm and suffering that they cause to others.
In this scenario, the exchange would be entirely different. This is because, were I to present them with data showing the harm and suffering caused to people by these practices, they would, presumably, judge it positively, rather than negatively. Here, again, it is their opinion which leads them to this reading of the data. In this respect, the data by itself does nothing to take my assertion about the wrongness of those practices beyond being ‘mere’ opinion and into the articulation of some more objective kind of moral truth.
If the point I’m making here wasn’t clear before, I hope it is now. Your judgement of the data and the interpretation you make of it as the basis for deciding what you ought to do is not grounded in the data: it is grounded in your opinions. Your opinions about what is right, wrong, permissible and obligatory come prior to the data presented to you, and it is virtue of the opinions you make the particular judgement that you do about the data. The justificatory grounding for a claim goes in the opposite direction from how it is presented in the quote; from the opinion to the data, and not the other way round. To be clear, my claim is that the practices listed are moral wrongs. Of course, that is my opinion, but I am prepared to argue that it’s not only my opinion, but a view that aligns with what we ought to value morally in a more objective sense.
So, if the moral validity, truth, objective reality, and so on of one’s opinions is not grounded by the data, what is it grounded by? Well, it’s grounded in the adequacy of the reasons that you have for holding the opinions that you do. And, unfortunately, assessing the adequacy of reasons is a far more difficult, time-consuming, intellectually demanding task than just pointing at some data as ‘evidence’ for what you think and assuming that this is sufficient. Unfortunately, there is just no way around this, and next I’ll demonstrate why.
The foundational importance of reasons
Let’s imagine our scenario again and extend it even further. Suppose I present you with the data showing the harm and suffering done by the immoral practices listed, and you, being a fair and ethically-minded individual accept it as evidence for why my claim that those practices are morally impermissible is not ‘just’ my ‘opinion’. Having been satisfied by the data, you assume that this is the end of the matter.
However, I decide to push you on why you accept the data as evidence, proof, and so on. So, I say to you ‘OK, great, you agree with me. So, please explain why you believe that the harm and suffering caused to people by these practices is evidence for the moral wrongness of those practices in a way that now makes my assertion more than ‘mere’ opinion’ What would you say in response?
Well, presumably, you might respond by, for example, telling me that they are wrong because all people are morally equal, and since we shouldn’t treat people differently on the basis of differences that have no bearing on their moral status, it’s arbitrary to mistreat individuals based on their sex, skin colour, sexual orientation, physical capabilities and so on. Likewise, you might say that capital punishment is wrong because, for example, the risk of irreversible miscarriages of justice is too high a risk to justify it; and / or that if we regard deliberate killing – that is to say, murder – as morally impermissible, then the state is not morally entitled to deliberately kill either, even if as a means of punishment. And you might say that rape and torture are moral wrongs because nobody is morally entitled to inflict such profound and irreversible psychological and physical harm and suffering on another person either. All of these ‘opinions’ are entirely reasonable to any fair, ethically-minded person, because we can give reasons why we ought to commit to their being more objective moral truths.
Why ‘opinions’ matter
But what if I continue to push you and ask for the ‘data’ which backs up your ‘opinion’ that all humans are morally equal, and the equality of moral status is not altered by sex, skin colour, sexual orientation, physical capabilities, and so on and so on. Remember, of course, that the ‘data’ we are referring to here is precisely what I’ve just asked you to unpack with respect to how and why it shows that the practices are moral wrongs. So, what other ‘data’ could you possibly present to show this, in a more fundamental way that goes beyond simply pointing to the harm done by those practices, and saying that the data are conclusive because causing harm is wrong?
Eventually, to defend the view that you also hold about the immorality of the practices, you would be forced to ground that defence in something other than data, because data is always open to competing evaluations. You would be forced to ground your defence of why your view is not ‘just’ an opinion by appeal to reasons. And, presumably, you could do this relatively easily, because you could explain that, for example, skin colour, is an arbitrary basis on which to favour or discriminate against individuals. You could also defend the moral wrongness of torture by appeal to what we all know about pain, and make the claim that if we reflect on why we avoid pain, we will be able to see why it is wrong to inflict it on others, given that we have also already given a justification for everyone having the same moral status, such that no person is any more or less deserving of being tortured.
And so on, and so on. The essential point here, is that, eventually, however much of a data evangelist you might be, when you really get down to brass tacks about what you, I, someone else, or an organisation ought to do and why, the data will not tell you whether or not the ‘opinion’ you hold about that it is right or wrong. We can keep appealing to the data as long as we like, but, eventually, we will have to hit the ground if we want to stop being pushed for an answer. The answer can only come by appeal to the reasons why you judge the data in the way that you do. So, relegating a point of view about right or wrong to being ‘just’ an opinion in the absence of any data to back it up is a risky business, because if we really do believe that an opinion is only validated by data, this belief yields two unacceptable conclusions.
First, it entails the conclusion that my ‘opinion’ that sexism, for example, is a moral wrong is only a matter of opinion without some ‘data’ to prove that the ‘opinion’ is true in some more objective sense. Second, it entails the conclusion that, given some people will approve of the only data we could present, namely, the harm caused to people by sexist prejudice, if the quote were true, it would follow that the sexist person can show that their ‘opinion’ that sexism is fine becomes more than ‘just’ opinion simply by presenting you with data that shows the harm caused by sexism, because they approve of that data, because they happen to think, mistakenly, that sexism is morally acceptable.
I’ll leave this exhaustive treatment of the scenario here, as I trust that I’ve made it sufficiently clear why the quote in question is not to be taken seriously, and is, in fact, nonsense.
But why does this matter?
What I’ve unpacked here matters for several important moral and ethical reasons.
Back to reasons again
First, it matters because if the quote were true, then any view would be ‘just’ an opinion without any data to back it up. And, to return to the short example, I don’t think any of us are comfortable, and nor should we be, with what this would entail; namely, that if I say racism is morally permissible and you say it’s morally impermissible, there is no fundamental sense in which either of us is more right or wrong than the other about that. If it’s ‘just’ an opinion that racism is right, it’s ‘just’ an opinion that racism is wrong. And, if we think that something like basic human rights are valuable and must be upheld, we cannot accept that conclusion. So, if the data can’t ground the truth, we have to ground it in some more foundational account, based on a justification from first principles, why all people have equal moral status irrespective of skin colour and other differences. If we’re not prepared to ground that defence in something more foundational than whatever data happen to have been collected or are available – that is to say, if we’re not prepared to ground our defence in reasons – then we are without the kinds of robust ethical moorings that we need for informing the opinions that we hold, and for how we interpret the data available on the basis of those opinions.
Data **can’t** tell you what **ought** to think or do
Second, and relatedly, the point I’m making here matters because if the quote were true, it implies that, in some sense, it will be obvious just from looking at the data what we ought to do. But this cannot be true either. Given the data-driven, data-ubiquitous, and indeed, data-incontinent world that we now live in, we know that data is necessarily partial, can be unreliable, is highly vulnerable to bias, and so on. As such, it would be irresponsibly naïve of anyone who works in data just to say ‘look at the data’ if you want evidence to shore up your ‘opinions’, because the provenance of the data should not be assumed and should be analysed and assured before we judge it trustworthy, objective, a reliable basis for making or supporting judgements, and so on.
To give those who posted the quote the benefit of the doubt, again, we should assume that they mean ‘without data you’re just another person with an opinion’ **on the basis of having first satisfied themselves about the provenance of the data**. But here it is us who have been generous enough to give them the benefit of the doubt: they have not demonstrated that this is in fact what they mean, because they have not made it clear. And here we can approach the end of this piece by thinking about what the analysis here means in practical terms for who you should look to if you need coherent, credible, analysis about challenges in data ethics. Before I conclude, however, a quick digression.
If data can’t tell you what to do, why are you working in data ethics?
One reasonable question that could be levelled at me, or any other data ethics professional making the same argument, is: if the data are irrelevant to forming moral judgements or can’t help us to distinguish between what is ‘just’ an opinion and what is, in fact, true in a more objective sense, why are you working in data ethics? The answer to this is straightforward. It’s because of the ubiquity, complexity, and potential unreliability of the data-ubiquitous world that we live in, that we need to: think on the basis of reasons what the conditions are for us to be able to judge something as true or false; and be able to apply critical scrutiny to data to assess its reliability, validity, trustworthiness, and so on. For sure, we need to make decisions based on the data available, but how we engage with it and the degree of scrutiny that we apply to it as a basis for making our decisions, in combination with the principles that we believe, as a matter of reason, that we ought to uphold, is up to us and can’t be read off the data alone.
Why you should choose IGS
In the final section of this article I’m going to spell out why, if you are facing challenges in any aspect of data ethics, you should hire IGS to help you navigate them and decide what to do.
The extent to which the quote was met with approval was alarming to me. I argued in an earlier article that if you want expertise in data ethics, you should hire a professional in data ethics, where an unavoidably necessary component of that expertise is high-level academic training in philosophy in general, and theoretical and applied ethics in particular. I do not need to repeat the argument here. Suffice to say, I hope the analysis in this article gives further weight to that argument.
In that earlier article, I gave examples to show that when subjected to ‘serious critical scrutiny’, statements or claims that seem to be ethically benign or have no troubling implications can sometimes be revealed to be incoherent, ill-thought out, one-dimensional, to lead to unacceptable risks if used unreflectively as a guide for action, and so on. What I find most concerning in the example I’ve given here is that it did not even take ‘serious’ critical scrutiny to show why the quote is nonsensical and entails potentially dangerous conclusions. The short version of the explanation demonstrates in a matter of sentences, having subjected the quote to only the most rudimentary critical scrutiny, why it should not be taken seriously.
This, in turn, is concerning, because judging by the resounding approval for the post, many people found the quote plausible. Given that degree of approval, I infer that posts like this are taken as guidance or – for want of a better term – ‘thought-leadership’ that is reliable, trustworthy, coherent, and so on. On the basis of how many people appeared to find the quote plausible, I can only assume that this is what passes for wisdom for many in our industry. But if this is what passes for wisdom, and thus, presumably, something sufficiently trustworthy as a basis for forming conclusions about what you ought to do, then at IGS we think we have a responsibility to provide a better alternative to our industry.
To give the poster their due, where some of the responses pointed out some of the shortcomings or inaccuracy of the quote, they acknowledged this to some extent. But this too is problematic. If they recognise the quote’s shortcomings and inaccuracies, then it ought not to have been posted; if they didn’t recognise the quote’s shortcomings and inaccuracies, then they either did not or were not capable of subjecting it to the requisite critical scrutiny. Given that, as I have tried to show here, it only required very limited critical scrutiny to show why the quote is specious and not to be trusted, it worries me that they either did not or could not do even this before deciding whether or not to post it.
Deciding what to do or to think on the basis of a quote like this is extremely risky. People’s data matters morally because people matter morally. And my claim that people matter morally is, I’m sure we would all agree, not ‘just’ an ‘opinion’ needing to be backed up by data before it can be taken seriously and regarded as true.
At IGS, our data ethics service is offered by ex-academics with doctorates in philosophy, theoretical and applied ethics. Of course, it’s not that having a PhD necessarily makes someone wise, any more than, as I said in the previous article, having expertise in ethics by virtue of having acquired the requisite body of formal philosophical knowledge and analytical skills, makes them a more perfect moral agent in their own lives. Nevertheless, if you find the analysis in this article plausible in terms of debunking the quote in question, I hope it will demonstrate what difference that body of knowledge and range of skills make for the clarity and quality of the conclusions that can be drawn.
If you want data ethics advice of the highest quality, advice which you can be sure is reliable and grounded in appropriately rigorous critical analysis, you should hire us rather than someone who does not have that formal knowledge or skill available to deploy. If you want to make sure that the data ethics advice you are getting is credible, and optimally likely to help you protect, rather than undermine, the interests of both your organisation and the individuals of whose data you are custodians, then IGS is available to help you.