Mapping out the entire world of objects

ImageNet is a large crowd-sourced database of coded images, widely used for machine learning. This database can be traced to an idea articulated by Fei-Fei Li in 2006: We’re going to map out the entire world of objects. In a blogpost on the Limitations of Machine Learning, I described this idea as naive optimism.

Such datasets raise both ethical and epistemological issues. One of the ethical problems thrown up by these image databases is that objects are sometimes also subjects. Bodies and body parts are depicted (often without consent) and labelled (sometimes offensively); people are objectified; and the objectification embedded in these datasets are then passed on to the algorithms that use them and learn from them. Crawford and Paglen argue convincingly that categorizing and classifying people is not just a technical process but a political act. And thanks to some great detective work by Vinay Prabhu and Abeba Birhane, MIT has withdrawn Tiny Images, another large image dataset widely used for machine learning.

But in this post, I'm going to focus on the epistemological and metaphysical issues - what constitutes the world, and how can we know about it. Li is quoted as saying Data will redefine how we think about models. The reverse should also be true, as I explain in my blogpost on the Co-Production of Data and Knowledge.

What exactly is meant by the phrase the entire world of objects and what would mapping this world really entail? Although I don't believe that philosophy is either necessary or sufficient to correct all of the patterns of sloppy thinking by computer scientists, even a casual reading of Wittgenstein, Quine and other 20th century philosophers might prompt people to question some simplistic assumptions of the relationships between Word and Object underpinning these projects. According to Donna Haraway, what counts as an object is precisely what world history turns out to be about.

The first problem with these image datasets is the assumption that images can be labelled according to the objects that are depicted in them. But as Prabhu and Birhane note, real-world images often contain multiple objects. Crawford and Paglen argue that images are laden with potential meanings, irresolvable questions, and contradictions and that ImageNet’s labels often compress and simplify images into deadpan banalities.

One photograph shows a dark-skinned toddler wearing tattered and dirty clothes and clutching a soot-stained doll. The child’s mouth is open. The image is completely devoid of context. Who is this child? Where are they? The photograph is simply labeled toy. Crawford and Paglen

Implicit in the labelling of this photograph is some kind of ontological precedence - that the doll is more significant than the child. As for the emotional and physical state of the child, ImageNet doesn't seem to regard these states as objects at all. (There are other image databases that do attempt to code emotions - see my post on Affective Computing.)

Given that much of the Internet is funded by companies that want to sell us things, it would not be surprising if there is an ontological bias towards things that can be sold. (This is what the word everything means in the Everything Store.) So that might explain why ImageNet chooses to focus on the doll rather than the child. But similar images are also used to sell washing powder. Thus the commercially relevant label might equally have been dirt.

But not only do concepts themselves (such as toys and dirt) vary between different discourses and cultures (as explored by anthropologists such as Mary Douglas), the ontological precedence between concepts may vary. People from a different culture, or with a different mindset, will jump to different conclusions as to what is the main thing depicted in a given image.

The American philosopher W.V.O. Quine argued that translation was indeterminate. If a rabbit runs past, and a speaker of an unknown language, Arunta, utters the word gavegai, we might guess that this word in Arunta corresponds to the word rabbit in English. But there are countless other things that the Arunta speaker might have been referring to. And although over time we may be able to eliminate some of these possibilities, we can never be sure we have correctly interpreted the meaning of the word gavegai. Quine called this the inscrutability of reference. Similar indeterminacy would seem to apply to our collection of images.

The second problem has to do with the nature of classification. I have talked about this in previous posts - for example on Algorithms and Governmentality - so I won't repeat all that here.

Instead, I want to jump to the third and final problem, arising from the phrase the entire world of objects - what does this really mean? How many objects are there in the entire world, and is it even a finite number? We can't count objects unless we can agree what counts as an object. What are the implications of what is included in everything and what is not included?

I occasionally run professional workshops in data modelling. One of the exercises I use is to display a photograph and ask the students to model all the objects they can see in the picture. Students who are new to modelling can always produce a simple model, while more advanced students can produce much more sophisticated models. There doesn't seem to be any limit to how many objects people can see in my picture.

ImageNet boasts 14 million images, but that doesn't seem a particularly large number from a big data perspective. For example, I guess there must be around a billion dogs in the world - so how many words and images do you need to represent a billion dogs?

Bruhl found some languages full of detail
Words that half mimic action; but
generalization is beyond them, a white dog is
not, let us say, a dog like a black dog.
Pound, Cantos XXVIII

---

Kate Crawford and Trevor Paglen, Excavating AI: The Politics of Images in Machine Learning Training Sets (19 September 2019)

Mary Douglas, Purity and Danger (1966)

Dave Gershgorn, The data that transformed AI research—and possibly the world (Quartz, 26 July 2017)

Donna Haraway, Situated Knowledges (Feminist Studies 14/3, 1988) pp 575-99

Vinay Uday Prabhu and Abeba Birhane, Large Image Datasets: A pyrrhic win for computervision? (Preprint, 1 July 2020)

Katyanna Quach, MIT apologizes, permanently pulls offline huge dataset that taught AI systems to use racist, misogynistic slurs Top uni takes action after El Reg highlights concerns by academics (The Register, 1 July 2020) 

W.V.O. Quine, Word and Object (MIT Press, 1960)

Stanford Encyclopedia of Philosophy: Feminist Perspectives on Objectification, Quine on the Indeterminacy of Translation


Related posts:  Co-Production of Data and Knowledge (November 2012), Have you got big data in your underwear (December 2014), Affective Computing (March 2019), Algorithms and Governmentality (July 2019), Limitations of Machine Learning (July 2020)

Naive Epistemology

One of the things I learned from studying maths and philosophy is an appreciation of what things follow from what other things. Identifying and understanding what assumptions are implicit in a given argument, what axioms required to establish a given proof.

So when I see or hear something that I disagree with, I feel the need to trace where the disagreement comes from - is there a difference in fact or value or something else? Am I missing some critical piece of knowledge or understanding, that might lead me to change my mind? And if I want to correct someone's error, is there some piece of knowledge or understanding that I can give them, that will bring them around to my way of thinking?

(By the way, this skill would seem important for teachers. If a child struggles with simple arithmetic, exactly which step in the process has the child failed to grasp? However, teachers don't always have time to do this.)

There is also an idea of the economy of argument. What is the minimum amount of knowledge or understanding that is needed in this context, and how can I avoid complicating the argument by bringing in a lot of other material that may be fascinating but not strictly relevant. (I acknowledge that I don't always follow this principle myself.) And when I'm wrong about something, how can other people help me see this without requiring me to wade through far more material than I have time for.

There was a thread on Twitter recently, prompted by some weak thinking by a certain computer scientist. @jennaburrell noted that computer science has never been very strong on epistemology – either recognizing that it implicitly has one, that there might be any other, or interrogating its weaknesses as a way of understanding the world.

Some people suggested that the solution involves philosophy.

People in CS and machine learning have been haphazardly trying to reinvent epistemology while universities make cuts to philosophy departments. Instead of getting more STEM majors we might be better off if we figured out how to send more funding to the humanities.

— 🦄 Dietrich Epp 📡 (@DietrichEpp) June 24, 2020

I completely agree with Dietrich about the value of philosophy and other humanities in general. However, I felt it was overkill for addressing the specific weaknesses identified by Professor Burrell, as her argument against this particular fallacy didn't seem to require any non-STEM knowledge or understanding.

While I sympathize with this sentiment, you don't always need a philosophy degree to spot incoherent CS thinking/epistemology. You just need to stay awake in your statistics class.

— Richard Veryard (@richardveryard) June 24, 2020

Of course, statistics is not the whole answer; but then neither is philosophy. I mentioned statistics as an example of a STEM discipline in which students should have the opportunity to unlearn naive epistemology; but of course any proper scientific discipline should include some understanding of scientific method. Although computing often calls itself a science, it is largely an engineering discipline; if you use the word methodology with computer people, they usually think you are talking about design methods. Social scientists (I believe Professor Burrell's PhD is in sociology) tend to have a much better understanding of research methodology.

And of course, it's not just epistemology but also ethics.

Nothing scares me as much as seeing naive engineers with no knowledge of structural injustice, pervasive power asymmetries, or conservative and racist history of the field of AI, being endowed with the power to make tech that infiltrates the social sphere.

— Abeba Birhane (@Abebab) June 29, 2020

One of the problems with professional philosophy is that it can be quite compartmentalized. There are philosophers who promote themselves as experts on technology ethics, but their published papers don't reference any recent literature on the philosophy of science and technology, or reveal any deep understanding of the challenges faced by scientists and engineers.

So although there is undoubtedly good reasons for broader education in both directions, I'm sceptical about expecting clever people in one discipline to acquire a small but dangerous amount of expertise in some other discipline. I'm much more interested in promoting dialogue between disciplines. In his tribute to Steve Jobs, @jonahlehrer called this Consilience.

What set all of Steve Jobs’s companies apart ... was an insistence that computer scientists must work together with artists and designers—that the best ideas emerge from the intersection of technology and the humanities.

The final word should go to @abebab

In contrast to an underspecified data for good model, embracing pluralism offers a way to work toward a model of data for co-liberation. This means transferring knowledge from experts to communities and explicitly cultivating community solidarity in data work.

— Abeba Birhane (@Abebab) July 4, 2020

---

Jonah Lehrer, Steve Jobs: “Technology Alone Is Not Enough” (New Yorker, 7 October 2011)

Related posts:  From Convenience to Consilience - “Technology Alone Is Not Enough"  (October 2011), The Habitual Vice of Epistemology (June 2019), Limitations of Machine Learning (July 2020), Mapping out the entire world of objects (July 2020)