Data Ethics Club: AI Snake Oil Book Club#
What’s this?
This is summary of the discussions held in Data Ethics book club summer 2025, where we spoke and wrote about AI Snake Oil by Arvind Narayanan and Sayash Kapoor. The summary was written by Jessica Woodgate, who tried to synthesise everyone’s contributions to this document and the discussion. “We” = “someone at Data Ethics Club”. Huw Day helped with the final edit.
Chapter 1 – Introduction#
Chapter Summary#
The term “AI” has been appropriated for a multitude of different technologies and applications, confusing discussions about its strengths and limitations, and misleading people about its capabilities. The introduction begins by distinguishing between generative AI, which produces various types of content such as text or images, and predictive AI, which produces outputs forecasting the future to inform decision-making in the present.
The waters surrounding the definition of AI are muddied; influenced by historical usage, marketing, and more. Some problems AI attempts to address, like vacuuming, are solvable; others, like predicting whether someone will commit a crime, are not. Some applications were once thought difficult, but AI has proven to be very good at, such as facial recognition. However, even if the mechanism is good, the tool can fail in practice, e.g., if the data is noisy, biased, or abused for malicious intent.
Sold as an accurate and capable technology, predictive AI has been widely adopted. However, there are significant issues with the premises predictive AI is based on, as it attempts to translate and reduce high dimensional phenomena to computationally feasible outputs. In practice, there is missing data, bias, and participants attempting to game the system. Generative AI, the chapter argues, holds more promise and yet also runs into significant issues. Factual inaccuracies in outputs are common, leading to rampant misuse such as error-filled news articles or AI-generated books.
Misinformation, misunderstandings, and mythology about AI are fed by a combination of researchers, companies, and the media. In research, the buzzier the research topic, the worse the quality of research tends to be. Companies feed off of hype, seeking to maximise profits whilst rarely being transparent about the accuracy of their products. Crumbling revenue in the media combined with access journalism, where outlets rely on good relationships with companies to be able to interview subjects, further propagate hype narratives dictated by AI companies.
The limitations, hype, and misconceptions surrounding AI leads the chapter to analogise AI as a “snake oil” product: a miracle cure advertised under false pretences. The book aims to identify AI snake oil – AI that does not and cannot work.
Discussion Summary#
What do “easy” and “hard” mean in the context of AI? Does it refer to computational requirements, or the human effort needed to build AI to perform a task, or something else? And what does easy/hard for people mean?#
The idea of “easy” or “hard” problems for AI weren’t specifically defined in the chapter, but it did give some examples of problems previously thought to be difficult for AI which the technology is now very capable of solving, such as image classification. Spellcheck is another example of an application previously thought to be hard but is today so embedded in our everyday lives that we might not even think of it as AI.
Sometimes what is “easy” or “hard” for AI is contrary to what is easy or hard for humans – a phenomenon coined Moravec’s paradox. Moravec’s paradox is the observation that reasoning (which is difficult for humans) requires little computation, but sensory and perception skills (which are easy for humans) are highly computationally expensive. Computational complexity is a well-studied field examining the amount of resources required to run an algorithm. We wondered if easy and hard in relation to AI are somehow related to how difficult it is for AI to produce an accurate output.
Tasks that are easy for humans are those that we can do quickly and without much concentration. Pinpointing what these tasks are is tricky; we don’t know what we know. People aren’t good at breaking tasks down into smaller levels than humans are typically used to or deducing which of these tasks are easy. The difficulty with breaking down tasks is why some people find programming really tricky. What counts as easy might depend on the sample population, e.g., a roomful of mechanics will find it much easier to replace an engine than a roomful of mathematicians.
Discerning what is easy or hard for AI is not straightforward, as it depends on the lens you examine the tool through and the context it is situated within. LLMs now seem to be very good at predicting text, giving the impression that the task is easy. However, an extraordinary amount of resources goes into training an LLM, suggesting it is actually a hard problem. Facial recognition is accurate under the right conditions, but there are many drawbacks for how it is used, making it hard to delineate appropriate use cases. Perhaps easy and hard aren’t the right terms; we should instead be asking how much context is needed to solve the problem, and how hard the execution is.
The underlying message of the introduction is that certain tasks are made easier or harder from the application of certain AI systems, and the way we sell those systems affects how they are used. For example, linear regression is really good at some statistical modelling problems. However, if we sold it as a silver bullet to solve any problem, it would be (and, as the chapter demonstrates, has proven to be) rubbish.
The barriers to entry for AI have been drastically lowered over recent years. On the deployment side, LLMs are now very easy to set up on a personal laptop. On the user side, generative AI interfaces are widely accessible. It is actually starting to require more effort not to use LLMs in some settings such as searching, where search engines are displaying an “AI overview” before presenting the results. Tasks that LLMs previously found difficult are slowly getting easier and easier, making their deployment and user friendliness increasingly accessible. For example, maintaining context over time was something that LLMs previously found very difficult.
However, the context window, which dictates how much input the LLM can consider when calculating its output, is slowly getting larger. A larger context window entails that an LLM can consider more information in its answers. In addition, ChatGPT now incorporates some “memory” functionality, where information from previous conversations is maintained. However, sometimes relevant context goes beyond conversations, such as body language. This raises important questions about what kind and how much information goes into the model, e.g. whether it should consider current affairs or interpersonal relationships.
More data and computational power could mean that AI gets better, but this is not always true. It may depend on the use case and type of AI. For example, given enough training, AI can beat the best human chess player, but how long will it be before AI can make a cup of tea to your preferences, and is more computational power enough to solve this? If the task is something that the user doesn’t know much about, or is difficult to express, the usefulness of AI goes down. Training and human feedback is required on both sides. It is difficult to know where the improvements will stop or how the evolution of AI will unfold, especially considering the transformative effect of previous industrial revolutions.
Based on your definitions of these terms, pick a variety of tasks and try to place them on a 2-dimensional spectrum where the axes represent people’s and computers’ ease of performing the task. What sort of relationship do you see?#
Difficulty |
Computer |
Human |
Both |
---|---|---|---|
EASY |
Coding |
Sense of right and wrong |
Image classification |
Pattern recognition in big sets of data |
Ironing |
Speech to text |
|
Analytical decision-making (depending on algorithm and input data) |
|||
Chess |
|||
Remembering lots of information |
|||
Recall (for some computers but difficult for LLMs) |
|||
Writing a poem in a certain style |
|||
Sucking in created content |
|||
HARD |
Ironing |
Coding |
Creativity |
Ethical decisions / judicial |
Content creation |
Reading emotions |
|
Holding/understanding context |
|||
Sarcasm/jokes |
|||
Moving around in a new environment |
|||
Certainty of answers |
|||
Language manipulation |
|||
The text gives many examples of AI that quietly work well, like spellcheck. Can you think of other examples? What do you think are examples of tasks that AI can’t yet perform reliably but one day will, without raising ethical concerns or leading to societal disruption?#
Applications of AI that we could envision being less ethically concerning include scientific pursuits, such as animal conservation for tracking animals, biodiversity monitoring, birdsong recognition, weather prediction, pollution analysis, or finding biomarkers for diseases. These use cases do not directly dictate outcomes for people’s lives and can be verified by complementary scientific methods.
With respect to AI being used in everyday life, AI should act as a facilitator for human flourishing, rather than a supplement. We would much rather have AI do our laundry whilst we make art, rather than have AI generate “art” whilst we do laundry. If ChatGPT was reliably accurate, we could imagine it being useful as a learning tool e.g. by quizzing students on their homework. Tools like Grammarly are great in certain contexts, because native English speakers can be harsh when assessing people’s writing. When people don’t have English as a first language, grammar correcting tools can help them to be sure they are saying the right things.
There are many AI tools we use every day without noticing, but that does not mean that they are working well. Spellcheck is integrated in many applications, however, we do not think it works well. If you write in more than one language, it falls apart and doesn’t understand what context it’s in. We tend to have counterintuitive expectations of the abilities of these tools; we recently encountered someone praising ChatGPT for being helpful with writing code about 60% of the time. If someone was asking us for help, and we were getting it wrong 40% of the time, we would not expect them to ask us for help again.
We would be hard pushed to find an AI tool that won’t raise ethical concerns or cause some disruption. If something could replace a job, it could cause disruption. There are lots of areas in which people don’t care how good the labour is, they just want an output – even if it’s inaccurate generative AI nonsense. Even seemingly innocuous applications like spellcheck or Grammarly have rippling implications, e.g. for students. We have experienced spellcheck changing our answers leading to a quiz fail. If students can use something to do their work, it affects the work of teachers.
Once a tool can provide an answer it would take a human a while to come up with, it is easy to slip into cognitive offloading. We’ve noticed that tools like Grammarly are now pitched at native English speakers, potentially discouraging them from improving their own grammatical abilities. Lots of native English speakers who should already have these skills will defer to grammar checking tools, assuming that the tools will be better. People have authority bias, doubting their own knowledge because the tool has told them something different.
The disruptive potential of AI may depend on the application and sensitivity of the data. It is important to take into account the whole pipeline, considering not just what the tool does but the energy it consumes and where the data used to train it comes from.
What change would you like to see on the basis of this piece? Who has the power to make that change?#
Careful consideration of our attitudes towards AI is crucial in shaping the role it plays in our lives. It is difficult to maintain healthy scepticism of AI in the face of overwhelming hype. In general, the chapter sets a good tone, balancing criticisms with an awareness of potential benefits. Some of us have come to the book with a pro-AI attitude, looking to engage with it as a challenging view. Predictive and generative AI have beneficial use cases, e.g., generative AI can help people get started with a prototype for an idea even if they don’t know how to code. Others among us are looking to ask whether these benefits are worth the costs.
Some of us are becoming increasingly hardline anti-AI, finding that the chapter doesn’t go in as hard as it could (or maybe should) against AI. The chapter focuses, rightly, more on predictive AI, but it could have been more critical of generative AI. Fundamentally, many of us believe generative AI cannot do anything new. The outputs might be useful but are fundamentally unreliable in the sense that there is no validation of their correctness. The chapter talks about how “facial recognition works well enough to be harmful, and badly enough to be harmful” – this probably applies to LLMs as well.
To highlight the problems with applying AI, we considered the difference between using facial recognition to block someone from attending an event, and having a bouncer at the door with a list of people not allowed in. The difference may lie with accountability; you can argue with a person or ask them to take action, but with an automated system you will get nowhere. A person might lose their job if they do it incorrectly; if a system makes a mistake, it does not pay for the consequences, and the costs of rollback are often too high to warrant a system change. A human security guard can’t be scaled up to surveil 100,000 people at once and get 5% wrong with no recourse.
Distinguishing between the technology itself and the people behind it is increasingly tricky, as the people behind it are increasingly distanced and invisibilised from the tool itself.
Chapter 2 - How predictive AI goes wrong#
Chapter Summary#
The chapter discusses the precedence of companies making strong claims made about the utility of automated decision-making systems using predictive AI in order to sell them. Models are claimed to be accurate, efficient (requiring little to no input from humans), and fair. One of the appeals of predictive AI is that it can reuse datasets that have already been collected for other purposes, such as record keeping or bureaucratic tasks. Models are also appealing through the promise of attempting to predict the future, as people struggle to deal with uncertainty or randomness and seek methods that enable control. These systems are used to allocate resources, provide or withhold opportunities, and predict peoples’ future behaviour.
Yet, whilst a model may make a decision from an input without human involvement, human decisions still exist throughout the pipeline. Humans dictate the design of the model, the data that is collected, and the methods of deployment. It cannot be guaranteed that these decisions are unbiased or fair. Models tend to make predictions that are correct according to the way they were designed, but issues can arise in deployment. Important data may be missed or misunderstood, the system may change, or people may employ gaming strategies.
Once in place, systems are extremely hard to reverse, and people are unable to challenge decisions. Decision subjects are frequently unaware that they are being evaluated by AI, yet the decisions made can have life changing implications and mistakes are hard to fix. Even if human oversight is in place, it is often inadequate. Costs of flawed AI disproportionately harm groups that have been systematically excluded and disadvantaged in the past. Instead of treating people as fixed and their outcomes as predetermined, the chapter argues that we need to accept the inherent randomness and uncertainty in life. Institutions should be built with an appreciation that the past does not predict the future.
Discussion Summary#
Predictive models make “common sense” mistakes that people would catch, like predicting that patients with asthma have a lower risk of developing complications from pneumonia, as discussed in the chapter. What, if anything, can be done to integrate common-sense error checking into predictive AI?#
The idea of being able to model common sense is difficult to accept, as it is a changeable and contested concept. Common sense mistakes happen with or without AI, thus perhaps automated decisions are not necessarily worse than those made by humans. Models are a product of how they are trained; a model is only as good as the modeller. If what goes into a model is rubbish, you can expect that what comes out will be rubbish too: “garbage in, garbage out”. Algorithms are trained on data that reflects biases and prejudices held in society, such as racial inequities. We’ve found evidence of this in research we’ve conducted investigating the ability of a neural network to look for differences in skin tones in images of skin cancer lesions, finding that the network performed poorly. In the case of detecting skin cancer lesions, the implications this has for detection rates for varying racial groups is troubling.
In addition to the data that goes into a model, the design of the model itself, such as the variables included, influences the predictions the model makes. It is crucial and difficult to select variables that are meaningful with respect to the question being asked. Identifying causal inference is hard, and not every relationship you see in data is a causal relationship. Many AI models today have far too many variables to exhaustively check; ChatGPT-4 is estimated to have 1.8 trillion parameters.
Considering the inherent difficulties with building common sense into a system, perhaps the best approach is to focus on employing common sense in the application of AI. To cultivate common sense application, the general perception of computational systems as infallible will need to change. People tend to treat AI systems as more knowledgeable than humans, thinking that computers don’t make mistakes and over-crediting their veracity. Humans transfer social trust onto machines, projecting an idea of “expertise” as systems appear to give confident answers on the base of some hidden knowledge. There may to be some correlation between size and scepticism; there tends to be more apprehension around small studies compared to larger ones, and similarly we are more likely to mistrust a smaller model compared to a massive one. However, just because something is big doesn’t mean it is right.
To properly evaluate a model, the broader context must be considered. Caution does not tend to be incorporated into models yet plays an important role in the way humans make and apply decisions. It is also important to distinguish between different sorts of problems, asking if the model itself is bad, or if the application of it is inappropriate. One should consider how the data was collected and what the shape of the problem looks like. For example, in predicting blood pressure, most people don’t have issues until they are older, creating a “U” shape in the model. Therefore, application of a linear model to this problem will not fit.
Incorporating human checks and oversight throughout the AI pipeline could help mitigate unwanted side effects. To avoid responsibility “washing”, wherein companies claim to have oversight without implementing proper procedures, oversight will need to be carefully defined including the likely failure modes to detect. Systems will need to be explainable so that overseers can understand what they are looking at. Perhaps legislation is one solution to require companies to explain the weaknesses of their models and highlight where oversight should be more closely applied, similar to how companies are required to list side effects for medication.
Statistical modelling is generally thought of as reliable, requiring deliberate choices which are made explicit. In machine learning (ML) contexts, sometimes those choices are not as transparent. There tends to be a lot of opacity in decision-making that goes into the development and application of models. To understand how black box models are working, the criteria that the model uses to make decisions should be pulled out. Sometimes, machines are making connections we would not know to make, or inferences that are not what they seem. For example, in healthcare applications, image classifiers have been found to be taking into account other elements of the image in their decision such as the use of rulers in an image.
If we are incapable of understanding a model, such as a cancer screener, we wondered whether or not we should be using it. It may not be essential to understand the mechanics of all the technology we use. Most of us use black box applications every day without question, such as computers, central heating, or aspirin. Sometimes more transparency might not be appealing if revealing the criteria for decisions facilitates applicants gaming the system. In these settings there might be specific audiences for whom the system should be transparent to, e.g. the system should be transparent to hiring managers, but not to applicants. On the other hand, perhaps applicants would like to know the criteria they are being judged by, and withholding this information may be unfair.
Think about a few ways people “game” decision-making systems in their day-to-day life. What are ways in which it is possible to game predictive AI systems but not human-led decision making systems? Would the types of gaming you identify work with automated decision-making systems that do not use AI?#
There is an increasing sense for upcoming generations that to stay ahead they need to game systems, and those who don’t realise systems are gameable are put at a disadvantage. For example, understanding how to answer personality tests prevents job applicants from minority groups being screened out. Part of the hacker mindset entails you are smart if you are able to hack things.
Even in non-AI contexts, there are plenty of examples of systems being gamed. People frequently game each other through manipulative tactics. Knowing what to say can get you the right hospital treatment or make you eligible for benefits in pre-screening processes. We wondered where the line is between tailoring appropriate communication to a particular audience and gaming a system. Perhaps there is some relation to whether one’s actions are working towards a particular desired outcome.
Settings in which people game predictive AI systems include search engine optimisation, contributing to the enshittification of the internet by filling websites with algorithmic buzzwords; job applications; social media. Industry resumes tend to be shorter than academic resumes, incentivising the use of buzzwords and listing skills like “leadership”, “Java”, etc., that will bump the applicant up the list. To get your profile seen, we feel an increasing pressure to make our job applications and cover letters “LinkedIn friendly”. A funny side effect of this is that when someone says you are good at LinkedIn it feels like an insult by being seemingly inauthentic.
Buzzwords are also used to game funding or grant applications. We have seen examples of proposals suggesting a project will be using “AI” in its methods, where in reality it is not really an appropriate application of AI. Sometimes people want to use AI because it is trendy, whether or not it would actually solve the problem. Many people do not really understand what AI is, or the what the differences are between specific terms such as ML and AI (ML is a subfield of AI). Being clear about terminology, especially in the mainstream, is complicated by the hype cycle, which propagates uncertainty and sensationalist narratives.
The hype cycle encourages people to adopt trending methods even if they do not have sufficient background or training to understand how the methods work, which is detrimental to the quality of research. Bad research deteriorates public trust in science and scientific outcomes. Interdisciplinary collaboration is essential to better support researchers, and perhaps the importance of various disciplines should be discussed at lower levels of education to foster this. Psychology has suffered credibility crises, but this has led to stronger research practices such as hypothesis pre-registration and more publishing of negative outcomes in journals.
In which kinds of jobs are automated hiring tools predominantly used ? How does adoption vary by sector, income level, and seniority? What explains these differences?#
We weren’t sure which fields automated hiring tools are currently being used in, so instead discussed the fields such tools might be best used in. Hiring tools could be helpful for positions with a high ratio of applicants, as sifting through thousands of applications is a time consuming and repetitive task. Another appropriate application for automated tools may be entry level jobs in which group interviews are already commonplace, to help speed up the process.
What change would you like to see on the basis of this piece? Who has the power to make that change?#
People gaming automated systems may change the systems themselves, depending on how and what the algorithms are learning. For instance, hiring algorithms may learn to favour people who figured out how to game them and thereby further incentivise those behaviours. Gaming the system can push things towards homogeneity, which we have seen in other applications such as TikTok, where monetisation depends on the length of engagement and so videos tend towards a minimum length.
As AI is adopted by both hirers and applicants, a feedback loop is formed where LLMs are used to write and apply for jobs, other AI systems sift through the applications, and each side learns to game the other. Increasing dependence on LLMs will lead to more mistakes: LLMs are producing less and less accurate results, and are shown to repeatedly hallucinate and backtrack. It is important that society finds ways to resist slipping into homogeneity and error-strewn information as a consequence of LLM overuse.
Chapter 3 - Why can’t AI predict the future?#
Chapter Summary#
The sheer number of correlations that certain AI methods can identify in data, and the usefulness of those methods in identifying which correlations are important, has contributed to the popularisation of AI to predict the future. Despite the efficacy of AI in identifying correlations, the world is profoundly complicated, and phenomena are often compounded by a myriad of variables. The weather, which humans have been trying to predict for thousands of years, is a good example of how easily real-world complexity interferes with prediction methods. In the 1960s, Edward Norton Lorenz found that rounding the numbers of weather simulations to three decimal places instead of six gave vastly different results. This finding led to the coining of “the Butterfly Effect”: the observation that small errors in measurement (e.g. of temperature) lead to exponentially increasing errors later. The farther away the prediction, the larger the error.
There are two main paradigms for predicting the future: simulation, and ML. Simulation is based on the idea that the future evolution of a system can be predicted using the current state of the system and equations describing how the system changes over time based on the interactions between its components. Simulations have proven useful for some applications, like the weather, and terrible at others, such as social phenomena like modelling a whole city. ML on the other hand uses past data to learn underlying patterns and make predictions about the future, without fixed rules about how the future will play out given the past. Predictions and patterns can adapt and change over time, but the “rules” determining this are based on how the system behaved in the past. The chapter argues that simulation is more suitable for predicting things about collective or global outcomes, whereas ML is better suited to predict things about individuals. ML is appropriate in settings where there is a lot of data to train on, such as examples of spam or not spam emails, whereas simulation is appropriate where there is domain knowledge but not enough examples, such as food shortages.
The line between what is and isn’t feasible in prediction social phenomena is ill defined. It is too broad to state that we can’t predict any social phenomena, as some dynamics like traffic or how busy a store will be on a certain day can be predicted reasonably well. However, predicting a person’s future is hard because of a combination of the real-world utility of what can be done with the prediction, the moral legitimacy of the application, and the prediction’s irreducible error (error that won’t go away with more data and better models). Statistical or AI models used to predict life outcomes have repeatedly been shown to be little better than random, with complex models offering no substantial improvement above simple baseline models.
A problem in predicting the future is the difficulty in defining the accuracy of a prediction: the range of output that is considered accurate (e.g. for the weather, whether temperature is accurate within one degree or five degrees); the type of output (e.g. if it will reliably predict it will rain regardless of temperature); how the accuracy of one domain compares to another (e.g. weather patterns compared to sales patterns). To assess if a prediction task can be done well, there are qualitative criteria that can be examined, such as looking at what kinds of prediction tools people actually use, what can be done with a tool, relevant power dynamics and moral implications, and whether a prediction will improve with more data and better models.
It is not yet possible to conclude whether there are fundamental limits to predicting human lives, as has been proven in other domains such as planetary orbits or thermodynamics. We do not know what the irreducible error is for social prediction; it may be possible that predictability of life outcomes would be improved by the availability of more data. Yet, it seems likely that the irreducible error is high as chance events can drastically alter the path of a person’s life, either through large shocks or small (dis)advantages that compound over time. Patterns underlying social phenomena are not fixed and differ greatly based on context, time, and location. The sheer amount of data needed to predict life outcomes would be very high because social datasets have a lot of noise and the number of samples needed to create accurate models sharply increases as the samples become noisier. There are thus some limits to predicting the future that could be overcome with more and better data, whilst other limits seem intrinsic.
Discussion Summary#
Cumulative advantage implies a lot of success comes down to luck, which challenges a meritocratic world view. How do you feel about this?#
Putting too much emphasis on luck can be detrimental; we don’t want the takeaway to be that everything is luck so there is no point in trying. It’s still important to strive for things we think or know are impossible, as incremental progress can be made. Some of us thought that we should strive for more meritocracy because of this. We wondered if we would be able to predict success better if we lived in a more meritocratic society. Overemphasis on meritocracy may however run into the problem of predicting and thereby pre-empting success, promoting those who are predicted to be successful rather than those that have proven to be successful. One of the main criticisms of meritocracy is that it ends up supporting those who already have an unfair advantage due to societal power dynamics.
Cumulative advantage is something that we have observed in our day-to-day life such as by clicking more on the most downloaded songs and other demonstrations of herd behaviour. For music and movies there are other factors that intertwine with herd behaviour and may factor into cumulative advantage such as the effect of nostalgia and role of cultural background.
What change would you like to see on the basis of this piece? Who has the power to make that change?#
Limitations to ML methods entail that when approaching a problem, it is important to ask if ML is actually the appropriate tool to use, or if there is a faster and cheaper way. In the book, the examples of successful predictive models were interpretable mathematical models and the ML examples either failed or were worse than “simpler” models. We wondered if there are any examples of ML models that work, and work better than simpler models. The lack of examples of successful ML may be because ML is a (relatively) young field. ML may become increasingly better than simpler models as the field matures.
Chapter 4 - The Long Road to Generative AI#
Chapter Summary#
Generative AI comes behind a long history of computational innovations. From the development of the perceptron in the 1940s, based on a mathematical model of a neuron, to the boom and busts of neural networks over the latter half of the 20th century, eventually bringing us where we are today with massive neural networks that can generate diverse outputs from text, to images, to audio, and more. The chapter argues that generative AI can be hugely beneficial for a range of use cases from assisting knowledge workers, who “think for a living”, to improving accessibility such as for the partially sighted.
One application that deep neural networks have proven really effective in is image classification. Historically, labelling and classifying images required a marathon of human effort. Today, it is something that can be done in seconds using AI. Building on the success of neural networks, over recent years generative AI has received an escalating level of attention. There is a huge community of people working on the technology, contributing to its application in an ever increasing expanse of use cases and rapid evolution of the technology itself.
As we see time and time again, alongside excitement for a new AI technology comes a range of harms. The rapid uptake of a technology before it is properly understood and potential harms have been rigorously studied opens the door for destructive consequences. Generative AI has produced outputs that are racist and inappropriate, oversexualised or claiming to be sentient, declared sentient by prominent engineers, been used in real law cases to generate citations proven fake, associated with suicides when used as companion bots, and used for non-consensual porn deepfakes. Copyright issues have led to a number of legal cases currently passing through the courts concerning the appropriation of artists’ work to train models. Models do not just adapt the works they have been trained on to produce new outputs, but in some cases can almost exactly replicate instances of their training data, such as the Mona Lisa.
After training LLMs on copious amounts of data, additional fine tuning has led to highly convincing chatbots which are good at producing believable responses but lack real-world grounding and verification. On a practical level, chatbots are at their core statistical engines. They have incomplete internal representations of the world around them and thus lack true “understanding”. LLMs have a tendency to produce “bullshit”, which are outputs that attempt to persuade without regard for the truth. Researchers’ understanding of these internal representations is still rudimentary due to the lack of interpretability in what neural networks encode. We do know, however, that because LLMs learn from historical data, they will also learn any historical biases contained within that data. Attempts to mitigate these harms come in the form of more human annotation, which raises more issues with worker rights as the labour tends to be outsourced to poorer countries than where the company is located so that they can capitalise on low wages and less regulation.
Discussion Summary#
Various AI systems, such as the neural networks in the ImageNet competitions, are designed to automate something that is expensive to do on mass (e.g. labelling images) and where the process in doing so is not valuable to the humans doing it. What tasks do you see GenAI (both LLMs and image generators) being used for and how much are they being used for processes where there is value for humans performing it?#
Many tech leaders are loudly proclaiming the promise of generative AI to automate work as we witness a wave of mass firings and human workers being replaced by AI (explored in a previous Data Ethics Club). Generative AI proponents advocate that the technology will free humans up for meaningful work. The implication of these messages is that the work people are attempting to replace with AI, including some of our jobs or things that we do in our jobs, are not meaningful.
Defining what it really means for something to be valuable to humans is critical to ensuring that generative AI is used in ways that truly benefit society and yet seems to be often brushed aside. The current landscape of generative AI tools tends to implement them in obvious applications that the tools can speed up, such as writing emails, coding, or labelling images, but without delving deeper to ask questions about if these are the sorts of things that we actually should be using it for. If we want to use AI to label people in photographs, who is that valuable to, and why is it valuable? Many tools have been proposed as a solution for efficiency, without critically assessing whether there is more besides efficiency that needs consideration.
Perhaps generative AI should be used for doing tasks that nobody wants to do and that we would rather a machine do than a human. If people do not understand or appreciate the effort required for a task, perhaps it is not worthwhile putting a lot of time into it, and we should use whatever means available to us to expediate it. An application we have seen for computer vision that seems intrinsically valuable to us is medical imaging, where previously we worked on teams with many people spending hours labelling images. Many of those labelling images found the task to be boring, and did not leave them room to grow through their jobs. Today, AI demonstrates high utility for completing these sorts of tasks quick and at high volume, leading to significant scientific contributions such as creating a complete map of the neurons in a fly’s brain and tumour identification.
Other touted applications of AI may not be as appealing as they seem when we delve deeper into them. We have conducted work anonymising text data about medical conditions, which is a task that could arguably be automated by AI with advantages for efficiency. However, whilst we found the experience to be quite heartbreaking it also really motivated us to work on the project and helped us appreciate the stories within the data. We wouldn’t want to lose this connection with stories; if you just deal with the summary level you do not get same the richness as reading the actual interviews. These experiences highlight how there is value that lies in painstaking qualitative data analysis. Some of us think that a hybrid approach could work, where generative AI can be used for qualitative data analysis but still needs to be checked.
Concerns with the trustworthiness of generative AI gives us reason to be cautious in our use of it and question how valuable it is for day-to-day use. A lot of the marketing surrounding generative AI tools centres around automating admin tasks, but we were unsure if we would trust an AI agent to perform these tasks correctly and in privacy preserving ways. It may take a significant amount of effort and careful wording to prompt an AI in such a way that it completes your requests truly according to your specifications.
There is an important distinction between the value of the product and the value of the process. Something that is valuable to do intrinsically is not the same as something that is valuable by its outcome. Sometimes the process itself is valuable, such as in making art. Often artists are admired for the effort that goes into creating art. The hype narrative centring on using AI to automate “boring” tasks is a narrow view that just focusses on speeding up the process to get to the product faster, rather than thinking holistically about the whole pipeline. Using AI to write code has become very popular, with many users just wanting something that runs without error even if it does not evaluate the right answer. A lot of students and staff we encounter use LLMs like they’re a search engine and expect the output to always be correct. Outsourcing the search strategy to an LLM shortcuts what could be a valuable process of evaluating sources and their utility. Many people don’t know this process exists, let alone recognise its value. Searching and analysing sources is a critical skill taught by librarians that should be more highly recognised.
A question that seems to have been skimmed over by the AI hype narrative is whether we really want everything to be maximally efficient. There is an appeal to doing mundane work and many of us want some parts of our jobs to be boring. We can’t be turned on and 100% high functioning all the time, and there are other things that matter to us in life aside from work which we want to save energy for. Getting bored gives space for creativity to flourish, to get curious, and find meaning in life. Monotony also provides opportunities to socialise. We have had jobs doing repetitive factory work but found it to be very social as we chatted with the people around us while we worked. Across many sectors of society more and more distance is being created between people, disincentivising interaction with others. In our offices people used to spend more time standing around the water fountain chatting. Shopping used to involve talking to someone at the checkout and someone helping you pack your bags but now that checkouts are automated it is harder and harder to talk to an actual live person. Sometimes when interacting with a company you don’t even know if you are interacting with a human or a bot. We’ve had interactions that seemed like we were talking to a bot but turned out to actually be people in India.
Throughout our lifetimes, we have seen many fads come and go. For example, there has been excitement about robots performing surgery since the 1970s yet we are still operated on primarily by humans. The extent of the hype and disruption around AI, however, makes us wonder if it is a craze that will not tail off. Hype cycles are problematic (explored in a previous Data Ethics Club). One of the many issues is that hype generates over trust in machines leading to technologies not being properly checked. Consequences of improper checking can be life changing for people, such as the Post Office scandal in the UK where hundreds of people were prosecuted because of faulty accounting software. Rather than investigate and fix the problems with the software itself, the Post Office blamed branch operators for financial discrepancies.
It is also important to acknowledge that it is possible to automate systems without using AI, such as using statistical models for medical imaging versus AI and ML techniques. For medical imaging, ML methods do not address causation which is central to identifying diseases. If we are using ML for these sorts of applications, then proper checks and balances are indispensable.
In automating processes and evaluating models we must be clear about how we verify what is considered good or bad. This is easier in some domains than others. For example, if we know what a good scientific image diagram looks like, where there are well-defined rules and criteria, we can evaluate whether an AI has done a good job in producing one. It is harder to evaluate whether an AI summary of a long article is good if we haven’t read the original article or know the domain in depth. Some of us wondered if perhaps no-one without a PhD should be using certain AI tools, and outputs should be checked by multiple people with PhDs. Similar to how new drugs are evaluated in pharmaceuticals, all consequences and side effects should be considered. In AI settings, this includes bias and institutional and historical power dynamics; as ML relies on and continues to propagate the past, issues of the past will percolate into the future if unmitigated.
Our discussions make some of us yearn for a simple formula to work out if something is a good use case for AI. Important questions to ask would include how boring the work is, how verifiable the results are, how risky the setting is, whether there is enough data to make an accurate prediction, and whether it removes the accountability for a decision from a human. These are all essential considerations but must be held alongside the knowledge that there is often not a simple answer and ethics is grey.
Generative AI is built using the creative output of journalists, writers, photographers, artists, and others — generally without consent, credit, or compensation. Discuss the ethics of this practice. How can those who want to change the system go about doing so? Can the market solve the problem, such as through licensing agreements between publishers and AI companies? What about copyright law — either interpreting existing law or by updating it? What other policy interventions might be helpful?#
Our legal systems are currently in the midst of working out how to address copyright in light of the new AI landscape. There are lessons from the past that could be useful here; the power dynamics of the tech industry with a handful of titans dominating the market draws more than a few resemblances to railroad companies in the US towards the end of the 1800s. Railroads were deemed to be monopolies, leading to the introduction of antitrust leglislation. Antitrust legislation banned monopolies that placed unreasonable restrictions on trade, which enabled the federal government to break up big companies. Whether antitrust legislation really worked is debatable, but perhaps some similarly disruptive legislation is long overdue for big tech companies (this podcast gives a perspective on how we got to where we are today by Cory Doctorow, who coined the term “Enshittification”, explored in a previous Data Ethics Club).
A case regarding Meta’s use of millions of books to train AI systems has recently been deemed fair use. Our reading of the outcome is that using art to train AI was deemed fair use, but pirating art to train AI was not. District judge Vince Chhabria noted in his decision that the ruling does not entail the use of copyrighted materials to train LLMs is lawful, but that the plaintiffs made the wrong arguments. Chhabria suggested that a potentially winning argument in the case would be that Meta had caused market dilution by flooding the market with endless amounts of content thereby causing damage to copyright holders.
Companies profiting off of artists’ work without their consent seems intrinsically bad to us, but it might be too late to turn back the clock on how wide we allow the reach of big companies to be. There is an inequality between the rich and everyone else in how copyright is enforced. Ordinary people, including a 12 year old girl, have been prosecuted for sharing MP3 tracks yet today generative AI companies are seemingly getting off scot-free for using millions of songs, images, books, and more – and making money from it.
One imagination for after the court cases have been worked out is that the market for artists will look more similar to something like Spotify or iTunes, where artists are compensated a little bit but not much. This idea is believable but perhaps not desirable; the monetisation structure of these platforms means artists have to be fairly large to earn real money and it is almost impossible for smaller artists to make a living.
To redress power imbalances, perhaps companies that profit off of AI trained on other people’s data or creative works should be required to repay the public, as explored in the idea of data labour. Some of us think that either we acknowledge that data or creative works were given freely and consensually, or they were stolen. While we have been taking photographs and posting them on Facebook over the last 20 years, we have been doing so to share them with our friends; we did not think that the photos could be taken and used for other purposes. The publicness of the internet is a consequence of the mechanisms by which we can share stuff. Yet, artists posting their portfolios online are doing so for employment opportunities, not to condone the misuse of their work for others to make money. We do not think other people should profit from our work, which is why some of us who write poetry never post it online. We want our work to be out there and free for people to appreciate, but we don’t want it to be appropriated and fed into AI. Taking our work to train AI feels like our creativity being stolen.
It should be much easier to decline your data being given away and used to train AI models in order to protect peoples’ autonomy. Currently, it is nearly impossible to not use AI products such as on our phones or browsers, and it is not clear how those AI products handle our data. Similar issues crop up with facial recognition compared to DNA swabs. In the US, police need a warrant to collect DNA swabs yet facial recognition has much lower restrictions and is increasingly ubiquitous. With many of the big players in tech being US based, the perspective on privacy is highly US centric. However, views on privacy and expectations of public observation differ across the world. Germany restricted Google Streetview until 2023 as people weren’t comfortable with pictures of their houses being online. The presumption that if something is on the internet, it’s fair game, is flawed and does not align with global privacy preferences. Despite this, we must acknowledge that this view is the default and do our best to raise the importance of other views.
The assumption that posting something online removes your ownership of it also has implications for responsibility and accountability. When the systems trained on public information cause harm, distance from those whose information was used to train the system, as well as the developers and deployers of the system itself, makes it much more difficult to attribute accountability. In our workplaces, we see a lack of accountability in the way people use LLMs for daily tasks and try to discourage this by picking people up on using generative AI to write emails or generate images.
Given that most of the content we post online is spread across multiple different platforms, the way our data is used and any attempts at compensation will depend on each platform’s terms of service. Under some circumstances, perhaps payment should go through the platform itself. DeviantArt is an example of a platform that has fallen prey to the boom of generative AI through a plague of bots and failure to protect users. The events surrounding DeviantArt made us wonder how outspoken users need to be when a platform changes its terms of services, and if they should have to speak up every time. It seems unfair to place the burden of holding platforms accountable on the users, however, it’s important that users do not become desensitised to changes in terms of service. Stack Overflow has also seen a drop off in users coinciding with the rise in AI coding assistants on top of growing dissent regarding the perceived culture of elitism.
Discuss the environmental impact of generative AI. What, if anything, is distinct about AI’s environmental impact compared to computing in general or other specific digital technologies with a large energy use such as cryptocurrency?#
An important distinction between generative AI and other energy intensive technologies, such as mining cryptocurrency or running complicated physics simulations, is that generative AI is extremely easy to use. The ease of access prevents people from thinking about environmental concerns; because it’s easy to prompt, the impression is given that it’s computationally easy to “do”. Many people see computers as magic boxes that do things and are not aware of their computational costs. Even for technical people it’s hard to find information about the environmental impact of computers, especially regarding the emissions associated with manufacturing. For example, manufacturers do not provide information about the impact of GPUs compared to CPUs. It’s important to widen awareness of the mechanics behind AI tools, how they work, and what they’re good and bad at. Equipping people with this knowledge helps them to be more selective in their use, so that instead of jumping straight to an LLM they first use a calculator or Wikipedia.
Chapter 5 - Is Advanced AI an Existential Threat?#
Chapter Summary#
Artificial General Intelligence (AGI) is an ill-defined term used to describe the stage of AI development represents a threshold for large-scale risks becoming serious. As AGI is a philosophically contentious concept, the chapter adopts the definition of AGI as “AI that can perform most or all economically relevant tasks as effectively as any human”. This definition, the chapter argues, bypasses the need to delve into questions about subjectivity and consciousness.
Stories about advanced AI have captured our imagination over and over again in books (Asimov’s Foundation series), films (2001: A Space Odyssey, The Terminator), and even ancient mythology of man-made artificial beings (Golem, Talos). Today, the presence of AI in everyday life combined with generative AI’s quasi-realistic depiction of sentience narrows the gap between those imagined worlds and our own experiences.
Many prominent figures in tech lean into the science-fiction narratives surrounding AI, with some claiming to be directly working towards “superintelligence” and others voicing concerns about AI causing catastrophic harm. Researchers concerned about existential risks of AGI argue that it should be a global priority alongside threats such as pandemics and nuclear war. Yet, arguments for existential threats of AGI are based on a tower of fallacies. Reasons for the prevalence of these false narratives could be due to selection bias, where people are drawn to AI research exactly because of the prospect of building an all-powerful technology, and cognitive bias, wherein the idea of imminent and terrifying AGI adds an aura of grandeur to one’s work.
Whilst AI has become increasingly generalised over the years, at each stage it has proven difficult to tell whether the current dominant paradigm can be further generalised or if it is a dead end. The latest rung in the ladder of generality is the use of LLMs to specify a computational task without having to do any programming. This step has turned AI into a consumer tool, making it accessible to people regardless of their understanding of programming or computation.
Whilst advances in generative AI have facilitated a jump in generalisability, the history of AI research has shown us that once one aspect is automated, other aspects previously unrecognised tend to reveal themselves as bottlenecks. Automating the whole pipeline of programming, to the extent that AI can program itself, seems unlikely without a colossal amount more data than is currently available. The gap between chatbots and embodiment is also significant, with robotics facing several massive obstacles. Even if it were possible to create AI that is truly intelligent, it seems unlikely to go “rogue”. There is so much context to consider in catastrophic scenarios like turning the whole world into paperclips that the context under consideration probably contains enough information for the AI to identify that it is a bad idea. If an AI were to attempt to take over the world, it seems unlikely it would be able to acquire enough power to do so; it would surely make similarly misjudged mistakes along the way, such as breaking social norms, that would lead to it being shut down.
Technical constraints, as advocated for by the AI safety community, are likely infeasible. A better way to mitigate AI harms is to focus on the risks of how people can misuse AI such as cybersecurity, biological misuse, and disinformation. Portraying AI as all-powerful overstates its capabilities and underemphasises its limitations, playing into the hands of companies that prefer less scrutiny and making it less likely that people will spot and challenge AI snake oil.
Rather than scaling AI progress in terms of “intelligence”, which is hard to measure in any meaningful way, perhaps it is better to measure progress in terms of power. Power, defined as the ability to modify the environment, aptly frames how technological capability has accelerated at certain points throughout history. A focus on power reframes the argument to demonstrate that it is us who are being altered and made more powerful. Shifts in power dynamics will continue as AI capabilities improve: “we are the ‘superintelligent’ beings” and we should be far more concerned with what people do with AI than what AI will do on its own.
Discussion Summary#
In AI safety policy, entrenched camps have developed, with vastly divergent views on the urgency and seriousness of catastrophic risks from AI. While research and debate are important, policymakers must make decisions in the absence of expert consensus. How should they go about this, taking into account differences in beliefs as well as values and stakeholders’ interests?#
Policymakers should leverage expert opinion to inform decision-making regarding AI, however, defining what constitutes expertise is tricky and often unclear. Expert opinion should be considered alongside the idea of situated knowledge, which emphasises the importance of the context in which knowledge was produced. A balanced view of expertise should be considered, similar to how experts are used in court cases to present supporting evidence for explanations. In court, the judge and jury must compare each side and fathom which argument is most credible. The duty of expert witnesses is to the court, rather than to the defence or prosecution. Similarly, experts working with policymakers should view their duty as to society, rather than to the interests of a particular company or organisation. It can be more difficult to quantify conflicts of interest in the intellectual realm compared to economic conflicts of interest.
Whilst a balanced view of diverse expert opinion supports more informed decision-making, on a practical level there are some scenarios (such as the pandemic) where it is almost unhelpful for experts to cast doubt on scientific messaging. We wondered if disagreement among experts is also unhelpful in other policy situations. As scientists, we embrace discussion and disagreement, but scientific disagreement isn’t always appreciated by the press and politicians.
There is something about sitting in ambiguity that many people don’t like. Often, decision-makers would prefer to justify the course of action as being correct, rather than admitting that they aren’t certain what is best but have decided upon a particular route because of some justifiable reasons. In our experience as expert witnesses, we have found that we have to dress up the evidence in a way that is appealing and gives people a sense of security. Especially in medical settings, people want to be really clear about what’s happening to them. In saying this, research has also found that people are more comfortable with uncertainty than we might assume.
To make decisions about AI policy, it is important that policymakers consider a wide scope of possible risks. Threats policymakers should be taking seriously include the impact of AI on education, the environment, and cognitive functioning; we would have liked more discussion about these kinds of risks in the chapter. AI can cause significant threat to society through its integration into vital infrastructure, such as banking systems. We have seen the harm that faulty AI can cause, such as the Robodebt and Post Office scandals. AI also poses risks to the integrity of creative industries and the propensity of creatives to generate an income; the UK government has proposed a copyright exemption for commercial generative AI training.
We wondered if there would be more regulation and transparency if AI tools were primarily developed by non-profit companies. OpenAI was originally non-profit, but transitioned to a for-profit model in 2019. We wondered if OpenAI was acting mainly in the interests of its funders when the company was non-profit, despite claiming that its aim was “to build value for everyone rather than shareholders” – shareholders that included big tech companies such as Amazon and figures like Peter Theil.
What change would you like to see on the basis of this piece? Who has the power to make that change?#
We should hold AI products to the same standards as other (e.g. diagnostic) tools. There should be someone that can be held accountable for any decisions; someone who checks the outputs and makes the final call. Human accountability is crucial because we can’t just blame the tools; tools cannot feel remorse or make efforts to correct mistakes.
Whilst important, human control is not a tick box and must be careful thought out. It is a very human tendency to avoid blame, so efforts must be made to mitigate this. There have been instances of Tesla cars giving control back to human drivers split seconds before a collision, which is clearly not enough time for a person to course correct. Regardless of how effective this is, as the control has been returned to the human, there is room for Tesla to claim that the fault of the accident does not lie with the technology but with the human.
AI does make mistakes, but we think it will continue to improve as development progresses. Some of us self-describe as AI optimists and have high hopes for the potential of AI to benefit humanity, aside from the climate impact which we acknowledge needs addressing. Copyright issues could conceivably be worked out in the future, as the legal system adapts to the new landscape we find ourselves in. There will probably still be a demand for artists and creatives, as there is something special about the human element in these domains.
At the scale of AI use within organisations, we do think it will be possible to turn it off or stop using it. We wondered what the ethical route is if you realise your company is causing harm; whether you just stop working and let others take over, or if it is possible to switch paths and keep doing the work in a more ethical way. For individuals, we think that stopping usage of AI might be harder, depending on what their skills are and how comfortable they are with the technology.
Chapter 7 - Why do myths about AI persist?#
Chapter Summary#
AI is surrounded by myths that overstate its capabilities. These myths contribute to a general willingness of society to accept AI products without proper scrutiny leading to drastic misuse. One example of misuse is healthcare company Epic’s sepsis prediction model, which was used by tens of thousands of clinicians before being evaluated by independent researchers who found it had just 64% relative accuracy (where relative accuracy is the probability that a patient who would go on to develop sepsis would be rated as higher risk than a patient who wouldn’t develop sepsis).
Epic’s sepsis model exemplifies a phenomenon known as the hype cycle. Conceptualised by Gartner, the hype cycle depicts the four key stages a technology goes through from introduction to adoption. After a new technology is introduced, there is a sharp rise in visibility that builds to the “peak of inflated expectations”. Invariably, expectations are not met, which results in a sharp dip in visibility labelled as the “trough of disillusionment”. As society configures its expectations and disillusionment, visibility climbs the “slope of enlightenment” before balancing out in the “plateau of productivity”, wherein the product gains mainstream adoption and success.
In practice, however, technologies rarely evolve along the Gartner hype cycle. Every year Gartner releases a list of technologies alongside their place on the cycle. Most of those products appear for only one year before dropping off. The issues with the Gartner hype cycle are even more apparent with AI, as AI encompasses a broad range of technologies with varying fundamental limits and applications.
Each participator in the AI hype cycle has varying characteristics and behaviours delineating the role they play in perpetuating hype. Companies have commercial interests in spreading hype, as it helps them to sell their products. Developers game accuracy to make models appear to be more capable than they really are. Public figures viewed as experts (such as tech CEOs) have portrayed AI as a form of unknowable omnipotence, which should be held alongside consideration of their significant vested interests. News outlets are chronically underfunded, meaning that journalists often do not have the time or expertise to comprehensively verify the claims of researchers, public figures, and companies. The media also has a tendency to rely on grand metaphors that misrepresent AI capabilities and propagate sensationalist stories. Humans share cognitive biases making us susceptible to hype, such as anthropomorphising AI and difficulties in recognising the limits of our own knowledge. Misinformation is amplified by the illusory truth effect, which is where the mere repetition of inaccurate information can lead us to believe something is true.
Academia and researchers also contribute to spreading hype. AI researchers have a tendency to focus on improving benchmark performance and engineering breakthroughs, rather than prioritising scientific understanding. Prominent researchers have made speculative claims about AI which have been assumed to be true because of the author’s credibility, despite a lack of empirical evidence. Researchers can also misuse language to imply that tools perform better than they actually do and dismiss the value of domain experts (such as radiologists).
Discussion Summary#
One difference between AI research and other kinds of research is that most AI research is purely computational and doesn’t involve (for instance) experiments involving people or arduous measurements of physical systems. In what ways does this make it easier to have confidence in the claims of AI research? In what ways does it make it harder?#
Reliance on computational experiments rather than humans or physical systems can promote confidence in AI research via the ability to precisely dictate relevant variables, inputs, and hyperparameters. Knowing these details facilitates more exact reproduction of experiments compared to human studies, as it is possible to numerically specify and recreate a computer program, but harder to do the same for people. Stability in computational experiments involving AI models may depend on whether the approach is static or dynamic. ML, a dynamic approach where models are iteratively updated, is less affected by data completeness than other types of studies, as it predicts patterns between data points and can infer missing data given the patterns it has learnt. This functionality can be useful in predicting rare phenomenon or counterfactuals, which is very difficult to do when the real world data that is available isn’t very clean. ML may offer a way to mitigate poor quality data.
However, it is difficult to ascertain the accuracy of dynamic predictive models with respect to rare outcomes. Dynamic approaches may be more vulnerable to changes in the environment and thus less stable. For example, changes in input data can result in phenomena like catastrophic forgetting. It can be hard to discern when environmental changes are significantly affecting model performance, as ML approaches are largely black box. Explainability challenges with black box methods obfuscate model performance assessment and identification of when models are struggling. Data-heavy weather prediction models may improve on deterministic and static systems if there are good, clean data sources available.
Computational AI research is vulnerable to difficulties with defining and bridging accuracy gaps between the lab and the real world, limiting the confidence that we can have in it. Model accuracy measured in the lab is not reflective of model performance in the real world. Accuracy can be assessed by exposing the model to test sets comprised of data that the model hasn’t seen before, but there are a lot of assumptions that have to be made. The design of test sets can be quite contrived and distant from real world examples, as datasets have been cleaned and are designed to test the model in question, possibly in ways to improve the appearance of the model’s performance.
In the real world, there are a range of effects and complications that should be considered in model evaluation but cannot be foreseen until deployment has already happened. In some cases it is nearly impossible to articulate what counts as accuracy. For example, is a model “wrong” if it predicts 80% chance candidate A wins and 20% chance of candidate B wins, and candidate B wins? We need to have better standards of measurement than reductionist classification accuracy metrics, so that we can look into whether a system actually makes peoples’ lives better.
The end of the chapter highlights cognitive biases present when we hear claims about AI. Which biases have you experienced? Which ones have you witnessed at data ethics club?#
Cognitive biases we have seen in relation to AI tend to fall into two main camps: anti-technology bias and authority bias. Anti-technology bias manifests as strong criticism of AI capabilities and the use of AI in everyday life. Authority bias can manifest as overconfidence in AI ability, techno-solutionism, thinking that AI is better than humans, or fears that AI is going to take over the world. We have seen bias against AI at DEC, however, we feel that a healthy level of scepticism is important to ensure that systems are critically examined so that they can be improved. We have seen AI authority bias in students who over-use AI tools in their work, for example, by using generative AI for tasks that it is not suitable for like generating references. Some of these students seem to place more belief in the abilities of AI than their own. Many people take the outputs of generative AI as correct, without appreciating that those outputs may be faulty or based on low quality sources.
Overconfidence in the abilities of AI is mirrored in the commercial domain, where many companies are investing in AI without sufficient analysis of true costs and benefits. Companies are motivated to integrate AI products to cut costs yet are not considering the long term effects of this. Despite the way that AI is marketed, difficulties with measuring accuracy when AI is deployed in the real world mean that the true costs of AI may not justify replacing people. The immediate financial incentives, however, imply that people will continue to be replaced by AI despite its performance issues. We would prefer to see AI products used as an assistive tool used alongside and continuously audited by a human.
AI authority bias in the public reveals shortcomings in technology education and an overestimation in AI capabilities. For example, there is a pervasive association between AI and robots with much depiction of AI in the media confabulating AI with robotics. However, there is still a big leap between AI and embodiment through robotics.
What are some ways to improve accountability for companies making unsubstantiated claims? These could include legal remedies as well as non-legal approaches.#
One of the biggest accountability hurdles is inadequate education of those whose responsibility it is to enforce it. If accountability enforcers do not understand what they’re looking for, it is likely that bad practices will slip through the net. Hype worsens education deficits, for which the media hold a level of responsibility. Even negative media about AI propagates hype by framing the technology in a particular way and amplifying discussion about it. The long term view of hype is that it waxes and wanes, yet we feel somewhat unequipped to critique it, given the pervasiveness of hype about AI and its influence on society.
The ubiquity of AI hype limits accountability, and influential actors have taken advantage of this. Persistently emphasising the imminency of AGI exaggerates the appearance of AI capabilities. There seems to be a megalomanic culture within the AI industry with people thinking that everything they’re doing is world changing. This narrative aggrandises AI by masking its important limitations and encourages the illusory truth effect.
AI companies have been chipping away at core societal values by reframing previously unacceptable behaviours like copyright theft and student cheating in ways that make them increasingly normalised practices. The hijacking of societal values to suit commercial interests is unethical and companies should be held to account, but their sway over the AI narrative muddies the ability of accountability enforcers to identify and penalise bad behaviour.
Financial incentives underpin much of the behaviour surrounding AI, and making these incentives explicit is something the media covers relatively well. The press has historically been excellent at communicating how much money has been invested in what businesses and by who. Despite this, there has been increasing attention on completely false claims (such as by political world leaders) which arguably the press, and we as a society, are not doing enough to combat.
Improving public technology education and the quality of discourse around AI is also the responsibility of academics. Across various academic disciplines, we have witnessed the communication of research being a bit stretched to increase the likelihood of funding. Academic funding constraints have also contributed to a growing entanglement between industry and university research. There is increasing encouragement by universities for academics to productise the results of their research. It is important that academic research is primarily incentivised by the pursuit of knowledge rather than money, and that this knowledge is effectively disseminated to society.
Engendering good public understanding of technology requires trust in facts and experts. If there is a lack of trust, it is very difficult to police (in)accuracies as everything becomes conjecture. Often, we believe the claims of others depending on how trustworthy we find the person. Whether or not this is the best way of determining trustworthiness, factors to consider tend to include what we know about the person’s expertise and how (un)certain they appear in their knowledge.
There are other key factors that should be considered but are often ignored in determining trustworthiness, such as the motivation of the speaker. There are bizarre theories floating around technology epicentres, such as the popularity of simulation theory in Silicon Valley, which has some weird motivational consequences if people really believe it. Foundational moral values become no longer important, and life just becomes a levelling-up game. Public figures have explicitly connected AI with god and presented religious visions of superintelligent AI. At the minimum, there seems to be a dismissal of the value of human expertise in favour of machine automation.
To improve technology education and support accountability, it is important that efforts are made to counter anthropomorphising AI. People have a tendency to assign human qualities to things, and AI is no exception. AI anthropomorphism is a form of hype as it exaggerates its capabilities, with detrimental effects such as distorting moral judgments. To mitigate anthropomorphism, we must be careful when using analogies between humans and machines such as describing machines as “thinking”, “feeling”, “learning”, or “believing”.
Being intentional with the language we used is essential in order to avoid misconceptions about AI and yet avoiding these kinds of analogies whilst concisely expressing complicated topics is hard. There is something slightly magical in the way we talk about AI. AI seems especially difficult to describe without anthropomorphism as it is so deeply embedded in its foundation: the pursuit of the discipline is literally to develop artificial “intelligence”.
It is much easier to avoid human analogies in domains like maths, so perhaps AI should be more commonly framed as a maths problem. The wider public does not need to understand the underlying maths, but they should have an intuition that ML models are probabilistic. Despite the depiction of AI as somewhat unknowable, we do actually understand the maths and statistical theory behind them quite well; it is essential to demystify this. We see a growing need for AI education throughout schooling, however, are concerned that perhaps it may be too late to incorporate the necessary structural changes.
What change would you like to see on the basis of this piece? Who has the power to make that change?#
At the moment, much AI research falls into minimal risk categories and so has relatively low scrutiny. Research at universities may be perceived as minimal risk because the impact tends to be lower than commercial research insofar as outputs generally aren’t productised and deployed across large populations. In this sense, risks from academic research may lie more with miscommunication of research.
However, overestimating the gap between AI research and research involving human subjects has ethical implications. If a study does not have direct human subjects, but it does affect people on some level, perhaps those studies should also come under some sort of institutional review board. For example, there seems to be a growing use of LLMs in politics, such as to summarise judicial decisions. This does not usually need ethics approval as it does not deal directly with human subjects, but it is handling human data and therefore perhaps the ethics around how the data is interpreted should be considered.
In addition to people directly included in studies, ethical review boards should consider more secondary risks which may involve the researchers or students running the studies themselves. There are grey areas where even asking the question can be ethically contentious, such as emotion or gender recognition and image generators (e.g. for impersonation). Risk assessment should also consider temporal aspects, and we have heard talk at our intuitions of incorporating long term ethics into processes. For AI research, long term ethics should encompass the effects of AI infrastructure and environmental impact.
To foster more ethical research practices, it is important that researchers from a wide range of disciplines come together to help each other understand their fields. Many researchers do not understand how their fields are different, and the methods or hoops that are specific to each area. The contrast between steps that people have to take in research (such as the ethics processes for research involving or not involving human subjects) can be quite large. It would be useful if different fields talked to each other to voice concerns and share tips about research practices.
Chapter 8 - Where do we go from here?#
Chapter Summary#
To help us understand how the role of AI in society is likely to evolve, we can look to to the history of the internet. At its beginnings, the internet was a platform for which people logged onto for specific purposes. Now, the internet has shifted into a medium over which an immense amount of communication and information exchange happens. It seems likely that generative AI will similarly retreat into the background as a medium for knowledge work, instead of a tool used for particular purposes.
Parallels between the internet and AI include a shift from public funding and expertise to private ownership. The internet began as a largely government funded project and has become increasingly privatised. Similarly, until recently most AI research was relatively open, yet competitive pressure has led to AI companies reducing the amount of research they share. Privatisation of the internet has centred focus on cost related metrics such as engagement and ad revenue, causing a plethora of harms such as the exacerbation of inequality in accessibility, cost, and content moderation.
As AI is likely to continue to have a big impact on society, it is important to shape its trajectory in a way that promotes the public interest. Reshaping the role of AI requires a change in largely social factors, such as the incentives of the institutions that use it. Journalism and educational institutions are often financially constrained and overburdened, creating a need for solutions that promise increased efficiency and cost cutting. AI companies market products to these organisations with promises of reduced human decision-making and thereby reduced costs.
We can help change harmful practices by countering flawed proposals in the companies we work for and participating in local democratic processes to ensure that the voices of local communities are heard in the adoption of AI technologies. We should look for strategies or policies that achieve modest efficiency gains whilst being simple enough for all stakeholders to understand, so that trust can be built between decision-makers and decision subjects. Embracing a reasonable level of randomness and explicitly acknowledging the randomness that already exists in decisions can help counter flawed prediction attempts. Incorporating more randomness reduces the incentive for people to game systems and optimise unimportant goals, leaving more space to focus on the things that matter.
As companies are driven by profit, the reputational damage from AI harms can sometimes provide sufficient motivation for companies to fix problems. For example, offensive comments by early language models like Microsoft’s Tay led to them being taken offline. Other harms might be indirect or too diffuse, meaning that it isn’t in the company’s interest to fix them, like the loss of income faced by artists and lost time of teachers. When companies have no internal incentive to address the harms they bring about, regulation is essential.
Regulatory capture is when a regulator is co-opted to serve a company’s interests rather than the public’s. Regulatory capture has been repeatedly attempted by big tech, where companies have poured hundreds of millions of dollars into advertising to try to influence the direction of regulation. Regulatory capture happens when regulators are misinformed or lack resources to operate independently and can be avoided by proper funding of regulatory bodies.
Regulation is key to ensuring safety for society, as well as to create space for people and companies to innovate safely. The frameworks needed to regulate AI already exist in many jurisdictions, despite the pervasive myth that they are in their infancy. AI regulation in the US is vertical, with federal agencies having the authority over specific sectors. AI regulation in the EU is horizontal, with rules that apply over many sectors such as the General Data Protection Regulation (GDPR), the Digital Services Act (DSA), and the Artificial Intelligence Act (AIA). China initially had vertical regulation and is planning to draft a horizontal AI law. This trajectory is similar to China’s strategy for internet regulation where narrow vertical regulation gave way the broader Cybersecurity Law.
General principles that underpin law, such as freedom of speech, can be used as guiding mechanisms for how the law should address evolving technologies. The goal should be to make regulation more responsive, flexible, and informed; facilitated by increasing funding so that regulatory agencies have the capacity to develop and enforce better frameworks.
Discussion Summary#
The chapter makes the point that broken AI appeals to broken institutions. What are some examples of broken institutions enamoured by other dubious technologies? Is there something about AI, as opposed to other technologies, that makes it liable to be misused this way?#
Many of us were disappointed by the chapter and the societal issues highlighted, but also weren’t that surprised at the states of affairs presented. The UK government has been very vocal about the promises of AI to increase productivity and are pushing for accelerated AI development. Yet, the government’s AI plans face serious challenges including requirements for enormous levels of investment to match the suggested plans and concerns with data privacy when integrating AI into fundamental infrastructure like the NHS.
Whilst there are issues with institutional adoption of AI, labelling institutions as “broken” seems a somewhat extreme characterisation. We would have liked further clarification of what the authors mean by broken, as well as some counter examples to illustrate institutions or organisations that are using technologies well. Often, there are people within institutions who are genuinely trying to do good things but are misinformed or ill-resourced. There seems no way to improve institutions without getting more money; cash strapped institutions will always seek magical solutions because they are strapped for cash.
Given how AI is marketed as a magic silver bullet, promising to solve an array of problems in mysterious and efficient ways, it is not surprising that it is so appealing to resource limited institutions. It’s concerning that there could be a vicious cycle where the most cash strapped institutions are the most vulnerable to snake oil and thereby more reliant on it. The organisations that benefit most from AI may end up being the ones that are the most well-resourced, as they can afford to be selective about what, where, and how AI is used.
What change would you like to see on the basis of this piece? Who has the power to make that change?#
The chapter promotes embracing randomness and suggests lottery-based systems as an approach to mitigate insufficient prediction mechanisms. We wondered whether lottery-based systems do actually help; perhaps they would be useful in settings where resources are scarce, or to avoid “rich get richer” situations.
Attendees#
Huw Day, Data Scientist, University of Bristol: LinkedIn, BlueSky
Jessica Woodgate, PhD Student, University of Bristol
Liz Ing-Simmons, RSE, King’s College London :wave:
Virginia Scarlett, Open Source Programs Specialist, UC Santa Barbara :coffee:
Beverly Shirkey, Medical Statistician (Clinical Trials) University of Bristol :smile:
Paul Matthews, Lecturer, UWE Bristol
Jessica Bowden, Research Associate, University of Bristol :cat:
Robin Dasler, data product manager, California
Julie-Myrtille Bourgognon, lecturer, University of Glasgow
Euan Bennet, Lecturer, University of Glasgow: LinkedIn, BlueSky
Martin Donnelly, Principal Research Data Steward, UCL, martin.donnelly@ucl.ac.uk
Nicolas Gold, Associate Professor, UCL
Social sciences focus on understanding causal mechanisms, not predicting associations as opposed to typical machine learners. What do both fields have to learn from one another?#
Traditional statistical models are based on explicit mathematical equations that can be interpreted by experts and explained to others. In comparison, understanding how a neural network makes a particular decision is very difficult. The connections the network learns between data are masked by the copious number of parameters involved, making it difficult to draw out what exactly happens within a model and how it is handling the data. This intrinsic implicitness complicates identifying how best to interpret the model and understand how the outputs are produced from the inputs.
Causality is a murky question and a philosophical rabbit hole, so there is a tendency among statisticians to avoid it. Yet, if we predict what will happen based on what has already happened, we will propagate and amplify historical issues into the future. Instead of using the past and all the baggage it comes with, predictions regarding human lives should place more emphasis on causality.
Attempts to understand causality must be accompanied by domain specific knowledge and a scientific mindset. A good example of the success of domain knowledge and scientific method coming together is the true story that was adapted into the film Moneyball. In Moneyball, baseball pundits had bad judgments about what made good players. Billy Beane, the manager of the Oakland Athletics baseball team, hired a Yale economics graduate and together they were able to derive what made players valuable and build a successful team.
ML models are very good at identifying patterns, but the utility of those patterns depletes if they are detached from domain expertise and we are unable to understand how the patterns are made. Domain expertise is essential to pursue actually useful models; without domain expertise to define what exactly we are looking for before starting the modelling process, there is little chance of finding useful patterns. In ML development there seems to be a tendency to chuck a bunch of data into black box models that the developers can’t explain, rather than putting in the work to incorporate domain expertise. The mystery that is so embedded in this process lends favour to the image of “tech bros” who are smart at STEM, so people assume they are good at everything else. This characterisation is bad – it stops people from questioning and criticising them, enabling bad decision-making and unaccountability.
As more and more information becomes available to us, we need more and more time to become experts. It is crucial to recognise expertise across fields and support experts in communicating with each other. Some of us wondered if this is an opening for generative AI to emulate expertise across an expanse of fields. Others worried that the use of generative AI here would mean skipping a communication step, and perhaps it would be more useful to use it to augment and support communication rather than replace it. If generative AI is used in replacement, we wondered if there is a point in proving a theorem if no-one can understand or replicate it. However, generative AI surpassing human ability in certain settings might not be important if the application of the technology furthers human knowledge; there are cases where geniuses like Stephen Hawking have proven something that nobody else could have.
To delineate (in)appropriate uses for generative AI we need to be clear about what our goal is; whether we want to learn something, or whether we just want a task to be completed. For some of us, proving something that no humans can understand is less interesting even if we don’t think it’s less valuable. At the very least, someone should be able to understand it. In education settings, there is too often an overemphasis on doing assignments and turning them in, rather than spending the effort to truly learn the lesson underpinning the assignment.