In selecting this blog’s title, I appreciate that I may ruffle some feathers. That is not my intention, though it is a bonus from a search marketing perspective. Truthfully, I selected the title because I believe it to be true.
That said, I should be a good anthropologist and acknowledge my own biases. I started in technology before deciding to get an anthropology degree, and I have worked in tech for about 15 years. However, I don’t think that is coloring my decision to a significant degree.
Artificial intelligence (AI) has been part of my anthropological workflow since 2018. While I may be on the earlier side of the technology adoption life cycle, AI is the future of anthropology, and it is coming for most anthropologists like a freight train, like it or not.
Now I appreciate that this is a bold claim, so I will provide some common examples of AI that you may already be using, as well as examples of emerging technologies that I believe will shape anthropological theory, ethnography, and our discipline as a whole.
Before doing that, however, it seems prudent to operationalize AI and discuss the relevant subfields.
On Artificial Intelligence and its Subfields
Artificial Intelligence (AI) is a field of computer science and engineering focused on developing systems that can perform tasks that typically require human intelligence, such as visual perception, speech recognition, decision-making, and language understanding.
To be clear, when using the term AI, I am not referring to artificial general intelligence (AGI), typically used to describe a type of AI capable of performing an intellectual task as a human would.
Machine Learning (ML)
Machine Learning (ML) is a subset of AI that involves training computer systems to learn from data without being explicitly programmed. Machine learning algorithms can make predictions or decisions based on data, such as image classification, natural language processing, and anomaly detection.
Deep Learning & Neural Networks (NN)
Deep learning and Neural Networks NN are machine learning algorithms modeled after the human brain’s structure and function. Neural networks consist of layers of interconnected nodes, called artificial neurons, that process and transmit information. Neural networks can perform various tasks, such as image and speech recognition, natural language processing, and decision-making.
Natural Language Processing (NLP)
Natural Language Processing (NLP) is a field of AI that focuses on developing systems that can understand, interpret, and generate human language. It involves using computational techniques to analyze, understand, and generate natural language data, such as text or speech.
Natural Language Understanding (NLU)
Natural Language Understanding (NLU) is a subfield of NLP that involves the ability of a computer system to understand the meaning of human language input, such as text or speech. This can include sentiment analysis, named entity recognition, and intent recognition.
Natural Language Generation (NLG)
Natural Language Generation (NLG) is another subfield of NLP that involves the ability of a computer system to generate human-like text or speech. This can include text summarization, text completion, and dialogue generation. To the chagrin of academic departments everywhere, this NLG has become quite popular in the last few weeks because of OpenAI’s chatGPT, which is based on the firm’s deep learning autoregressive language model, GPT-3.
Common Examples of AI in Anthropology
If you have used any of the following applications, then you have used AI as part of your anthropological work practice.
Research and Writing
We all spend much of our time doing secondary research and writing, and increasingly AI is typically involved in that process. When conducting secondary research on academic websites or in Google searches, recommender systems based on machine learning are increasingly behind the content suggestions. When writing, our sentences are often auto-completed in products like Google for Education/Business and Office 365. This may seem minor in the big picture, but it is AI, and you are already using it.
Transcription Services
If you have tried any automated transcription services like Otter, Rev, or Temi, you have used AI, and in particular, the board family of natural language processing. Now I know what you may be thinking. These tools don’t work that well. Agreed, but I am already testing Open AI’s Whisper, and I can tell you it is a giant leap forward. When it or similar speech-to-text models arrive in consumer tools within the next year, you will feel differently about automated transcriptions.
Using & Editing Multimedia
Most of us use multimedia as part of our work. Be it audio, photos, or video, we often use multimedia to support our text in telling impactful stories, and AI is in the loop. From how we collect that data to how we edit it, AI is used to improve our abilities. For example, many of us use smartphones today to capture photos and videos, and those phones use AI to capture the highest quality possible.
AI is also present if we do any post-processing or audio/video editing. Be it as simple as touching up a photo in Adobe Lightroom or using editing programs like Descript to intelligently remove background noise and filler words. AI is helping us accomplish these tasks faster and with greater precision than in the past.
Geographic Information Systems
While some readers may think my above examples lean towards cultural anthropology, don’t be alarmed, AI will touch all branches. If you work as an archeologist, you have likely used geographic information systems (GIS). These systems already incorporate deep learning and computer vision to provide more accurate and sophisticated maps.
Language Studies & Preservation
Similarly, linguistic anthropologists have a lot to discuss. Large language models (LLM) are in the early stages but improving exponentially over the past 18 months. For example, Meta recently announced No Language Left Behind (NLLB), which can translate across 200 different languages, even with languages such as Kamba and Lao, which were not previously supported by existing translation tools. Indigenous computer scientists such as Running Wolf are even starting to study the use of language models to make language learning more accessible, immersive, and engaging.
Imagine the potential for linguistic anthropologists and computer scientists to work together in this field. Or for cultural anthropologists to be engaged in creating a truly ethnographic algorithm. The potential abounds.
Emerging Applications of AI to Anthropology
While these common examples are helpful, they were primarily used to make a point about the ubiquity of AI in recent years. But my genuine interest in AI and anthropology, or more broadly, the social sciences, lie ahead and are being researched and designed by scholars worldwide as I write. For a summary of some of that work, check out the UNESCO/LiiV Center’s first global report on digital anthropology.
However, it should be noted that, like with any emerging technology, there is much hype around AI anthropology, and some of the promises may not come true. That may even apply to assertations that follow. But one should not be dissuaded by such realities. That is expected in any emerging field, and as we know, anthropologists are not futurists, even if we work to understand, define, and design potential futures.
So then, what interests me as a scholar and practitioner? Here are a few of my key takeaways.
AI-Assisted Ethnography
AI-assisted ethnography will leverage the capabilities of AI to collect, analyze, and interpret qualitative and quantitative data in ethnographic research. It builds on the common examples above but will do so in much more powerful ways by bringing advanced techniques to all researchers through low-code and no-code offerings.
The offerings will take many forms, but the obvious and imminent opportunities include the following:
- Speech-to-text and large language models will transcribe and translate between languages accurately so researchers can work across the language barrier more easily. This will apply to audio captured as part of interviews and observations, and digital traces scraped from the web in the form of videos and podcasts.
- NLP will analyze text data, be it colls text or transcribed from audio and video. Text extraction techniques like keyword and entity extraction, topic modeling, and sentiment analysis will all be used to assist in coding and analysis.
- NLP will also summarize existing research as part of the secondary research process. Entire academic articles will be summarized into their essential points to assist us in finding research that is helpful to the task.
- Computer vision will identify patterns and themes in visual data, including photos and video, once converted to text labels and combined with existing text data sets.
- Network analysis will be used to understand online networks and the relationships within corpses of texts.
Again, while these are not new, and many anthropologists are already using such methods, the difference in the future will be that the technology will become democratized. All researchers will have the ability to scale their capabilities and improve their efficiency.
It is worth mentioning that these tools should not be viewed as a replacement for us as researchers but instead as assistive tools that we use as part of our creative work practice.
Multimodal AI
Anthropologists have long appreciated the value of multimodal data in understanding the complexity of human culture. Sadly, many publishers have often acted like AI models, which is to say, they rarely support multimodal data. Thankfully that is changing, at least on the AI side of the equation, because in 2021, Google released Multimodal Unsupervised Machine learning (MUM)
MUM is a model capable of understanding and analyzing multiple forms of data, including text, images, audio, and video. It can identify patterns and relationships across these different modalities and can be used for various tasks such as image captioning, image retrieval, and audio-visual scene understanding.
Multimodal AI models like MUM will become the norm over the next few years. With that, anthropologists will have the opportunity to use fewer tools that can handle more of their AI-assisted ethnographic needs. With this change, we will again see our abilities scale as we achieve greater efficiency in our workflow and realize new opportunities to focus our creativity on finding rich insights and making recommendations.
Anthropology-Specific AI Tools
One of my concerns with AI tools, such as applications built on top of the existing large language (LLMs) models, is that they are not anthropology specific or even very anthropologically aware. Most are trained on publicly available data sets scrapped from the web. While this has resulted in surprisingly exceptional tools like chatGPT, the tools are still incapable of producing rich and nuanced interpretations like a human-anthropologist.
While that might sound like a breadth of fresh air to many anthropologists, I view it as a threat. If the public uses tools that miss the depth of human culture to make decisions, there is a tremendous opportunity for error. Likewise, I argue that we need anthropology-specific AI tools, such as an anthropology large language model (LLM) trained on the anthropological corpus.
An anthro LLM trained specifically on anthropological data and designed to understand the cultural and social context of the information it analyzes would provide added benefits over other models today, especially if it used a knowledge graph and elements of symbolic AI. An example is that it would allow researchers to gain a more comprehensive understanding of human behavior, cultural dynamics, and cultural-specific nuances, which general LLMs lack, and speak back to us in the language of anthropology.
Such a model would powerfully move AI anthropology forward and help overcome a significant hurdle today, which is that the current tools can’t make the leap from data and findings to anthropological insight. They are fantastic aids that help us see patterns and inform our own insights, but the current crop of AI tools falls short of what we should expect from an intelligent agent.
Thankfully, I see that changing, and while I am again not suggesting an AI should be a replacement for anthropologists, I do believe that an imagined future in which every researcher has an AI agent capable of acting as a co-contributor will be a valuable asset to the discipline given our increasingly time and budget crunched work cycles.
Closing
Like any tool before, AI will shape human society for good and bad. While I have opted to focus on some of the positive changes I see coming to the future of anthropology, genuine contemporary concerns need to be addressed. Ethical issues such as bias and fairness, transparency and explainability, privacy, autonomy and agency, and job displacement are significant concerns and need to be examined. We also need proper governance structures in place if we are going to ensure AI is used for good.
But despite those concerns, we should also recognize that AI is creating new opportunities that, as anthropologists, we should appreciate. As we know, culture is learned, shared, and constantly adapted. We need to do the same when it comes to AI.
Whether studying AI as a field site, working with engineers to improve AI models, or using the tools in our work, AI is the future of anthropology.