Bowdoin Science Journal

Computer Science and Tech

Biological ChatGPT: Rewriting Life With Evo 2

by

What makes life life? Is there underlying code that, when written or altered, can be used to replicate or even create life? On February 19th 2025, scientists from Arc Institute, NVIDIA, Stanford, Berkeley, and UC San Francisco released Evo 2, a generative machine learning model that may help answer these questions. Unlike its precursor Evo 1, which was released a year earlier, Evo 2 is trained on genomic data of eukaryotes as well as prokaryotes. In total, it is trained on 9.3 trillion nucleotides from over 130,000 genomes, making it the largest AI model in biology. You can think of it as ChatGPT for creating genetic code—only it “thinks” in the language of DNA rather than human language, and it is being used to solve the most pressing health and disease challenges (rather than calculus homework).

Computers, defined broadly, are devices that store, process, and display information. Digital computers, such as your laptop or phone, function based on binary code—the most basic form of computer data composed of 0s and 1s, representing a current that is on or off. Evo 2 centers around the idea that DNA functions as nature’s “code,” which, through protein expression and organismal development, creates “computers” of life. Rather than binary, organisms function according to genetic code, made up of A, T, C, G, and U–the five major nucleotide bases that constitute DNA and RNA.

Although Evo 2 can potentially design code for artificial life, it has not yet designed an entire genome and is not being used to create artificial organisms. Instead, Evo 2 is being used to (1) predict genetic abnormalities and (2) generate genetic code.

11 Functions of Evo 2 in biology at the cellular/organismal, protein, RNA, and epigenome levels.
Functions of Evo 2 at different levels. Adapted from https://www.biorxiv.org/content/10.1101/2025.02.18.638918v1.full

Accurate over 90% of the time, Evo 2 can predict which BRCA1 (a gene central to understanding breast cancer) mutations are benign versus potentially pathogenic. This is big, since each gene is composed of hundreds and thousands of nucleotides, and any mutation in a single nucleotide (termed a Single Nucleotide Variant, or SNV) could have drastic consequences for the protein structure and function. Thus, being able to computationally pinpoint dangerous mutations reduces the amount of time and money spent testing each mutation in a lab, and paves the way for developing more targeted drugs.

Secondly, Evo 2 can design genetic code for highly specialized and controlled proteins which provide many fruitful possibilities for synthetic biology (making synthetic molecules using biological systems), from pharmaceuticals to plastic-degrading enzymes. It can generate entire mitochondrial genomes, minimal bacterial genomes, and entire yeast chromosomes–a feat that had not been done yet.

A notable perplexity of eukaryotic genomes is their many-layered epigenomic interactions: the complex power of the environment in controlling gene expression. Evo 2 works around this by using models of epigenomic structures, made possible through inference-time scaling. Put simply, inference-time scaling is a technique developed by NVIDIA that allows AI models to take time to “think” by evaluating multiple solutions before selecting the best one.

How is Evo 2 so knowledgeable, despite only being one year old? The answer lies in deep learning.

Just as in Large Language Models, or LLMs (think: ChatGPT, Gemini, etc.), Evo 2 decides what genes should look like by “training” on massive amounts of previously known data. Where LLMs train on previous text, Evo 2 trains on entire genomes of over 130,000 organisms. This training—the processing of mass amounts of data—is central to deep learning. In training, individual pieces of data called tokens are fed into a “neural networks”—a fancy name for a collection of software functions that are communicate data to one another. As their name suggests, neural networks are modeled after the human nervous system, which is made up of individual neurons that are analogous to software functions. Just like brain cells, “neurons” in the network can both take in information and produce output by communicating with other neurons. Each neural network has multiple layers, each with a certain number of neurons. Within each layer, each neuron sends information to every neuron in the next layer, allowing the model to process and distill large amounts of data. The more neurons involved, the more fine-tuned the final output will be. 

This neural network then attempts to solve a problem. Since practice makes perfect, the network attempts the problem over and over; each time, it strengthens the successful neural connections while diminishing others. This is called adjusting parameters, which are variables within a model that can be adjusted, dictating how the model behaves and what it produces. This minimizes error and increases accuracy. Evo 2 was trained with 7b and 40b parameters to have a 1 million token context window, meaning the genomic data was fed through many neurons and fine-tuned many times.

Example neural network
Example neural network modeled using tensorflow adapted from playground.tensorflow.org

The idea of anyone being able to create genetic code may spark fear; however, Evo 2 developers have prevented the model from returning productive answers to inquiries about pathogens, and the data set was carefully chosen to not include pathogens that infect humans and complex organisms. Furthermore, the positive possibilities of Evo 2 usage are likely much more than we are currently aware of: scientists believe Evo 2 will advance our understanding of biological systems by generalizing across massive genomic data of known biology. This may reveal higher-level patterns and unearth more biological truths from a birds-eye view.

It’s important to note that Evo 2 is a foundational model, emphasizing generalist capabilities over task-specific optimization. It was intended to be a foundation for scientists to build upon and alter for their own projects. Being open source, anyone can access the model code and training data. Anyone (even you!) can even generate their own strings of genetic code with Evo Designer. 

Biotechnology is rapidly advancing. For example, DNA origami allows scientists to fold DNA into highly specialized nanostructures of any shape–including smiley faces and China–potentially allowing scientists to use DNA code to design biological robots much smaller than any robot we have today. These tiny robots can target highly specific areas of the body, such as receptors on cancer cells. Evo 2, with its designing abilities, opens up many possibilities for DNA origami design. From gene therapy, to mutation-predictions, to miniature smiley faces, it is clear that computation is becoming increasingly important in understanding the most obscure intricacies of life—and we are just at the start.

 

Garyk Brixi, Matthew G. Durrant, Jerome Ku, Michael Poli, Greg Brockman, Daniel Chang, Gabriel A. Gonzalez, Samuel H. King, David B. Li, Aditi T. Merchant, Mohsen Naghipourfar, Eric Nguyen, Chiara Ricci-Tam, David W. Romero, Gwanggyu Sun, Ali Taghibakshi, Anton Vorontsov, Brandon Yang, Myra Deng, Liv Gorton, Nam Nguyen, Nicholas K. Wang, Etowah Adams, Stephen A. Baccus, Steven Dillmann, Stefano Ermon, Daniel Guo, Rajesh Ilango, Ken Janik, Amy X. Lu, Reshma Mehta, Mohammad R.K. Mofrad, Madelena Y. Ng, Jaspreet Pannu, Christopher Ré, Jonathan C. Schmok, John St. John, Jeremy Sullivan, Kevin Zhu, Greg Zynda, Daniel Balsam, Patrick Collison, Anthony B. Costa, Tina Hernandez-Boussard, Eric Ho, Ming-Yu Liu, Thomas McGrath, Kimberly Powell, Dave P. Burke, Hani Goodarzi, Patrick D. Hsu, Brian L. Hie (2025). Genome modeling and design across all domains of life with Evo 2. bioRxiv preprint doi: https://doi.org/10.1101/2025.02.18.638918.

 

Unsupervised Thematic Clustering for Genre Classification in Literary Texts

by

Book genres
(Chapterly 2022)

Summary

In the last decade, computational literary studies have expanded, yet computational thematics remains less explored than areas like stylometry, which focuses on identifying stylistic similarities between texts. A 2024 study by researchers from the Max Planck Institute and the Polish Academy of Sciences investigated the most effective computational methods for measuring thematic similarity in literary texts, aiming to improve automated genre clustering.

Key Findings and Assumptions

  • Key Assumptions: 
    • Text pre-processing to emphasize thematic content over stylistic features could improve genre clustering. 
    • Unsupervised clustering would offer a more scalable and objective approach to genre categorization than manual tagging by humans.
    • Four genres were selected (detective, fantasy, romance, science fiction) for their similar level of broad qualities.
    • If the genres are truly distinct in terms of themes, computers should be able to separate them into clusters.
  • Best Performance: The best algorithms were 66-70% accurate at grouping books by genre. Thus showing unsupervised genre clustering is feasible despite the complexity of literary texts.
  • Text Pre-Processing: Medium and strong levels of text pre-processing significantly improved clustering, while weak pre-processing performed poorly.
  • Which methods worked best: Doc2vec, a method that captures word meaning and context, performed the best overall, followed by LDA (Latent Dirichlet Allocation), which finds major topics in texts. Even the simpler bag-of-words method, which just counts how often words appear, gave solid results.
  • Best way to compare genres: Jensen-Shannon divergence, which compares probability distributions, was the most effective metric, while simpler metrics like Euclidean distance performed poorly for genre clustering.

Methodology 

Sample Selection

The researchers selected canonical books from each of the four genres, ensuring they were from the same time period to control for language consistency.

Sample Pre-Processing and Analysis 

The researchers analyzed all 291 combinations of the techniques in each of the three stages: text pre-processing, feature extraction, and measuring text similarity. 

Stage 1: Different Levels of Text Pre-Processing  

  • The extent to which the text is simplified and cleaned up.
    • Weak → lemmatizing (reducing words to their base or dictionary form (e.g., “running” to “run”), removing 100 Most Frequent Words
    • Medium → lemmatizing, using only nouns, adjectives, verbs, and adverbs, removing character names
    • Strong → Same as medium, but also replaced complex words with simpler versions.

Stage 2: Identifying Key Text Features through Extraction Methods

  • Transforming pre-processed texts into feature lists.
    • Bag-of-Words → Counts how often each word appears.
    • Latent Dirichlet Allocation (LDA) → Tries to discover dominant topics across books.
    • Weighted Gene Co-expression Network Analysis (WGCNA) → A method borrowed from genetics to find clusters of related words.
    • Document-Level Embeddings (doc2vec) → Captures semantic relationships (connections between words based on their meanings (e.g., “dog” and “cat”)) for similarity assessment.

Stage 3: Distance metric (Measuring Text Similarity)

  • Quantifying similarity with metrics. 6 metrics were chosen: 
    • Euclidean, Manhattan, Delta, Cosine Delta, Cosine, Jensen-Shannon divergence 

To minimize the influence of individual books on the clustering results, rather than analyzing the full corpus at once, the researchers used multiple smaller samples. Each sample consisted of 30 books per genre (120 books total), and this sampling process was repeated 100 times for each combination. Additionally, models requiring training (LDA, WGCNA, and doc2vec) were retrained for each sample to reduce potential biases.

Clustering and Validation

The researchers applied Ward’s clustering algorithm on the distances, grouping texts into four clusters based on genre similarity. They then checked how well these clusters matched the actual genres of the books. To do this, they used a scoring system called the Adjusted Rand Index (ARI), which gives a number between 0 (least accurate) to 1 (most accurate). 

The results were visualized using a map projection, grouping similar books closer together, and revealing the underlying thematic structures and relationships among the novels.

Core Findings and Figures  

Results  

The best algorithms grouped literary texts with 66-70% accuracy, demonstrating that unsupervised clustering of fiction genres is feasible despite text complexity. Successful methods consistently used strong text pre-processing, emphasizing the importance of text cleaning and simplification to focus more on a book’s themes rather than its writing style.

Among the top features, six of the ten were based on LDA topics, proving its effectiveness in genre classification. Additionally, eight of the best distance metrics used Jensen–Shannon divergence, suggesting it is highly effective for genre differentiation.

Generalizability  

To assess generalizability, five statistical tests were used to analyze interactions between text pre-processing, feature extraction methods, distance metrics, and other factors. These models provided insights into the broader effectiveness of various methods for thematic analysis.

Text Pre-Processing and Genre Clustering  

Text pre-processing improves genre clustering, with low pre-processing performing the worst across all feature types. Medium and strong pre-processing showed similar results, suggesting replacing complex words with simpler words offers minimal improvements in genre recognition. 

The benefits of strong text pre-processing for document embeddings, LDA, and bag-of-words were minimal and inconsistent. The figure below suggests a positive correlation between Most Frequent Words and ARI and the degree of text pre-processing and ARI. This demonstrates that how we prepare texts matters just as much as what algorithms we use. Moreover, researchers can save time by avoiding replacing complex words with simpler words if medium and strong pre-processing show similar results. 

Figure depicting the influence of the number of Most Frequent Words, used as text features, on the model’s ability to detect themes, measured with ARI.
Fig 1. The influence of the number of Most Frequent Words, used as text features, on the model’s ability to detect themes, measured with ARI (Sobchuk and Šeļa, 2024, Figure 6).

Feature Types and Their Performance  

Doc2vec, which looks at how words relate to each other in meaning, performed best on average, followed by LDA, which remained stable across various settings, such as topic numbers and the number of Most Frequent Words. Perhaps researchers can use this method without excessive parameter tuning. The simple bag-of-words approach performed well despite its low computational cost, perhaps suggesting even basic approaches can compete with more complex models. WGCNA performed the worst on average, suggesting methods from other fields need careful adaptation before use.

LDA Performance and Parameter Sensitivity  

The performance of LDA did not significantly depend on the number of topics or the number of Most Frequent Words being tracked. The key factor influencing thematic classification was text pre-processing, with weak pre-processing significantly reducing ARI scores. Hence, this underscores the need for further research on text pre-processing, given its key role in the effectiveness of LDA and overall genre classification.  

Bag-of-Words Optimization

The effectiveness of Bag-of-Words depended on a balance between text pre-processing and how many Most Frequent Words are tracked. While increases in Most Frequent Words from 1,000 to 5,000 and medium text pre-processing significantly improved accuracy scores, further increases provided minimal gains. This ‘sweet spot’ means projects can achieve good results without maxing out computational resources, making computational thematics more accessible to smaller research teams and institutions.

Best and Worst Distance Metrics for Genre Recognition  

Jensen–Shannon divergence, which compares probability distributions, was the best choice for grouping similar genres, especially when used with LDA and bag-of-words. The Delta and Manhattan methods also worked reasonably well. Euclidean was the worst performer across LDA, bag-of-words, and WGCNA despite its widespread use in text analysis, suggesting further research is needed to replace industry-standard metrics. Cosine distance, while effective for authorship attribution, was not ideal for measuring LDA topic distances. Doc2vec is less affected by the comparison method used. 

Figure depicting the influence of distance metrics on ARI scores for each feature type.
Fig 2. The influence of distance metrics on ARI scores for each feature type (Sobchuk and Šeļa, 2024, Figure 3).

Main Findings  

Unsupervised learning can detect thematic similarities, though performance varies. Methods like cosine distance, used in authorship attribution, are less effective for thematic analysis when used with minimal preprocessing and a small number of Most Frequent Words.

Reliable thematic analysis can improve large-scale problems of inconsistent manual genre tagging in digital libraries and identifying unclassified or undiscovered genres. Additionally, it can enhance book recommendation systems by enabling content-based similarity detection instead of solely relying on user behavior. Much like how Spotify suggests songs based on acoustic features.

Conclusion  

This study demonstrates the value of computational methods in literary analysis, showing how thematic clustering can enhance genre classification and literary evolution. It establishes a foundation for future large-scale literary studies.

Limitations  

Key limitations include the simplification of complex literary relationships in clustering, which despite reducing complex literary relationships into more manageable structures, may not work the same way with different settings or capture every important textual feature.

The study also did not separate thematic content from elements like narrative perspective. Additionally, genre classification remains subjective and ambiguous, and future work could explore alternative approaches, such as user-generated tags from sites like Goodreads.

Implications and Future Research  

This research provides a computational framework for thematic analysis, offering the potential for improving genre classification and book recommendation systems. Future work should incorporate techniques like BERTopic and Top2Vec, test these methods on larger and more diverse datasets, and further explore text simplification and clustering strategies.

Bibliography 

Sobchuk, O., Šeļa, A. Computational thematics: comparing algorithms for clustering the genres of literary fiction. Humanit Soc Sci Commun 11, 438 (2024). https://doi.org/10.1057/s41599-024-02933-6

Book genres. (2022). Chapterly. Retrieved May 4, 2025, from https://www.chapterly.com/blog/popular-and-lucrative-book-genres-for-authors.

Motor Brain-Computer Interface Reanimates Paralyzed Hand

by

Over five million people in the United States live with paralysis (Armour et al., 2016), representing a large portion of the US population. Though the extent of paralysis varies from person-to-person, most with paralysis experience unmet needs that subtract from their overall life satisfaction. A survey of those with paralysis revealed “peer support, support for family caregivers, [and] sports activities” as domains where individuals with paralysis experienced less fulfillment—with lower household income predicting a higher likelihood of unmet needs (Trezzini et al., 2019). Consequently, individuals with sufficient motor function have turned to video games as a means to meet some of these needs, as video games are sources of recreation, artistic expression, social connectedness, and enablement (Cairns et al., 2019). Oftentimes, however, these individuals are limited by what games they are able to engage with—as they often “avoid multiplayer games with able-bodied players” (Willsey et al., 2025). Thus, Willsey and colleagues (2025) explore brain-computer interfaces as a valuable potential solution for restoring more sophisticated motor control of not just video games, but of digital interfaces used for social networking or remote work.

Brain-computer interfaces (BCIs) are devices that read and analyze brain activity in order to produce commands that are then relayed to output devices, with the intent of restoring useful bodily function (Shih et al., 2012). Willsey et al. explain how current motor BCIs are unable to distinguish between the brain activity corresponding to the movement of different fingers, so BCIs have instead relied on detecting the more general movement of grasping a hand (where the fingers are treated as one group). This limits BCIs to controlling fewer dimensions of an instrument: just being able to control a computer’s point-and-click cursor control as compared to typing on a computer. Hence, Willsey et al. seek to expand BCIs to allow for greater object manipulation—implementing finger decoding that will differentiate the brain output signals for different fingers, allowing for “typing, playing a musical instrument or manipulating a multieffector digital interface such as a video game controller.” Improving BCIs would also involve continuous finger decoding, as finger decoding has mostly been done retrospectively, where finger signals are not classified and read until after the brain data is analyzed. 

Willsey et al. developed a BCI system that is capable of decoding three independent finger groups (with the thumb decoded into two dimensions), allowing for four total dimensions of control. By training on the participant’s brain over nine days as they attempt to move individual fingers, the BCI can learn to distinguish brain regions that correspond to finger movements. These four dimensions of control are well reflected in a quadcopter simulation, where a patient with an implemented BCI is able to manipulate a virtual hand to fly a quadcopter drone through various hoops of an obstacle course. Many applications, even beyond video games, are apparent. These finger controls can be extended to a robotic hand or could reanimate the paralyzed limb. 

Finger movement is decoded into three distinct groups (differentiated by color).
Finger movement is decoded into three distinct groups (differentiated by color; Willsey et al., 2025).
Participant navigates quadcopter through a hoop through decoded finger movements.
Participant navigates quadcopter through a hoop through decoded finger movements (Willsey et al., 2025).

Download Full Video

The patient’s feelings of social connectedness, enablement and recreation were greatly improved. Willsey et al. note how the patient often looked forward to the quadcopter sessions, frequently “[asking] when the next quadcopter session was.” Not only did the patient find enjoyment in controlling the quadcopter, but they found training not to be tedious and the controls intuitive. To date, this finger BCI proves to be the most capable kind of motor BCI, and will serve as a valuable model for non-motor BCIs, like Brain2Char, a system for decoding text from brain recordings.

However, BCIs raise significant ethical considerations that must be addressed alongside their development. Are users responsible for all outputs from a BCI, even with outputs unintended? Given that BCIs decode brain signaling and train on data from a very controlled setting, there is always the potential for natural “noise” that may upset a delicate BCI model. Ideally, BCIs are trained on a participant’s brain in a variety of different circumstances to mitigate these errors. Furthermore, BCIs may further stigmatize motor disabilities by encouraging individuals toward restoring “normal” abilities. I am particularly concerned about the cost of this technology. As with most new clinical technologies, implementation is expensive and ends up pricing out individuals with lower socioeconomic statuses. These are often the individuals that face the greatest need for technologies like BCI. As mentioned earlier, lower household income predicts more unmet needs for individuals with paralysis. Nonetheless, so long as they are developed responsibly and efforts are made to ensure their affordability, there is great promise in motor BCIs.

 

References

Armour, B. S., Courtney-Long, E. A., Fox, M. H., Fredine, H., & Cahill, A. (2016). Prevalence and Causes of Paralysis—United States, 2013. American Journal of Public Health, 106(10), 1855–1857. https://doi.org/10.2105/ajph.2016.303270

Cairns, P., Power, C., Barlet, M., Haynes, G., Kaufman, C., & Beeston, J. (2019). Enabled players: The value of accessible digital games. Games and Culture, 16(2), 262–282. https://doi.org/10.1177/1555412019893877

Shih, J. J., Krusienski, D. J., & Wolpaw, J. R. (2012). Brain-Computer interfaces in medicine. Mayo Clinic Proceedings, 87(3), 268–279. https://doi.org/10.1016/j.mayocp.2011.12.008

Trezzini, B., Brach, M., Post, M., & Gemperli, A. (2019). Prevalence of and factors associated with expressed and unmet service needs reported by persons with spinal cord injury living in the community. Spinal Cord, 57(6), 490–500. https://doi.org/10.1038/s41393-019-0243-y

Willsey, M. S., Shah, N. P., Avansino, D. T., Hahn, N. V., Jamiolkowski, R. M., Kamdar, F. B., Hochberg, L. R., Willett, F. R., & Henderson, J. M. (2025). A high-performance brain–computer interface for finger decoding and quadcopter game control in an individual with paralysis. Nature Medicine. https://doi.org/10.1038/s41591-024-03341-8

Computer Vision Ethics

by

Computer vision (CV) is a field of computer science that allows computers to “see” or, in more technical terms, recognize, analyze, and respond to visual data, such as videos and images. CV is widely used in our daily lives, from something as simple as recognizing handwritten text to something as complex as analyzing and interpreting MRI scans. With the advent of AI in the last few years, CV has also been improving rapidly. However, just like any subfield of AI nowadays, CV has its own set of ethical, social, and political implications, especially when used to analyze people’s visual data.

Although CV has been around for some time, there is limited work on its ethical limitations in the general AI field. Among the existing literature, authors categorized six ethical themes, which are espionage, identity theft, malicious attacks, copyright infringement, discrimination, and misinformation [1]. As seen in Figure 1, one of the main CV applications is face recognition, which could also lead to issues of error, function creep (the expansion of technology beyond its original purposes), and privacy. [2].

Computer Vision technologies related to Identity Theft
Figure 1: Specific applications of CV that could be used for Identity Theft.

To discuss CV’s ethics, the authors of the article take a critical approach to evaluating the implications through the framework of power dynamics. The three types of power that are analyzed are dispositional, episodic, and systemic powers [3]. 

Dispositional Power

Dispositional power is defined as the ability to bring out a significant outcome [4]. When people gain that power, they feel empowered to explore new opportunities, and their scope of agency increases (they become more independent in their actions) [5]. However, CV can threaten this dispositional power in several ways, ultimately reducing people’s autonomy. 

One way CV disempowers people is by limiting their information control. Since CV works with both pre-existing and real-time camera footage, people might be often unaware that they are being recorded and often cannot avoid that. This means that technology makes it hard for people to control the data that is being gathered about them, and protecting their personal information might get as extreme as hiding their faces. 

Apart from people being limited in controlling what data is being gathered about them, advanced technologies make it extremely difficult for an average person to know what specific information can be retrieved from visual data. Another way CV might disempower people of following their own judgment is through communicating who they are for them (automatically inferring people’s race, gender, and mood), creating a forced moral environment (where people act from fear of being watched rather than their own intentions), and potentially leading to over-dependence on computers (e.g., relying on face recognition for emotion interpretations). 

In all these and other ways, CV undermines the foundation of dispositional power by limiting people’s ability to control their information, make independent decisions, express themselves, and act freely.

Episodic Power

Episodic power, or as often referred to as power-over, defines the direct exercise of power by one individual or group over another. CV can both give new power or improve the efficiency of existing one [6]. While this isn’t always a bad thing (for example, parents watching over children), problems arise when CV makes that control too invasive or one-sided—especially in ways that limit people’s freedom to act independently.

 With CV taking security cameras to the next level, opportunities such as baby-room monitoring or fall detection for elderly people open up to us. However, it also leads to the issues of surveillance automation, which can lead to over-enforcement in scales as small as private individuals to bigger corporations (workplaces, insurance companies, etc.). Another power dynamic shifts that need to be considered, for example, when the smart doorbells show far beyond the person at the door and might violate a neighbor’s privacy by creating peer-to-peer surveillance. 

These examples show that while CV may offer convenience or safety, it can also tip power balances in ways that reduce personal freedom and undermine one’s autonomy.

Systemic Power

Systematic power is not viewed as an individual exercise of power, but rather a set of societal norms and practices that affect people’s autonomy by determining what opportunities people have, what values they hold, and what choices they make. CV can strengthen the systematic power by making law enforcement more efficient through smart cameras and increase businesses’ profit through business intelligence tools. 

However, CV can also reinforce the pre-existing systematic societal injustices. One example of that might be flawed facial recognition, when the algorithms are more likely to recognize White people and males [7], which led to a number of false arrests. This might lead to people receiving unequal opportunities (when biased systems are used for hiring process), or harm their self-worth (when falsely recognized as a criminal). 

Another matter of systematic power is the environmental cost of CV. AI systems rely on vast amounts of data, which requires intensive energy for processing and storage. As societies become increasingly dependent on AI technologies like CV, those trying to protect the environment have little ability to resist or reshape these damaging practices. The power lies with tech companies and industries, leaving citizens without the means to challenge the system. When the system becomes harder to challenge or change, that’s when the ethical concerns regarding CV arise.

Conclusion

Computer Vision is a powerful tool that keeps evolving each year. We already see numerous applications of it in our daily lives, starting from the self-checkouts in the stores and smart doorbells to autonomous vehicles and tumor detections. With the potential that CV holds in improving and making our lives safer, there are a number of ethical limitations that should be considered. We need to critically examine how CV affects people’s autonomy, might cause one-sided power dynamics, and reinforces societal prejudices. As we are rapidly transitioning into the AI-driven world, there is more to come in the field of computer vision. However, in the pursuit of innovation, we should ensure the progress does not come at the cost of our ethical values.

References:

[1] Lauronen, M.: Ethical issues in topical computer vision applications. Information Systems, Master’s Thesis. University of Jyväskylä. (2017). https://jyx.jyu.fi/bitstream/handle/123456789/55806/URN%3aNBN%3afi%3ajyu-201711084167.pdf?sequence=1&isAllowed=y

[2] Brey, P.: Ethical aspects of facial recognition systems in public places. J. Inf. Commun. Ethics Soc. 2(2), 97–109 (2004). https:// doi.org/10.1108/14779960480000246

[3] Haugaard, M.: Power: a “family resemblance concept.” Eur. J. Cult. Stud. 13(4), 419–438 (2010)

[4] Morriss, P.: Power: a philosophical analysis. Manchester University Press, Manchester, New York (2002)

[5] Morriss, P.: Power: a philosophical analysis. Manchester University Press, Manchester, New York (2002)

[6] Brey, P.: Ethical aspects of facial recognition systems in public places. J. Inf. Commun. Ethics Soc. 2(2), 97–109 (2004). https://doi.org/10.1108/14779960480000246

[7] Buolamwini, J., Gebru, T.: Gender shades: intersectional accuracy disparities in commercial gender classification. Conference on Fairness, Accountability, and Transparency, pp. 77–91 (2018) Coeckelbergh, M.: AI ethics. MIT Press (2020)

AI – save or ruin the environment?

by

With the fast speed that AI is currently developing, it has the potential to alleviate one of the most pressing problems—climate change. AI applications, such as smart electricity grids and sustainable agriculture, are predicted to mitigate environmental issues. On the flip side, the integration of AI in this field can also be counterproductive because of the high energy demand of the systems. If AI helps us to transition to a more sustainable lifestyle, the question is, at what cost?

The last decade saw exponential growth in data demand and the development of Large Language Models (LLMs)–computational models such as ChatGPT, designed to generate natural language. The algorithms resulted in increased energy consumption because of the big data volumes and computational power required, as well as increased water consumption needed to refrigerate data centers with that data. This consequently leads to higher greenhouse gas emissions (Fig.1). For example, the training of GPT-3 on a 500 billion-word database produced around 550 tons of carbon dioxide, equivalent to flying 33 times from Australia to the UK [1]. Moreover, information and communications technology (ICT) accounts for 3.9% of global greenhouse gas emissions (surpassing global air travel) [2]. As the number of training parameters grows, so does the energy consumption. It is expected to reach over 30% of the world’s total energy consumption by 2030. These environmental concerns about AI implementation led to a new term—Green AI.

Fig 1: CO2 equivalent emissions for training ML models (blue) and real-life cases (violet). In brackets, the billions of parameters are adjusted for each model [3].

Green algorithms are defined in two ways: green-in and green-by AI (Fig. 2). Algorithms that support the use of technology to tackle environmental issues are referred to as green-by AI. Green-in-design algorithms (green-in AI), on the other hand, are those that maximize energy efficiency to reduce the environmental impact of AI. 

 

Fig. 2. Overview of green-in vs. green-by algorithms.

 

Green-by AI has the potential to reduce greenhouse gas emissions by enhancing efficiency across many sectors, such as agriculture, biodiversity management, transportation, smart mobility, etc. 

  • Energy Efficiency. Machine Learning (ML) algorithms can optimize heating, air conditioning, and lighting by analyzing the data from the smart buildings, making them more energy efficient [4][5]. 
  • Smart Mobility. AI can predict and avoid traffic congestion by analyzing the current traffic patterns and optimizing routes. Moreover, ML contributes to Autonomous Vehicles by executing tasks like road following and obstacle detection, which improves overall road safety [6].
  • Sustainable agriculture. Data from sensors and satellites analyzed by ML can give farmers insights into crop health, soil conditions, and irrigation needs. This enables them to use the resources with precision and reduce environmental impacts. Moreover, predictive analytics minimize crop loss by allowing farmers to aid the diseases on time [7].
  • Climate Change. Computer-vision technologies can detect methane leaks in gas pipes, reducing emissions from fossil fuels. AI also plays a crucial role in reducing electricity usage by predicting demand and supply from solar and wind power.
  • Environmental Policies. AI’s ability to process data, identify trends, and predict outcomes will enable policymakers to come up with effective strategies to combat environmental issues [8].

Green-in AI, on the other hand, is an energy-efficient AI with a low carbon footprint, better quality data, and logical transparency. To ensure people’s trust, it offers clear and rational decision-making processes, thus also making it socially sustainable. Several promising approaches to reaching the green-in AI include algorithm, hardware, and data center optimization. Specifically, more efficient graphic processing units (GPUs) or parallelization (distributing computation among several processing cores) can reduce the environmental impacts of training AI. Anthony et al. proved that increasing the number of processing units to 15 will decrease greenhouse gas emissions [9]. However, the reduction in runtime must be significant enough for the parallelization method not to become counterproductive (when the execution time reduction is smaller than the increase in the number of cores, the emissions deteriorate). Other methods include computation at the locations where the data is collected to avoid data transmissions and limit the number of times an algorithm is run. 

Now that we know about AI’s impact and the ways to reduce it, what trends can we expect in the future? 

  • Hardware: Innovation in hardware design is focused on creating both eco-friendly and powerful AI accelerators, which can minimize energy consumption [10].
  • Neuromorphic computing is an emerging area in the computing technology field, aiming to create more efficient computing systems. It draws inspiration from the human brain, which performs complex tasks with much less energy than conventional computers. 
  • Energy-harvesting AI devices. Researchers are exploring the ways in which AI can harvest energy from its surroundings, for example from the lights or heat [11]. This way, AI can rely less on external power and become self-sufficient.

In conclusion, while AI holds great potential in alleviating many environmental issues, we should not forget about its own negative impact. While training AI models results in excessive greenhouse gas emissions, there are many ways to reduce energy consumption and make AI more environmentally friendly. Although we discussed several future trends in green-in AI, it is important to remember this field is still continuously evolving and new innovations will emerge in the future.

References:

[1] D. Patterson, J. Gonzalez, Q. Le, C. Liang, L.-M. Munguia, D. Rothchild, D. So, M. Texier, J. Dean, Carbon emissions and large neural network training, 2021, arXiv:2104.10350.

[2] Bran, Knowles. “ACM TCP TechBrief on Computing and Carbon Emissions.” Association for Computing Machinery, Nov. 2021  www.acm.org/media-center/2021/october/tpc-tech-brief-climate-change  

[3] Nestor Maslej, Loredana Fattorini, Raymond Perrault, Vanessa Parli, Anka Reuel, Erik Brynjolfsson, John Etchemendy, Katrina Ligett, Terah Lyons, James Manyika, Juan Carlos Niebles, Yoav Shoham, Russell Wald, and Jack Clark, “The AI Index 2024 Annual Report,” AI Index Steering Committee, Institute for Human-Centered AI, Stanford University, Stanford, CA, April 2024. 

[4] N. Milojevic-Dupont, F. Creutzig, Machine learning for geographically differentiated climate change mitigation in urban areas, Sustainable Cities Soc. 64 (2021) 102526.

[5] T.M. Ghazal, M.K. Hasan, M. Ahmad, H.M. Alzoubi, M. Alshurideh, Machine learning approaches for sustainable cities using internet of things, in: The Effect of Information Technology on Business and Marketing Intelligence Systems, Springer, 2023, pp. 1969–1986.

[6] M. Bojarski, D. Del Testa, D. Dworakowski, B. Firner, B. Flepp, P. Goyal, L.D. Jackel, M. Monfort, U. Muller, J. Zhang, et al., End to end learning for self-driving cars, 2016, arXiv preprint arXiv:1604.07316. 

[7] R. Sharma, S.S. Kamble, A. Gunasekaran, V. Kumar, A. Kumar, A systematic literature review on machine learning applications for sustainable agriculture supply chain performance, Comput. Oper. Res. 119 (2020) 104926.

[8] N. Sánchez-Maroño, A. Rodríguez Arias, I. Lema-Lago, B. Guijarro-Berdiñas, A. Dumitru, A. Alonso-Betanzos, How agent-based modeling can help to foster sustainability projects, in: 26th International Conference on Knowledge-Based and Intelligent Information and Engineering Systems, KES, 2022.

[9] L.F.W. Anthony, B. Kanding, R. Selvan, Carbontracker: Tracking and predicting the carbon footprint of training deep learning models, 2020, arXiv preprint arXiv:2007.03051. 

[10] H. Rahmani, D. Shetty, M. Wagih, Y. Ghasempour, V. Palazzi, N.B. Carvalho, R. Correia, A. Costanzo, D. Vital, F. Alimenti, et al., Next-generation IoT devices: Sustainable eco-friendly manufacturing, energy harvesting, and wireless connectivity, IEEE J. Microw. 3 (1) (2023) 237–255.

[11]  Divya S., Panda S., Hajra S., Jeyaraj R., Paul A., Park S.H., Kim H.J., Oh T.H.

Smart data processing for energy harvesting systems using artificial intelligence

Machine learning and algorithmic bias

by

Algorithms permeate modern society, especially AI algorithms. Artificial intelligence (AI) is built with various techniques, like machine learning, deep learning, or natural language processing, that trains AI to mimic humans at a certain task. Healthcare, loan approval, and security surveillance are a few industries that have begun using AI (Alowais et al., 2023; Purificato et al., 2022; Choung et al., 2024). Most people will inadvertently continue to interact with AI on a daily basis.

However, what are the problems faced by an increasing algorithmic society? Authors Sina Fazelpour and David Danks, in their article, explore this question in the context of algorithmic bias. Indeed, the problem they identify is that AI perpetuates bias. At its most neutral, Fazelpour and Danks (2021) explain that algorithmic bias is some “systematic deviation in algorithm output, performance, or impact, relative to some norm or standard,” suggesting that algorithms can be biased against a moral, statistical, or social norm. Fazelpour and Danks use a running example of a university training an AI algorithm with past student data to predict future student success. Thus, this algorithm exhibits a statistical bias if student success predictions are discordant with what has happened historically (in training data). Similarly, the algorithm exhibits a moral bias if it illegitimately depends on the student’s gender to produce a prediction. This is seen already in facial recognition algorithms that “perform worse for people with feminine features or darker skin” or recidivism prediction models that rate people of color as higher risk (Fazelpour & Danks, 2021). Clearly, algorithmic biases have the potential to preserve or exacerbate existing injustices under the guise of being “objective.” 

Algorithmic bias will manifest through different means. As Fazelpour and Danks discuss, harmful bias will be evident even prior to the creation of an algorithm if values and norms are not deeply considered. In the example of a student-success prediction model, universities must make value judgments, specifying what target variables define “student success,” whether it’s grades, respect from peers, or post-graduation salary. The more complex the goal, the more difficult and contested will choosing target variables be. Indeed, choosing target variables is a source of algorithmic bias. As Fazelpour and Danks explain, enrollment or financial aid decisions based on the prediction of student success may discriminate against minority students if first year performance was used in that prediction since minority students may face additional challenges.

Using training data that is biased will also lead to bias in an AI algorithm. In other words, bias in the measured world will be reflected in AI algorithms that mimic our world. For example, recruiting AI that reviews resumes is often trained on employees already hired by the company. In many cases, so-called gender-blind recruiting AI have discriminated against women by using gendered information on a resume that was absent from the resumes of a majority-male workplace (Pisanelli, 2022; Parasurama & Sedoc, 2021). Fazelpour and Danks also mention that biased data can arise from limitations and biases in measurement methods. This is what happens when facial recognition systems are trained predominantly on white faces. Consequently, these facial recognition systems are less effective when individuals do not look like the data the algorithm has been trained on.

Alternatively, users’ misinterpretations of predictive algorithms may produce biased results, Fazelpour and Danks argue. An algorithm is optimized for one purpose, and without even knowing, users may utilize this algorithm for another. A user could inadvertently interpret predicted “student success” as a metric for grades instead of what an algorithm is optimized to predict (e.g., likelihood to drop out). Decisions stemming from misinterpretations of algorithm predictions are doomed to be biased—and not just for the aforementioned reasons. Misunderstandings of algorithmic predictions lead to poor decisions if the variables predicting an outcome are also assumed to cause that outcome. Students in advanced courses may be predicted to have higher student success, but as Fazelpour and Danks put it, we shouldn’t enroll every underachieving student in an advanced course. Models such as these should also be applied in a context similar to when historical data was collected. Doing this is more important the longer a model is used as present data begins to differ from historical training data. In other words, student success models created for a small private college should not be deployed at a large public university nor many years later.

Fazelpour and Danks establish that algorithmic bias is nearly impossible to eliminate—solutions often must engage with the complexities of our society. The authors delve into several technical solutions, such as optimizing an algorithm using “fairness” as a constraint or training an algorithm on corrected historical data. This quickly reveals itself to be problematic, as determining fairness is a difficult value judgment. Nonetheless, algorithms provide tremendous benefit to us, even in moral and social ways. Algorithms can identify biases and serve as better alternatives to human practices. Fazelpour and Danks conclude that algorithms should continue to be studied in order to identify, mitigate, and prevent bias.

References

Alowais, S. A., Alghamdi, S. S., Alsuhebany, N., Alqahtani, T., Alshaya, A. I., Almohareb, S. N., Aldairem, A., Alrashed, M., Saleh, K. B., Badreldin, H. A., Yami, M. S. A., Harbi, S. A., & Albekairy, A. M. (2023). Revolutionizing healthcare: the role of artificial intelligence in clinical practice. BMC Medical Education, 23(1). https://doi.org/10.1186/s12909-023-04698-z

Choung, H., David, P., & Ling, T. (2024). Acceptance of AI-Powered Facial Recognition Technology in Surveillance scenarios: Role of trust, security, and privacy perceptions. Technology in Society, 102721. https://doi.org/10.1016/j.techsoc.2024.102721

Fazelpour, S., & Danks, D. (2021). Algorithmic bias: Senses, sources, solutions. Philosophy Compass, 16(8). https://doi.org/10.1111/phc3.12760

Parasurama, P., & Sedoc, J. (2021, December 16). Degendering resumes for fair algorithmic resume screening. arXiv.org. https://arxiv.org/abs/2112.08910

Pisanelli, E. (2022). Your resume is your gatekeeper: Automated resume screening as a strategy to reduce gender gaps in hiring. Economics Letters, 221, 110892. https://doi.org/10.1016/j.econlet.2022.110892

Purificato, E., Lorenzo, F., Fallucchi, F., & De Luca, E. W. (2022). The use of responsible artificial intelligence techniques in the context of loan approval processes. International Journal of Human-Computer Interaction, 39(7), 1543–1562. https://doi.org/10.1080/10447318.2022.2081284

Getting the Big Picture: Satellite Altimetry and the Future of Sea Level Rise Research

by

Anthropogenic climate change is drastically affecting the natural processes of the Earth at unprecedented rates. Increased fossil fuel emissions coupled with global deforestation have altered Earth’s energy budget, creating the potential for positive feedback loops to further warm our planet. While some of this warming manifests through glacier melting, powerful storm systems, and rising global temperatures, it’s estimated that 93% of the total energy gained from the greenhouse effect is stored in the ocean, with the remaining 7% contributing to atmospheric warming (Cazenave et al. 2018, as cited in von Schuckmann et al. 2016). This storage of heat in the ocean is responsible for oceanic thermal expansion and in combination with glacier melt is contributing to global sea level rise. Currently, an estimated 230 million people live below 1 m of the high tide line and if we do not curb emissions, sea level rise projections range 1.1 – 2.1 m by 2100 (Kulp et al. 2019, Sweet et al. 2022). Sea level rise’s global impact has thus been a prominent area of scientific research with leading methods utilizing satellite altimetry to measure the ocean’s height globally over time. 

Originating in the 1990s, surface sea level data has been recorded using a multitude of satellites amassing information from subseasonal to multi-decadal time scales (Cazenave et al. 2018). NASA’s sea level change portal reports this data sub-annually, recording a current sea level rise of 103.8 mm since 1993 (NASA). Seeking more information on the current trend of satellite altimetry, I reached out to French geophysicist Dr. Anny Cazenave of the French space agency CNES and director of Laboratoire d’Etudes en Geophysique et Oceanographie Spatiale (LEGOS) in Toulouse, France. Dr. Cazenave is a pioneer in geodesy, has worked as one of the leading scientists on numerous altimetry missions, was lead author of the sea level rise report for two Intergovernmental Panel on Climate Change (IPCC) reports, and recently won the prestigious Vetlesen Prize in 2020 (European Space Sciences Committee). 

When asked about recent advancements in altimetry technology, Dr. Cazenave directed me towards the recent international Surface Water and Ocean Topography satellite mission (SWOT) launched in 2022. SWOT is able to detect ocean features with ten times the resolution of current technology, enabling fine-scale analysis of oceans, lakes, rivers, and much more (NASA SWOT). Specifically for measuring sea level rise, SWOT utilizes a Ka-band Radar Interferometer (KaRIn) which is capable of measuring the elevation of almost all bodies of water on Earth. KaRIn operates by measuring deflected microwave signals off of Earth’s surface using two antennas split 10 meters apart, enabling the generation of a detailed topographic image of Earth’s surface (NASA SWOT). With SWOT’s high-resolution capabilities for topographically mapping sea level change anomalies close to shore, more accurate estimations for how sea level rise can affect coastal communities will be accessible in the future.

The figure above displays the difference in resolution between Copernicus Marine Service of ESA (European Space Agency) data and SWOT surface height anomaly data (NASA SWOT).

Finally, in light of recent developments in AI and machine learning, Dr. Cazenave noted the power of these computational methods in analyzing large data sets. The high-precision data provided by SWOT requires advanced methods of analysis to physically represent sea level rise changes, posing a challenge for researchers (Stanley 2023). A few recent papers have already highlighted the use of neural networks that are trained on current altimetry and sea surface temperature data (Xiao et al. 2023, Martin et al. 2023). These neural networks are then able to decipher the high-resolution data, enabling for a greater understanding of ocean dynamics and sea surface anomalies. Dr. Cazenave explained that the key questions to answer regarding sea level rise are: (1) how will ice sheets contribute to future sea level rise, (2) how much will sea level rise in coastal regions, and (3) how will rising sea levels contribute to shoreline erosion and retreat. With novel computational analysis techniques and advanced sea surface monitoring, many of these questions are being answered with greater accuracy. As we navigate the effects of climate change, combining science and policy will allow us to design multifaceted solutions that enable a sustainable future for all.

References

  1. Anny Cazenave​. European Space Sciences Committee. (n.d.). https://www.essc.esf.org/panels-members/anny-cazenave%E2%80%8B/
  2. Cazenave, A., Palanisamy, H., & Ablain, M. (2018). Contemporary sea level changes from satellite altimetry: What have we learned? What are the new challenges? Advances in Space Research, 62(7), 1639–1653. https://doi.org/10.1016/j.asr.2018.07.017
  3. Home. (n.d.). NASA Sea Level Change Portal. Retrieved April 24, 2024, from https://sealevel.nasa.gov/
  4. Joint NASA, CNES Water-Tracking Satellite Reveals First Stunning Views. (n.d.). NASA SWOT. Retrieved April 24, 2024, from https://swot.jpl.nasa.gov/news/99/joint-nasa-cnes-water-tracking-satellite-reveals-first-stunning-views
  5. Kulp, S. A., & Strauss, B. H. (2019). New elevation data triple estimates of global vulnerability to sea-level rise and coastal flooding. Nature Communications, 10(1), 4844. https://doi.org/10.1038/s41467-019-12808-z
  6. Martin, S. A., Manucharyan, G. E., & Klein, P. (2023). Synthesizing Sea Surface Temperature and Satellite Altimetry Observations Using Deep Learning Improves the Accuracy and Resolution of Gridded Sea Surface Height Anomalies. Journal of Advances in Modeling Earth Systems, 15(5), e2022MS003589. https://doi.org/10.1029/2022MS003589
  7. Stanley, S. (2023, October 17). Machine Learning Provides a Clearer Window into Ocean Motion. Eos. http://eos.org/research-spotlights/machine-learning-provides-a-clearer-window-into-ocean-motion
  8. Xiao, Q., Balwada, D., Jones, C. S., Herrero-González, M., Smith, K. S., & Abernathey, R. (2023). Reconstruction of Surface Kinematics From Sea Surface Height Using Neural Networks. Journal of Advances in Modeling Earth Systems, 15(10), e2023MS003709. https://doi.org/10.1029/2023MS003709
  9. von Schuckmann, K., Palmer, M., Trenberth, K. et al. An imperative to monitor Earth’s energy imbalance. Nature Clim Change 6, 138–144 (2016). https://doi.org/10.1038/nclimate2876

Navigating the Unseen: Wireless Muon Technology Revolutionizes Indoor Positioning and Beyond

by

Cosmic rays have captivated scientists due to their enigmatic origins, imperceptibility, and natural abundance. Originating from celestial bodies ranging in distances from as close as our sun to as far as distant galaxies, these particles bombard our Earth at rates close to the speed of light. While these particles are responsible for the aurora borealis displays in the arctic, for the most part they go unnoticed and have been mainly researched in the context of astronomy and astrophysics (Howell 2018). However, recent development in muon tomography and research from Professor Hiroyuki Tanaka’s research group from the University of Tokyo has developed a wireless muometric navigation system (MuWNS) capable of using muons to create an indoor positioning system (Tanaka 2022).

Formation of muons from particle showers (Vlasov, 2023).

Muons are natural subatomic particles that are created from cosmic rays interacting with atoms in the atmosphere. With their mass around 207 times that of electrons, muons are capable of penetrating solid materials and water (Gururaj 2023). This unique property of muons has allowed for their use in mapping the interiors of hard-to-access places such as volcanoes, tropical storm cells, and even Egyptian pyramids (Morishima, 2017). Professor Tanaka’s team has now focused on improving the currently limited GPS system with a wireless muon detection system capable of navigation in places where radio waves used in GPS can not reach. This makes it an ideal technology for underground and underwater navigation, natural disaster relief, exploration of caves in planets, and much more. 

While the initial principle behind MuWNS involves the precise measurement of the timing and direction of cosmic-ray-generated muons through reference detectors, Professor Tanaka’s team had issues with the synchronization of time between the reference and receiver detectors (Tanaka, 2022). This precise time synchronization issue was displayed in their 2022 MuWNS prototype that had a navigation accuracy between 2-14 m, which Professor Tanaka claims is “far from the level required for the practical indoor navigation applications.” In a more recent article published in September 2023, Professor Tanaka has shifted his focus from using the timing of muons to measuring the directional vectors of incoming muons. Thus, instead of using the time of muon travel between the reference and receiver detectors for navigation, the next generation vector muPS (muometric positioning system) uses the angles of incoming muons through the reference and receiver detectors to locate the detector’s positioning. In essence, matching the angles of muons entering the two detectors confirms the same muon event. By identifying the same muon event, the angle and path of the muon is then used to determine the position of the receiver detector without relying on timing mechanisms. This approach minimizes the effects of time synchronization resulting in what he predicts as centimeter-level accuracy (Tanaka 2023). This new development has been greeted with excitement, earning Professor Tanaka’s team a spot in Time Magazine’s “The Best Inventions Of 2023” (Stokel-Walker 2023).

This image is from Professor Tanaka’s article on wireless muometric navigation systems. Image A depicts underwater navigation with floating reference detectors and muons marked as red lines. Image B depicts underground navigation using surface reference detectors to control the receiver. (Tanaka, 2022).

After being intrigued by Professor Tanaka’s work published in Nature (Tanaka 2023), I reached out to him asking a few questions for this article. The first question I asked was about the presence of muons and whether muon tomography could work on other celestial bodies. His response highlighted that muons are in fact generated in dust deposits on top of the surface of the Moon and Mars. Specifically, Professor Tanaka discussed how muons could be used to explore caves within the Moon. This would involve deploying a muPS navigating robot that uses muons generated in the regolith for navigation underground. This could allow us to explore hard to examine places on other planets without the physical presence of human exploration.

The second question involves the application of muPS within cell phones. Tanaka explains that our phones currently have a GPS receiver inside of them, allowing us to track their location when they are lost. However, if the cellphone is lost in an elevator, basement, cave, or room that has limited GPS signals, muPS could locate the phone instead. With 6.92 billion smartphone users worldwide, this application could be useful in natural disasters where individuals may be trapped under rubble and GPS signals cannot locate their phones (Zippia 2023). 

Finally, I asked Professor Tanaka what made him excited about muPS. He responded by discussing the current limitations with our present indoor/underground navigation systems and how they all rely on laser, sound, or radio waves to guide them through obstacles. This method he claims is not technically navigation because it does not provide coordinate information and thus is un-programmable. Tanaka states that “muPS is [the] only technique that provides the coordinate information besides GPS” and it can be used in locations where GPS is unavailable. 

In future technology, muon-based positioning systems may provide the opportunity to open new navigational and observational possibilities, propelling us into a world of new discoveries and exploration on Earth and beyond. 

 

Work Cited

  1. Gururaj, T. (2023, June 16). World’s first cosmic-ray GPS can detect underground movement. Interesting Engineering. https://interestingengineering.com/innovation/cosmic-ray-gps-underground-movement-disaster-management-muons 
  2. Howell, E. (2018, May 11). What are cosmic rays?. Space.com. https://www.space.com/32644-cosmic-rays.html 
  3. Morishima, K., Kuno, M., Nishio, A. et al. (2017). Discovery of a big void in Khufu’s Pyramid by observation of cosmic-ray muons. Nature 552, 386–390.. https://doi.org/10.1038/nature24647
  4. Stokel-Walker, C. (2023, October 24). Muon Positioning System: The 200 best inventions of 2023. Time. https://time.com/collection/best-inventions-2023/6326412/muon-positioning-system/ 
  5. Tanaka, H.K.M. Wireless muometric navigation system. Sci Rep 12, 10114 (2022). https://doi.org/10.1038/s41598-022-13280-4
  6. Tanaka, H.K.M. Muometric positioning system (muPS) utilizing direction vectors of cosmic-ray muons for wireless indoor navigation at a centimeter-level accuracy. Sci Rep 13, 15272 (2023). https://doi.org/10.1038/s41598-023-41910-y
  7. Vlasov, A. (2023, April 14). Muon Imaging: How Cosmic Rays help us see inside pyramids and volcanoes. IAEA. https://www.iaea.org/newscenter/news/muon-imaging-how-cosmic-rays-help-us-see-inside-pyramids-and-volcanoes 
  8. Zippia. 20 Vital Smartphone Usage Statistics [2023]: Facts, Data, and Trends On Mobile Use In The U.S. Zippia.com. Apr. 3, 2023, https://www.zippia.com/advice/smartphone-usage-statistics/

Telepathic Technology: Novel Brain Implants Help Paralyzed Patients Communicate Again

by

Whether through lively classroom discussions, profound conversations with friends, or political debates at family dinners, speech resides at the crux of human connection. Tragic, life-altering events, such as stroke, traumatic brain injury, and the development of neurodegenerative diseases such as ALS can contribute to a loss of speech, often via vocal paralysis. When the nervous system is ravaged by disease or injury, severe paralysis and consequent locked-in syndrome can occur. In cases of locked-in syndrome, a patient loses all motor abilities and can only communicate via blinking or very minimal movements, rendering traditional speech aids such as typing and writing tools useless. While years of research have produced several assistive speech devices targeted at patients with severe paralysis, these devices are often extremely limited in their vocabulary range and only offer output options that are choppy, slow, and inauthentic. Despite allowing patients to communicate to some degree, the shortcomings of these devices remove much of the character and connection a person derives through speech, leaving patients feeling socially and emotionally isolated (Ramsey and Crone, 2023).

A recent study published in Nature has partially succeeded in remedying this issue. Through the development and use of brain-computer interfaces (BCIs), scientists have created a pathway that translates a patient’s neural signals into personalized text, speech audio, and animated, expressive facial avatars. In a case study involving a 47 year-old woman suffering from severe paralysis and complete loss of speech as a result of a brainstem stroke sustained 18 years earlier, researchers Metzger et al. designed and implanted a BCI into the left hemisphere of the patient’s brain, centered on her central sulcus and spanning the regions of her brain associated with speech production and language perception. The first of its kind, this BCI harnesses electrocorticography (ECoG) to decode neural signals for attempted speech and vocal tract movements into corresponding words, phrases, and facial expressions. Similar to the more familiar and established electroencephalography (EEG), which uses small electrodes attached to the scalp to monitor electrophysiological activity in the brain, electrocorticography can directly adhere to exposed surfaces of the brain to decode electric signals (“Electrocorticography”). The specific BCI used in this study contains a high-density array of 253 individual ECoG electrodes, connected to a percutaneous pedestal connector, which allows for the signals to be decoded and displayed on a computer interface (Metzger et al., 2023).

Following surgical implantation, the BCI and its connector were hooked up to a computer, which contained deep learning models that had already been “trained” via probability data to have predictive abilities that enable them to decode certain phones, silences, and articulatory gestures into words and phrases. An additional probability-based feature called a connectionist temporal classification loss function was added to the neural decoding network to distinguish the timing of the attempted speech and signals, which allowed for order and pauses between words to be identified. After being processed through this system, the decoded signals could either be translated directly into letters, discrete vocal speech units, or discrete articulatory gestures, which respectively produce artificial text, speech, and facial expressions (Fig. 1, Metzger et al., 2023). 

 

Figure 1. Overview of multimodal speech decoding in a patient with vocal paralysis. a. Overview of the speech-decoding pathway, from neural signal to speech output. b. Magnetic resonance imaging scan (MRI) of the patient’s brain, showing stroke-induced atrophy in the brainstem resulting in severe paralysis. c. MRI of the patient’s brain, overlaid with the implanted BCI in its actual location. d. Examples of simple articulatory movements attempted by the patient, coupled with their corresponding electrode-activation maps. Bottom graphs depict evoked cortical activity for each type of movement, along with the mean ± standard error across trials.

 

This novel neuroprosthesis provides an unprecedented sense of personalization and authenticity to communication for patients with extreme paralysis. While previous assistive speech technology could at best reach a rate of 14 words per minute, the BCI used in the case study averages 78 words per minute, which is much closer to the average adult rate of speaking, roughly 150 words per minute. Additionally, the speech function can be personalized to the patient’s voice prior to their vocal paralysis, and the facial expressions can be projected onto an avatar resembling the patient, adding another layer of personalization (Metzger et al., 2023).

Innovations in speech-targeted neuroprosthetic technology have the potential to change the lives of thousands of people dealing with severe paralysis. Naturally, these solutions are not perfect– the error rate of this particular BCI is approximately 20% for direct text decoding, and 50% for direct speech decoding (Ramsey and Crone, 2023). Despite sounding high, these numbers are a remarkable improvement from more established technologies and point to the vast potential of BCIs a few years down the road. Additionally, while this case is still in its early stages, there are several other similar BCIs being developed for the same purpose (Willett et al., 2023).  As research continues, it will be fascinating to observe the continued effects of this revolutionary technology on patients’ lives.

 

References

Electrocorticography—An overview | sciencedirect topics. (n.d.). Retrieved November 4, 2023, from https://www.sciencedirect.com/topics/neuroscience/electrocorticography

Metzger, S. L., Littlejohn, K. T., Silva, A. B., Moses, D. A., Seaton, M. P., Wang, R., Dougherty, M. E., Liu, J. R., Wu, P., Berger, M. A., Zhuravleva, I., Tu-Chan, A., Ganguly, K., Anumanchipalli, G. K., & Chang, E. F. (2023). A high-performance neuroprosthesis for speech decoding and avatar control. Nature, 620(7976), 1037–1046. https://doi.org/10.1038/s41586-023-06443-4

Ramsey, N. F., & Crone, N. E. (2023). Brain implants that enable speech pass performance milestones. Nature, 620(7976), 954–955. https://doi.org/10.1038/d41586-023-02546-0

Willett, F. R., Kunz, E. M., Fan, C., Avansino, D. T., Wilson, G. H., Choi, E. Y., Kamdar, F., Glasser, M. F., Hochberg, L. R., Druckmann, S., Shenoy, K. V., & Henderson, J. M. (2023). A high-performance speech neuroprosthesis. Nature, 620(7976), 1031–1036. https://doi.org/10.1038/s41586-023-06377-x

ChatGPT Beats Humans in Emotional Awareness Test: What’s Next?

by

In recent times, it can seem like everything revolves around artificial intelligence (AI). From AI-powered robots performing surgery to facial recognition on smartphones, AI has become an integral part of modern life. While AI has affected nearly every industry, most have been slowly adapting AI into their field while trying to minimize the risks involved with AI. One such field with particularly great potential is the mental health care industry. Indeed, some studies have already begun to study the uses of AI to assist mental health work. For instance, one study used AI to predict the probability of suicide through users’ health insurance records (Choi et al., 2018), while another showed that AI could identify people with depression based on their social media posts (Aldarwish & Ahmed, 2017). 

Perhaps the most wide-spread AI technology is ChatGPT, a public natural language processor chatbot that can help you with a plethora of tasks, from writing an essay to playing chess. Much discussion has been done about the potential of such chatbots in mental health care and therapy, but few studies have been published on the matter. However, a study by Zohar Elyoseph has started the conversation of chatbots’ potential, specifically ChatGPT, in therapy. In this study, Elyoseph and his team gave ChatGPT the Levels of Emotional Awareness Scale (LEAS) to measure ChatGPT’s capability for emotional awareness (EA), a core part of empathy and an essential skill of therapists (Elyoseph et al., 2023). The LEAS gives you 20 scenarios, in which someone experiences an event that supposedly elicits a response in the person in the scenario, and the test-taker must describe what emotions the person is likely feeling. Two examinations of the LEAS, one month apart, were done on ChatGPT to test two different versions of ChatGPT. This was done to see if updates during that month would improve its ability on the LEAS. On both examinations, two licensed psychologists scored the responses from ChatGPT to ensure reliability of its score. On the first examination in January 2023, ChatGPT achieved a score of 85 out of 100, compared to the French men’s and female’s averages of 56.21 and 58.94 respectively. On the second examination in February 2023, ChatGPT achieved a score of 98: nearly a perfect score, a significant improvement from the already high score of 85 a month prior, and a score that is higher than most licensed psychologists (Elyoseph et al., 2023).

This study shows that, not only is ChatGPT more capable than humans at EA, but it is also rapidly improving at it. This has massive implications for in-person therapy. While there is more to being a good therapist than just emotional awareness, it is a major part of it. Therefore, based on this study, there is potential for chatbots like ChatGPT to rival, or possibly even replace, therapists if developers are able to develop the other interpersonal traits of good therapists. 

However, ChatGPT and AI needs more work to be done before it can really be implemented into the mental health field in this manner. To start, while AI is capable of the technical aspects of therapy, such as giving sound advice and validating a client’s emotions, ChatGPT and other chatbots sometimes give “illusory responses”, or fake responses that it claims are legitimate (Hagendorff et al., 2023). For example, ChatGPT will sometimes say “5 + 5 = 11” if you ask what 5 + 5 is, even though the answer is clearly wrong. While this is a very obvious example of an illusory response, harm can be done if the user is not able to distinguish between the real and illusory responses for more complex subjects. These responses can be extremely harmful in situations such as therapy, as clients rely on a therapist for guidance, and if such guidance were fake, it could harm rather than help the client. Furthermore, there are concerns regarding the dehumanization of therapy, the loss of jobs for therapists, and the breach of a client’s privacy if AI was to replace therapists (Abrams, 2023). ​​

Fig 1. Sample conversation with Woebot, which provides basic therapy to users. Adapted from Darcy et al., 2021. 

However, rudimentary AI programs are already sprouting that try to bolster the mental health infrastructure. Replika, for instance, is an avatar-based chatbot that offers therapeutic conversation with the user, and saves previous conversations to remember them in the future. Woebot provides a similar service (Figure 1), providing cognitive-behavioral therapy (CBT) for anxiety and depression to users (Pham et al., 2022). While some are scared about applications such as these, these technologies should be embraced since, as they become more refined, they could provide a low-commitment, accessible source of mental health care for those who are unable to reach out to a therapist, such as those who are nervous about reaching out to a real therapist, those who live in rural environments without convenient access to a therapist, or those who lack the financial means for mental health support. AI can also be used as a tool for therapists in the office. For example, an natural language processing application, Eleos, can take notes and highlight themes and risks for therapists to review after the session (Abrams, 2023). 

There are certainly some drawbacks of AI in therapy, such as the dehumanization of therapy, that may not have a solution and could therefore limit AI’s influence in the field. There is certainly a chance that some people would never trust AI to give them empathetic advice. However, people said the same when robotic surgeries began being used in clinical settings, but most people seem to have embraced that due to its superb success rate. Regardless of whether these problems are resolved, AI in the mental health industry has massive potential, and we must make sure to ensure that the risks and drawbacks of such technology are addressed and refined so that we can make the most of this potential in the future and bring better options to those who need it. 

 

Citations

Abrams, Z. (2023, July 1). AI is changing every aspect of psychology. Here’s what to watch for. Monitor on Psychology, 54(5). https://www.apa.org/monitor/2023/07/psychology-embracing-ai

 

Aldarwish MM, Ahmad HF. Predicting Depression Levels Using Social Media Posts. Proc – 2017 IEEE 13th Int Symp Auton Decentralized Syst ISADS 2017 2017;277–80.

 

Choi SB, Lee W, Yoon JH, Won JU, Kim DW. Ten-year prediction of suicide death using Cox regression and machine learning in a nationwide retrospective cohort study in South Korea. J Affect Disord. 2018;231(January):8–14.

 

Darcy, Alison & Daniels, Jade & Salinger, David & Wicks, Paul & Robinson, Athena. (2021). Evidence of Human-Level Bonds Established With a Digital Conversational Agent: Cross-sectional, Retrospective Observational Study. JMIR Formative Research. 5. e27868. 10.2196/27868. 

 

Elyoseph, Z., Hadar-Shoval, D., Asraf, K., & Lvovsky, M. (2023). ChatGPT outperforms humans in emotional awareness evaluations. Frontiers in psychology, 14, 1199058. 

 

Hagendorff, T., Fabi, S. & Kosinski, M. Human-like intuitive behavior and reasoning biases emerged in large language models but disappeared in ChatGPT. Nat Comput Sci 3, 833–838.

 

Pham K. T., Nabizadeh A., Selek S. (2022). Artificial intelligence and chatbots in psychiatry. Psychiatry Q. 93, 249–253.