Can we talk with whales?🐋 [HTC #25]
HelloOOOoo, WooOOorld!
Hello!
Welcome to Hold the Code #25.
Although not all of us can speak whale like Dori from Finding Nemo, in this edition we explore if AI may be able to do this for us. We also look at how AI can be used to predict renewable energy production, if AI knows when we are lying, and what an ethics bounty system is. Happy reading!
Decoding Whale Codas
The ability to speak to animals might seem more like science fiction than reality, but since March 2020, an interdisciplinary group of scientists from around the world has been trying to use artificial intelligence to decode the “codas” — clicking patterns — of sperm whales. The goal of Project CETI (Cetacean Translation Initiative) is to expand natural language processing (a branch of AI that deals with speech analysis) into the realm of nonhuman languages.
What is Natural Language Processing?
Over the past decade, advancements in the field have produced neural networks that can learn language structures through statistical analysis of language data — a process similar to how children learn language by recognizing speech patterns. More sophisticated machines can generate fake articles, summarize texts, and translate between languages. In principle, it is entirely conceivable to use these technologies to translate between human and nonhuman languages.
Can it Work on Whales?
Despite these promising advancements, however, the project also faces many obstacles. For example, huge amounts of data are needed to train a neural network. So far, CETI researchers have found that applying machine learning algorithms to a database of about 100,000 whale codas works relatively well for simple tasks, but deeper analysis would require much more data to work with. The researchers also need a means of annotating their coda recordings with data on the whales’ behaviors, in order to correlate meaning to the sounds.
Although there is still much work left to do, the researchers are optimistic that as technology advances, they will eventually be able to develop a system that can generate whale codas, and potentially even communicate with real whales. Whether these goals will be achieved remains to be seen, but there is no doubt that this project has the potential to change the way we use AI to analyze languages, human or otherwise.
Modeling Energy
Debates on renewable energy sources have created many questions regarding whether fossil fuels can be replaced as the dominant source of our power but new applications in machine learning may have potentially predicted a model that solves this problem.
Searching far and wide
Researchers have built a machine learning system that is capable of detecting solar facilities and then creates an assessment of the estimated power output across the globe. In total, it detected 68,661 solar facilities and predicted the total power output of these were estimated to be around 423 gigawatts (GW). This is extremely close to the measurements of the International Renewable Energy Agency’s estimation of 420 GW which speaks on the capabilities of this technology.
Typically, data on renewable energy output is calculated by questionnaires, limiting their ability to address areas where growth is needed. The machine learning model boasts the ability to calculate accurate energy production measurements while also directly pinpointing coverage gaps. The article states:
“As solar becomes more predictable, grid operators will need to keep fewer fossil fuel power plants in reserve, and fewer penalties for over- or under-generation will mean more marginal projects will be unlocked…Data like this allows us to study the precise conditions which are leading to the diffusion of solar energy, and will help governments better design subsidies to encourage faster growth.”
Implementing this model of machine learning towards various other renewable energy sources allows for a holistic approach towards proving energy that can measure energy production and assess shortages world wide.
Can AI tell if you’re lying?
Insurance company Lemonade received backlash on social media when it revealed it uses artificial intelligence to analyze videos that customers send whenever they open a claim. Customers must explain their losses in these videos, and AI analyzes them for “non-verbal cues” to detect instances of fraud.
Lemonade is not alone. A growing number of companies are investing in lie detection technologies, despite public criticisms.
Does this technology actually work?
According to Adam Tanner, the author of the article, top industry officials said that lie detection works—to a limited extent. They say these technologies can verify a caller’s identity to avoid fraud by comparing one’s voice to previous calls. The experts, however, do not believe that the technology can actually identify if what a caller is saying is true.
Why is lie detection technology controversial?
Although traditional passwords and other information like birthday and PIN number are flawed, advanced technologies that analyze eye movement or vocal sounds are problematic. Throughout history, humans have never invented a perfect truth detection machine.
John Kircher, chief scientist at Converus, whose product tracks eye movement to detect lies, recognizes the flaws of lie-detection technology.
“It’s not perfect. If it’s 90 percent accurate, one person out of 10, if you’re telling the truth, is going to fail it. So there’s a reason for people to be concerned.”
Weekly Feature: Headhunting Ethics Bugs
A few months ago, @capohai posted a screenshot of Google’s response to the query “what do terrorists wear on their head,” which returned a Palestinian keffiyeh as the top result. Outrage erupted over social media before making its way into more mainstream news sources, but the angry articles and tweets and comments seemed to fall on deaf ears.
Today, the keffiyeh is still the top result of this Google search.
Social Media Outrage
Ironically, social media is often the go-to method to express outrage and seek retribution from tech giants. Therefore, it makes sense that Twitter was the platform used to raise awareness of this ethical violation in Google’s search engine. Users are able to reach a wide audience and mobilize supporters to put pressure on these companies through public exposure and negative press. However, there is no set system of accountability, and social media outrage may be met with nothing but inaction.
Legislation Situation
Situations similar to this one have led to calls for greater regulation of the tech industry. While stronger legislation is undoubtedly necessary, the legislative process can be lengthy. Additionally, any legislation passed may be ineffective against unforeseen ethics failures. Algorithms have a tendency to break down in unexpected ways, especially when it comes to evaluating ethics, and often need correcting. If legislation alone cannot address all instances of ethical violation on tech platforms, what can?
What is an Ethics Bounty?
An ethics bounty would be a system by which users could be paid to report ethics violations to tech companies. Similar systems are already in place to help these companies catch “critical issues” in their systems. These issues are normally technical bugs affecting functionality or security.
Here’s how the process works at Apple:
The user has to notice the bug and be the first to report it.
The user must explain what happened and show evidence of the bug.
The user cannot publicly disclose the bug until Apple has had a chance to fix it.
The user receives compensation.
Currently, these systems are only in place for technical issues, but what if these companies included ethics violations as “critical issues” in their systems? Users could report these violations and receive compensation for their efforts. Companies would benefit in being able to fix their system to be more just and ethical. Additionally, these companies could bulk up their public record of addressing ethical issues promptly and responsibly.
Setting up an ethics bounty system would help companies quickly address ethical issues on their platform through user input in a more systematic and accountable way.
Global Accountability
An ethics bounty system would be necessary on a global scale. Ethics violations that occur outside of the US in languages other than English particularly suffer from a lack of attention by companies. A bounty system allows users to more easily get in contact with companies, report incidents, and see improvements made. Ideally, an ethics bounty system should be governed by a global nonprofit of ethicists to avoid cover-ups, nontransparency, and lying by internal ethics committees at tech companies. This nonprofit would determine the severity of the violations and the payout to users who flagged them.
Learn More
To read the full opinion piece on Wired, click here.
Written by Michelle Zhang, Dwayne Morgan, Molly Pribble
Edited by Molly Pribble