Artificial intelligence, otherwise known as AI, has been around for a bit longer than we might realize. Programs like Siri, Cortana, and Alexa were one of the first stepping stones towards true AI. They possess machine learning capabilities to figure out individuals’ preferences over time, are able to run commands based on user input and modify their behavior to fit corrections, and can access information on the internet. However, new evolutions of AI have become popularized in the past decade, and this next step in AI development has rapidly integrated into our society.
This development is the creation of generative AI programs and programs that have natural language processing capabilities. ChatGPT is a prime example of an AI program with both features. Programs like ChatGPT can take a text-based prompt and produce an output based on the various sources the program has been trained on. This includes texts, visuals, and as of recently, audio.
While many people find the growing capabilities of AI to be impressive and are excited for a possible future with AI as a central component of society, there are also problems surfacing with its ongoing development. One of these is the ongoing spread of misinformation and the visual misrepresentation of key facts, such as juvenile birds.
People first became aware of this issue with what some news sources have called the “fake baby peacock trend” produced by AI. AI-generated images of peacock chicks depicted what looked to be merely smaller adult peacocks, with a small portion of baby feathers on the front of the bird, have been appearing in Google searches when looking for photos of peacock chicks. These images also showed the chicks as having adult feather coloration already, despite this not being how actual baby peacock coloration appears in nature. These images are not labeled as AI, posing a problem for members of the public who are not experts on birds and want information on the differences between adult peacocks and their young.
More recently, we at Penguins International have become aware of this issue spreading to penguin chicks. Emperor Penguin chicks seem to be the sole issue thus far, but it is possible that it could spread to other penguin chicks. These images tend to show a large portion of the chick, or chicks, with a gray body, as is true. However, they extend the black feathers of an Emperor Penguin chick’s head down to various regions of the body, whether that be around the wings or extending down the main body of the chick. Actual Emperor Penguin chicks’ bodies are also more torpedo-shaped, while AI generates them as more short and round like more avian bird chicks’ bodies.
A picture of an emperor penguin with four juvenile emperor penguins, one hidden behind the adult. (Source – Dafna Ben nun from Wikimedia)
An AI-generated picture of emperor penguin chicks. These chicks have black feathers in improper places, and some have irregular body proportions, a common issue with AI images (Source – Dreamtime)
As you can see from the two pictures, AI-generated images have a likelihood to be misinforming when they produce pictures of penguin chicks. While no mainstream pictures of other AI-generated penguin species’ chicks have been seen in search results, it is not unlikely that generative AI produces inaccurate images of them as well.
Another critical issue with AI is misinformation through text in various forms. Whether it be mixing up the truth with myth, referencing outdated information, or misleading omission, just going off of AI-compiled search results could lead users to not having a complete and accurate picture of topics like penguin ecology or conservation status. During the research I did to write this paper, I searched for what penguin species are at risk of extinction. The AI auto-results did not present any species that are NOT currently at-risk, but either used incorrect labeling terms for the penguin’s conservation status or omitted particular species. It’s important to note that these results CAN change based on keyword orientation, but still pose some issues with misleading information.
Source: Bianca Anderson
A screenshot of the AI-generated results for “what penguin species are at risk”. Note: “White-Flippered Penguin” is another name for the Little Penguin
The above search result demonstrates AI pulling from both outdated information and omitting particular information. Both Galapagos Penguins, which are not mentioned here, and Northern Rockhopper Penguins are recognized as endangered, which is not reflected in this screenshot. Snares, Macaroni, and Southern Rockhopper Penguins are also not mentioned as being threatened, or more specifically vulnerable. The search results also seem to mix being near-threatened and being vulnerable. While this isn’t as major, it can still be confusing to a member of the public and is disingenuous to the true conservation status of a species.
An educated public is crucial when it comes to conservation and educational efforts, such as those Penguins International is focused on. Misinformation being spread via growing blind trust in AI has the potential to harm our efforts, and we must talk about these issues in the technology’s developments to limit the spread of these effects.
At the end of the day, while AI has its uses, it is important to not be reliant on it when it comes to finding accurate information. There are still many flaws with systems like generative AI, ChatGPT, and search-bot AI that have yet to be overcome, and they are not at a place yet where they should be depended on for accurate educational information in topics such as wildlife biology. It is important that internet users learn how to identify AI-produced misinformation and work to combat this growing problem.
To identify whether or not a text is misinformation when it’s coming from AI, the first check is to step for any unusual grammar patterns. If there’s any weird punctuation, improper spelling or capitalization, or other grammatical errors, there is a chance the information itself is not correct. Se`cond, look for whether or not something restated is different from what the AI said prior. Inconsistencies are a strong indication of potential false information in AI texts. As for images, the best way to check if an image is AI or not is to look for uncanny proportions of living subjects, odd overlapping of different aspects, such as a penguin merging into an iceberg, or out-of-place objects. Comparing the picture to surrounding images is a good way to check as to whether or not the subject is portrayed correctly. If you do use AI-generated search results for a quick skim on a topic, be sure to double-check any information you’re given with credible sources.
Reporting misinformation is crucial for fighting back against the rapid spread of misinformation online. If an AI response gives you incorrect information, locate the “report” button and report the information as being inaccurate. Include links to credible sources and give any necessary corrections that the AI needs to make for future search summaries on the topic at hand. When you spy an AI-generated image that is not labeled as AI, similarly reporting it and requesting it be flagged as generated content can warn other users that the image is likely to contain inaccuracies, and lower the risk of confusion between true pictures of wildlife and those made purely by computer programs.
How programs like Siri work (compared to ChatGPT):
https://www.tomsguide.com/ai/ive-been-using-chatgpt-voice-7-things-it-can-do-the-new-siri-cant – Comparisons between Siri and ChatGPT
https://www.tableau.com/learn/articles/machine-learning-examples – Broad scope of how digital assistants and voice-to-text programs have AI-based functions
ChatGPT Function Sources:
https://www.techtarget.com/whatis/definition/ChatGPT#:~:text=These%20are%20some%20uses%20for,that%20are%20created%20by%20AI. – What is ChatGPT in the world of AI?
https://barstow.libguides.com/generative-ai/timeline – Timeline of AI’s technological evolution
Outdated AI References:
https://www.worldatlas.com/articles/a-list-of-endangered-species-of-penguins.html – In reference to endangered penguins
Statements on AI’s Inaccuracies:
https://lib.guides.umd.edu/c.php?g=1340355&p=9880574 – Different things that AI gets wrong, including
- Omission of information
- Blatantly wrong information
- The creation of false information and blurring of truth and fiction
Conservation Status of Penguin Species:
- Each penguin species’ conservation status was checked on the Penguins International webpage pertaining to that particular species
