With AI embedded in all kinds of applications, it is hard for both adults and youth to avoid its use. If you do text messaging by phone, the predictive text feature is AI. If you use autocomplete to fill in a form, you are using AI. Youth who submit résumés created by AI are having recruiters screen and sort through them using AI.

The above examples are AI, but they are not generative AI, which is a whole other ball of wax. Generative AI creates humanlike text. It creates images, artwork, audio files and videos. It can be found on interactive platforms that facilitate conversations. Sometimes it is almost impossible to tell in a chat session whether you are texting a person or an AI. You can tell many times if an AI has been trained on too limited an amount or a poor dataset.

I have recently settled on using a generative AI called Perplexity.ai as my research assistant. That doesn’t mean I don’t continue to experiment with other vendor offerings. Perpexity.ai, however, has demonstrated consistency in the quality of its responses to my questions.

How do I use it to assist me in writing this blog?

Before generative AI tools appeared on the scene, I researched future articles by creating online searches for specific subject matter. These searches ended up piled in my inbox for me to sort through daily. I would select the ones that interested me to write about and then refine my search criteria to help me write the story. Sometimes my reading included jargon-filled scientific research papers. I had to get beyond new terminology to understand the content so that I could write about it using layman’s terms. This involved hours of reading and interpreting. Now, with Perplexity.ai, I can cut through the jargon to get to the essence of a journal article’s content. That doesn’t mean I don’t read the paper, because I do. Perplexity.ai, however, makes it easier for me to learn and saves me time.

When I ask Perplexity.ai a question, it provides answers with links. I spend time on the links. What I never do is accept the AI’s version of the facts until I have read those sources.

So far, Perplexity.ai hasn’t hallucinated or produced nonsense answers. I have, however, noticed that it repeats itself when I rephrase a question. It also repeats the question as an answer at the beginning of its response. I find this latter trait to be quite amusing since this was a sales interviewing technique I learned many years ago.

Salespeople beware!

American Psychological Association Publishes AI Health Advisory

In the past, I have written about the well-being of youth when exposed to social media and small screens. These are areas of interest to the American Psychological Association (APA), as is AI. It recently published a health advisory for youth addressing the subject. For its purposes, the APA defines youth as between the ages of 10 and 25 and asks who needs to be involved in ensuring that youth use AI responsibly.

The APA recommendations are a long list, and recommendations include talking to youth, parents, caregivers, educators, policymakers, and AI developers. The report notes the unique aspects of youth when interacting with AI, including the ages and stages from age 10 to 25 that affect brain development, emotional and physical maturity, and psychological competence. It also notes that the APA cannot overlook past harmful mistakes made in recommendations involving youth and social media.

The APA report describes how youth in different environments with different upbringings react to similar content in very different ways because of temperament, neurodiversity, exposure to stress or violence, social isolation, traumatic experiences, mental health, age, and exposure to socioeconomic or structural disadvantages.

It notes that AI is making discerning truth even more difficult and states that if an AI is trained on Internet content containing misinformation, how can youth discern if what the AI is stating is true or false?

The following is a partial list of recommendations for youth and AI aimed at developers, educators and parents:

• Let youth know that not all AI-generated content is accurate or age-appropriate.
• Discuss with youth the intent of some AI bots and provide age-appropriate privacy and interaction settings with limits.
• Educate youth about potential indicators of misinformation or marketing goals that may be underlying human-bot interactions using gamification as a honeypot, or manipulative notifications and other personalized response tools that can easily be coercive.
• Apply continuous testing with youth groups to identify and mitigate negative impacts of AI applications before they are released to the public.
• Include human oversight and support for youth to report concerns, seek help, or disengage from harmful AI interactions.
• Protect youth data privacy with developers, parents and educators discouraging the posting of images online that could be compromising.
• Provide parental controls and tools that let them set parameters that are suitable and age-appropriate.

For the adults in the room, the APA report includes these further recommendations:

• Educators need to establish an AI literacy core curriculum that spans computer science, social studies, and ethics courses. In addition, teachers need to be trained on AI concepts, algorithmic bias, and responsible AI use.
• Technology developers need to include features that include notifications and reminders to youth that they are interacting with it and not a human. In addition, they need to make AI algorithms and data collection practices transparent and provide educational tools and resources for users.
• Governments need to provide funding for research into AI literacy and develop guidelines and teacher training programs. In addition, governments need to pass regulations to protect youth’s mental and emotional health, ensuring that AI technologies used by youth are safe and beneficial.

Since OpenAI introduced ChatGPT I’ve studied Large Language Models (LLMs) and tested their evolving capabilities. ChatGPT created Monica as an enhancement to search engines. Others have produced competitors: Gemini (Google’s AI), Claude (Anthropic) and Grok (xAI, Elon Musk’s contribution).

When I read about two women in Brazil today dying from the Oropouche virus, one I had never heard of, and that it was mosquito-borne and spreading beyond the Amazon Basin to other areas of South America, I became somewhat concerned.

Having just read about Perplexity.ai as an AI search assistant, I decided to test it on a general question about pandemic threats. I asked:

“What viruses represent future pandemic threats?” 

This is what Perplexity wrote:

Based on the search results, several viruses represent potential future pandemic threats:

1. Novel coronaviruses: Similar to SARS-CoV-2 that caused COVID-19, new coronaviruses could emerge and cause future pandemics.
2. Influenza viruses: New strains of influenza, particularly avian influenza like H7N9, pose a significant pandemic risk.
3. Ebola virus: This has caused outbreaks and is considered a priority pathogen by the WHO.
4. Marburg virus: Another hemorrhagic fever virus that is on the WHO’s priority list.
5. Lassa fever virus: Included in WHO’s list of priority pathogens.
6. Nipah virus: Considered a threat for future outbreaks.
7. Zika virus: Has caused significant outbreaks and is on WHO’s watchlist.
8. MERS (Middle East Respiratory Syndrome) coronavirus.
9. Other zoonotic viruses: Many potential pandemic threats are zoonotic, originating in wildlife before spilling over to humans.
10. “Disease X”: This represents an unknown pathogen that could cause a serious international epidemic or pandemic.

The World Health Organization (WHO) maintains a list of priority pathogens for research and development efforts, which includes these and other potential threats. It’s important to note that while these viruses are known concerns, the next pandemic could also be caused by a currently unknown pathogen. This is why the concept of “Disease X” is included in pandemic preparedness efforts. Factors contributing to increased pandemic risk include environmental changes, human encroachment on natural habitats, and increased global travel and trade. To prepare for future pandemics, ongoing surveillance, research into these viral families, and development of vaccines and treatments are crucial.

This is a pretty comprehensive response and expressed well. What I didn’t understand was since Perplexity.ai was a search assistant, why wouldn’t it mention Oropouche when a Google search did?

I thought a fair test would be to pose the same question to other AIs. I asked ChatGPT’s Monica and it came back with two additions to the Perplexity.ai list. The one I couldn’t understand Perplexity.ai missing was Dengue, described by Monica as “A mosquito-borne virus that can cause severe illness and has been spreading to new regions due to climate change.” The other was unfamiliar to me called Hendra and was described as “similar to Nipah, it is also transmitted from bats and can infect horses and humans, leading to severe respiratory and neurological symptoms.”

I then asked Google’s Gemini. Gemini’s response grouped viruses into two categories: zoonotic and RNA-based. It also included “Disease X” as an unknown. Then it added a category neither of the previous two had mentioned: the growing threat of AMR (anti-microbial resistance) as a pandemic threat.

Gemini wasn’t finished in its response. Even though I hadn’t asked, Gemini added information about root causes including:

• Climate change: Altering ecosystems, increasing contact between humans and animals, and creating new environments for pathogens to thrive.
• Globalization: Rapid travel and trade facilitate the spread of diseases in a borderless world.
• Antimicrobial overuse: Contributing to the rise of AMR.  
• Deforestation and habitat destruction: Bringing humans into closer contact with wildlife and their pathogens.

In fairness to Perplexity and Monica, both included suggested follow-up questions to provide root-cause answers.

I went through the same process with Claude. Claude had two more to add to the list: Filoviruses, a group of viruses similar to Ebola, and Henipaviruses, a group of bat-borne viruses related to Nipah. Claude then asked me if I wanted it “to elaborate on any of these viruses or discuss pandemic preparedness strategies?”

My final pursuit in this exercise was to try and ask Grok the same question But with Grok I couldn’t because I needed to be an X Premium+ subscriber in the United States to use it.

Getting back to Perplexity.ai, it wasn’t alone in not mentioning Oropouche. But as a true search AI assistant, I couldn’t understand why it didn’t get mentioned.

I knew why the others had missed Oropouche. That’s because Monica, Gemini and Claude are AIs built with Large Language Model (LLM) algorithms that train with datasets of materials from the Internet that are often several months old.

To prove the difference, I went through a final exercise involving current events at the Paris Olympics. I asked Gemini and Claude who won medals today on August 5, 2024, and here are the two answers respectively:

Gemini: “There seems to be a slight confusion. The Paris Olympics are scheduled for July 26 to August 11, 2024. As of now, we are in the year 2023.” Whoops!!

Claude: “I apologize, but I don’t have access to real-time information about Olympic events…The most recent Olympic Games I have detailed information about were the 2022 Winter Olympics in Beijing. Sorry!

Ask Perplexity.ai the same question and here is its answer:

“Here are the key medal winners from the Paris Olympics on August 5, 2024:

• Gymnastics: Simone Biles won a silver medal in the floor exercise final, concluding her participation in the 2024 Paris Olympics without adding to her seven Olympic golds.
• Badminton: Lee Zii Jia secured Malaysia’s second medal at the Paris Games.
• Track and Field: Noah Lyles won gold in the men’s 100-meter sprint with a time of 9.784 seconds, narrowly beating Jamaica’s Kishane Thompson, who took silver.”

As the old saying goes, “One of these is not like the others” and it is Perplexity.ai. It knows what’s happening in the present. Not mentioning the Oropouche virus was the only surprise to me. Maybe the explanation is that Oropouche wasn’t seen as a pandemic threat as of yet.

Ignoring the Oropouche omission, it is pretty clear to me that as AIs go, the current frontrunners whose LLMs train on datasets from the past will soon be obsolete. If I were them, in the light of Perplexity.ai, I would be panicking.

Let’s hope Oropouche is right in its omission and that the virus doesn’t join other vector-borne diseases to spread across the planet.